merge with stable

contrib/byteify-strings.py

@@ -1,345 +1,346 b''
 #!/usr/bin/env python3
 #
 # byteify-strings.py - transform string literals to be Python 3 safe
 #
 # Copyright 2015 Gregory Szorc <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import, print_function
 
 import argparse
 import contextlib
 import errno
 import os
 import sys
 import tempfile
 import token
 import tokenize
 
 
 def adjusttokenpos(t, ofs):
     """Adjust start/end column of the given token"""
     return t._replace(
         start=(t.start[0], t.start[1] + ofs), end=(t.end[0], t.end[1] + ofs)
     )
 
 
 def replacetokens(tokens, opts):
     """Transform a stream of tokens from raw to Python 3.
 
     Returns a generator of possibly rewritten tokens.
 
     The input token list may be mutated as part of processing. However,
     its changes do not necessarily match the output token stream.
     """
     sysstrtokens = set()
 
     # The following utility functions access the tokens list and i index of
     # the for i, t enumerate(tokens) loop below
     def _isop(j, *o):
         """Assert that tokens[j] is an OP with one of the given values"""
         try:
             return tokens[j].type == token.OP and tokens[j].string in o
         except IndexError:
             return False
 
     def _findargnofcall(n):
         """Find arg n of a call expression (start at 0)
 
         Returns index of the first token of that argument, or None if
         there is not that many arguments.
 
         Assumes that token[i + 1] is '('.
 
         """
         nested = 0
         for j in range(i + 2, len(tokens)):
             if _isop(j, ')', ']', '}'):
                 # end of call, tuple, subscription or dict / set
                 nested -= 1
                 if nested < 0:
                     return None
             elif n == 0:
                 # this is the starting position of arg
                 return j
             elif _isop(j, '(', '[', '{'):
                 nested += 1
             elif _isop(j, ',') and nested == 0:
                 n -= 1
 
         return None
 
     def _ensuresysstr(j):
         """Make sure the token at j is a system string
 
         Remember the given token so the string transformer won't add
         the byte prefix.
 
         Ignores tokens that are not strings. Assumes bounds checking has
         already been done.
 
         """
         k = j
         currtoken = tokens[k]
         while currtoken.type in (token.STRING, token.NEWLINE, tokenize.NL):
             k += 1
             if currtoken.type == token.STRING and currtoken.string.startswith(
                 ("'", '"')
             ):
                 sysstrtokens.add(currtoken)
             try:
                 currtoken = tokens[k]
             except IndexError:
                 break
 
     def _isitemaccess(j):
         """Assert the next tokens form an item access on `tokens[j]` and that
         `tokens[j]` is a name.
         """
         try:
             return (
                 tokens[j].type == token.NAME
                 and _isop(j + 1, '[')
                 and tokens[j + 2].type == token.STRING
                 and _isop(j + 3, ']')
             )
         except IndexError:
             return False
 
     def _ismethodcall(j, *methodnames):
         """Assert the next tokens form a call to `methodname` with a string
         as first argument on `tokens[j]` and that `tokens[j]` is a name.
         """
         try:
             return (
                 tokens[j].type == token.NAME
                 and _isop(j + 1, '.')
                 and tokens[j + 2].type == token.NAME
                 and tokens[j + 2].string in methodnames
                 and _isop(j + 3, '(')
                 and tokens[j + 4].type == token.STRING
             )
         except IndexError:
             return False
 
     coldelta = 0  # column increment for new opening parens
     coloffset = -1  # column offset for the current line (-1: TBD)
     parens = [(0, 0, 0, -1)]  # stack of (line, end-column, column-offset, type)
     ignorenextline = False  # don't transform the next line
     insideignoreblock = False  # don't transform until turned off
     for i, t in enumerate(tokens):
         # Compute the column offset for the current line, such that
         # the current line will be aligned to the last opening paren
         # as before.
         if coloffset < 0:
             lastparen = parens[-1]
             if t.start[1] == lastparen[1]:
                 coloffset = lastparen[2]
             elif t.start[1] + 1 == lastparen[1] and lastparen[3] not in (
                 token.NEWLINE,
                 tokenize.NL,
             ):
                 # fix misaligned indent of s/util.Abort/error.Abort/
                 coloffset = lastparen[2] + (lastparen[1] - t.start[1])
             else:
                 coloffset = 0
 
         # Reset per-line attributes at EOL.
         if t.type in (token.NEWLINE, tokenize.NL):
             yield adjusttokenpos(t, coloffset)
             coldelta = 0
             coloffset = -1
             if not insideignoreblock:
                 ignorenextline = (
                     tokens[i - 1].type == token.COMMENT
                     and tokens[i - 1].string == "# no-py3-transform"
                 )
             continue
 
         if t.type == token.COMMENT:
             if t.string == "# py3-transform: off":
                 insideignoreblock = True
             if t.string == "# py3-transform: on":
                 insideignoreblock = False
 
         if ignorenextline or insideignoreblock:
             yield adjusttokenpos(t, coloffset)
             continue
 
         # Remember the last paren position.
         if _isop(i, '(', '[', '{'):
             parens.append(t.end + (coloffset + coldelta, tokens[i + 1].type))
         elif _isop(i, ')', ']', '}'):
             parens.pop()
 
         # Convert most string literals to byte literals. String literals
         # in Python 2 are bytes. String literals in Python 3 are unicode.
         # Most strings in Mercurial are bytes and unicode strings are rare.
         # Rather than rewrite all string literals to use ``b''`` to indicate
         # byte strings, we apply this token transformer to insert the ``b``
         # prefix nearly everywhere.
         if t.type == token.STRING and t not in sysstrtokens:
             s = t.string
 
             # Preserve docstrings as string literals. This is inconsistent
             # with regular unprefixed strings. However, the
             # "from __future__" parsing (which allows a module docstring to
             # exist before it) doesn't properly handle the docstring if it
             # is b''' prefixed, leading to a SyntaxError. We leave all
             # docstrings as unprefixed to avoid this. This means Mercurial
             # components touching docstrings need to handle unicode,
             # unfortunately.
             if s[0:3] in ("'''", '"""'):
                 # If it's assigned to something, it's not a docstring
                 if not _isop(i - 1, '='):
                     yield adjusttokenpos(t, coloffset)
                     continue
 
             # If the first character isn't a quote, it is likely a string
             # prefixing character (such as 'b', 'u', or 'r'. Ignore.
             if s[0] not in ("'", '"'):
                 yield adjusttokenpos(t, coloffset)
                 continue
 
             # String literal. Prefix to make a b'' string.
             yield adjusttokenpos(t._replace(string='b%s' % t.string), coloffset)
             coldelta += 1
             continue
 
         # This looks like a function call.
         if t.type == token.NAME and _isop(i + 1, '('):
             fn = t.string
 
             # *attr() builtins don't accept byte strings to 2nd argument.
             if fn in (
                 'getattr',
                 'setattr',
                 'hasattr',
                 'safehasattr',
                 'wrapfunction',
                 'wrapclass',
                 'addattr',
             ) and (opts['allow-attr-methods'] or not _isop(i - 1, '.')):
                 arg1idx = _findargnofcall(1)
                 if arg1idx is not None:
                     _ensuresysstr(arg1idx)
 
             # .encode() and .decode() on str/bytes/unicode don't accept
             # byte strings on Python 3.
             elif fn in ('encode', 'decode') and _isop(i - 1, '.'):
                 for argn in range(2):
                     argidx = _findargnofcall(argn)
                     if argidx is not None:
                         _ensuresysstr(argidx)
 
             # It changes iteritems/values to items/values as they are not
             # present in Python 3 world.
             elif opts['dictiter'] and fn in ('iteritems', 'itervalues'):
                 yield adjusttokenpos(t._replace(string=fn[4:]), coloffset)
                 continue
 
         if t.type == token.NAME and t.string in opts['treat-as-kwargs']:
             if _isitemaccess(i):
                 _ensuresysstr(i + 2)
             if _ismethodcall(i, 'get', 'pop', 'setdefault', 'popitem'):
                 _ensuresysstr(i + 4)
 
         # Looks like "if __name__ == '__main__'".
         if (
             t.type == token.NAME
             and t.string == '__name__'
             and _isop(i + 1, '==')
         ):
             _ensuresysstr(i + 2)
 
         # Emit unmodified token.
         yield adjusttokenpos(t, coloffset)
 
 
 def process(fin, fout, opts):
     tokens = tokenize.tokenize(fin.readline)
     tokens = replacetokens(list(tokens), opts)
     fout.write(tokenize.untokenize(tokens))
 
 
 def tryunlink(fname):
     try:
         os.unlink(fname)
     except OSError as err:
         if err.errno != errno.ENOENT:
             raise
 
 
 @contextlib.contextmanager
 def editinplace(fname):
     n = os.path.basename(fname)
     d = os.path.dirname(fname)
     fp = tempfile.NamedTemporaryFile(
         prefix='.%s-' % n, suffix='~', dir=d, delete=False
     )
     try:
         yield fp
         fp.close()
         if os.name == 'nt':
             tryunlink(fname)
         os.rename(fp.name, fname)
     finally:
         fp.close()
         tryunlink(fp.name)
 
 
 def main():
     ap = argparse.ArgumentParser()
     ap.add_argument(
         '--version', action='version', version='Byteify strings 1.0'
     )
     ap.add_argument(
         '-i',
         '--inplace',
         action='store_true',
         default=False,
         help='edit files in place',
     )
     ap.add_argument(
         '--dictiter',
         action='store_true',
         default=False,
         help='rewrite iteritems() and itervalues()',
     ),
     ap.add_argument(
         '--allow-attr-methods',
         action='store_true',
         default=False,
         help='also handle attr*() when they are methods',
     ),
     ap.add_argument(
         '--treat-as-kwargs',
         nargs="+",
         default=[],
         help="ignore kwargs-like objects",
     ),
     ap.add_argument('files', metavar='FILE', nargs='+', help='source file')
     args = ap.parse_args()
     opts = {
         'dictiter': args.dictiter,
         'treat-as-kwargs': set(args.treat_as_kwargs),
         'allow-attr-methods': args.allow_attr_methods,
     }
     for fname in args.files:
+        fname = os.path.realpath(fname)
         if args.inplace:
             with editinplace(fname) as fout:
                 with open(fname, 'rb') as fin:
                     process(fin, fout, opts)
         else:
             with open(fname, 'rb') as fin:
                 fout = sys.stdout.buffer
                 process(fin, fout, opts)
 
 
 if __name__ == '__main__':
     if sys.version_info[0:2] < (3, 7):
         print('This script must be run under Python 3.7+')
         sys.exit(3)
     main()

contrib/heptapod-ci.yml

@@ -1,82 +1,81 b''
 image: octobus/ci-mercurial-core
 
 # The runner made a clone as root.
 # We make a new clone owned by user used to run the step.
 before_script:
     - hg clone . /tmp/mercurial-ci/ --noupdate
     - hg -R /tmp/mercurial-ci/ update `hg log --rev '.' --template '{node}'`
     - cd /tmp/mercurial-ci/
-    - (cd tests; ls -1 test-check-*.*) > /tmp/check-tests.txt
+    - ls -1 tests/test-check-*.* > /tmp/check-tests.txt
 
 variables:
     PYTHON: python
     TEST_HGMODULEPOLICY: "allow"
 
 .runtests_template: &runtests
     script:
-        - cd tests/
         - echo "python used, $PYTHON"
         - echo "$RUNTEST_ARGS"
-        - HGMODULEPOLICY="$TEST_HGMODULEPOLICY" "$PYTHON" run-tests.py --color=always $RUNTEST_ARGS
+        - HGMODULEPOLICY="$TEST_HGMODULEPOLICY" "$PYTHON" tests/run-tests.py --color=always $RUNTEST_ARGS
 
 checks-py2:
     <<: *runtests
     variables:
         RUNTEST_ARGS: "--time --test-list /tmp/check-tests.txt"
 
 checks-py3:
     <<: *runtests
     variables:
         RUNTEST_ARGS: "--time --test-list /tmp/check-tests.txt"
         PYTHON: python3
 
 rust-cargo-test-py2: &rust_cargo_test
     script:
         - echo "python used, $PYTHON"
         - make rust-tests
 
 rust-cargo-test-py3:
     <<: *rust_cargo_test
     variables:
         PYTHON: python3
 
 test-py2:
     <<: *runtests
     variables:
         RUNTEST_ARGS: " --no-rust --blacklist /tmp/check-tests.txt"
         TEST_HGMODULEPOLICY: "c"
 
 test-py3:
     <<: *runtests
     variables:
         RUNTEST_ARGS: " --no-rust --blacklist /tmp/check-tests.txt"
         PYTHON: python3
         TEST_HGMODULEPOLICY: "c"
 
 test-py2-pure:
     <<: *runtests
     variables:
         RUNTEST_ARGS: "--pure --blacklist /tmp/check-tests.txt"
         TEST_HGMODULEPOLICY: "py"
 
 test-py3-pure:
     <<: *runtests
     variables:
         RUNTEST_ARGS: "--pure --blacklist /tmp/check-tests.txt"
         PYTHON: python3
         TEST_HGMODULEPOLICY: "py"
 
 test-py2-rust:
     <<: *runtests
     variables:
         HGWITHRUSTEXT: cpython
         RUNTEST_ARGS: "--rust --blacklist /tmp/check-tests.txt"
         TEST_HGMODULEPOLICY: "rust+c"
 
 test-py3-rust:
     <<: *runtests
     variables:
         HGWITHRUSTEXT: cpython
         RUNTEST_ARGS: "--rust --blacklist /tmp/check-tests.txt"
         PYTHON: python3
         TEST_HGMODULEPOLICY: "rust+c"

mercurial/archival.py

@@ -1,386 +1,390 b''
 # archival.py - revision archival for mercurial
 #
 # Copyright 2006 Vadim Gelfer <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import gzip
 import os
 import struct
 import tarfile
 import time
 import zipfile
 import zlib
 
 from .i18n import _
 from .node import nullrev
 from .pycompat import open
 
 from . import (
     error,
     formatter,
     match as matchmod,
     pycompat,
     scmutil,
     util,
     vfs as vfsmod,
 )
 
 stringio = util.stringio
 
 # from unzip source code:
 _UNX_IFREG = 0x8000
 _UNX_IFLNK = 0xA000
 
 
 def tidyprefix(dest, kind, prefix):
     '''choose prefix to use for names in archive.  make sure prefix is
     safe for consumers.'''
 
     if prefix:
         prefix = util.normpath(prefix)
     else:
         if not isinstance(dest, bytes):
             raise ValueError(b'dest must be string if no prefix')
         prefix = os.path.basename(dest)
         lower = prefix.lower()
         for sfx in exts.get(kind, []):
             if lower.endswith(sfx):
                 prefix = prefix[: -len(sfx)]
                 break
     lpfx = os.path.normpath(util.localpath(prefix))
     prefix = util.pconvert(lpfx)
     if not prefix.endswith(b'/'):
         prefix += b'/'
     # Drop the leading '.' path component if present, so Windows can read the
     # zip files (issue4634)
     if prefix.startswith(b'./'):
         prefix = prefix[2:]
     if prefix.startswith(b'../') or os.path.isabs(lpfx) or b'/../' in prefix:
         raise error.Abort(_(b'archive prefix contains illegal components'))
     return prefix
 
 
 exts = {
     b'tar': [b'.tar'],
     b'tbz2': [b'.tbz2', b'.tar.bz2'],
     b'tgz': [b'.tgz', b'.tar.gz'],
     b'zip': [b'.zip'],
     b'txz': [b'.txz', b'.tar.xz'],
 }
 
 
 def guesskind(dest):
     for kind, extensions in pycompat.iteritems(exts):
         if any(dest.endswith(ext) for ext in extensions):
             return kind
     return None
 
 
 def _rootctx(repo):
     # repo[0] may be hidden
     for rev in repo:
         return repo[rev]
     return repo[nullrev]
 
 
 # {tags} on ctx includes local tags and 'tip', with no current way to limit
 # that to global tags.  Therefore, use {latesttag} as a substitute when
 # the distance is 0, since that will be the list of global tags on ctx.
 _defaultmetatemplate = br'''
 repo: {root}
 node: {ifcontains(rev, revset("wdir()"), "{p1node}{dirty}", "{node}")}
 branch: {branch|utf8}
 {ifeq(latesttagdistance, 0, join(latesttag % "tag: {tag}", "\n"),
       separate("\n",
                join(latesttag % "latesttag: {tag}", "\n"),
                "latesttagdistance: {latesttagdistance}",
                "changessincelatesttag: {changessincelatesttag}"))}
 '''[
     1:
 ]  # drop leading '\n'
 
 
 def buildmetadata(ctx):
     '''build content of .hg_archival.txt'''
     repo = ctx.repo()
 
     opts = {
         b'template': repo.ui.config(
             b'experimental', b'archivemetatemplate', _defaultmetatemplate
         )
     }
 
     out = util.stringio()
 
     fm = formatter.formatter(repo.ui, out, b'archive', opts)
     fm.startitem()
     fm.context(ctx=ctx)
     fm.data(root=_rootctx(repo).hex())
 
     if ctx.rev() is None:
         dirty = b''
         if ctx.dirty(missing=True):
             dirty = b'+'
         fm.data(dirty=dirty)
     fm.end()
 
     return out.getvalue()
 
 
 class tarit(object):
     '''write archive to tar file or stream.  can write uncompressed,
     or compress with gzip or bzip2.'''
 
-    class GzipFileWithTime(gzip.GzipFile):
-        def __init__(self, *args, **kw):
-            timestamp = None
-            if 'timestamp' in kw:
-                timestamp = kw.pop('timestamp')
-            if timestamp is None:
-                self.timestamp = time.time()
-            else:
-                self.timestamp = timestamp
-            gzip.GzipFile.__init__(self, *args, **kw)
+    if pycompat.ispy3:
+        GzipFileWithTime = gzip.GzipFile  # camelcase-required
+    else:
+
+        class GzipFileWithTime(gzip.GzipFile):
+            def __init__(self, *args, **kw):
+                timestamp = None
+                if 'mtime' in kw:
+                    timestamp = kw.pop('mtime')
+                if timestamp is None:
+                    self.timestamp = time.time()
+                else:
+                    self.timestamp = timestamp
+                gzip.GzipFile.__init__(self, *args, **kw)
 
-        def _write_gzip_header(self):
-            self.fileobj.write(b'\037\213')  # magic header
-            self.fileobj.write(b'\010')  # compression method
-            fname = self.name
-            if fname and fname.endswith(b'.gz'):
-                fname = fname[:-3]
-            flags = 0
-            if fname:
-                flags = gzip.FNAME  # pytype: disable=module-attr
-            self.fileobj.write(pycompat.bytechr(flags))
-            gzip.write32u(  # pytype: disable=module-attr
-                self.fileobj, int(self.timestamp)
-            )
-            self.fileobj.write(b'\002')
-            self.fileobj.write(b'\377')
-            if fname:
-                self.fileobj.write(fname + b'\000')
+            def _write_gzip_header(self):
+                self.fileobj.write(b'\037\213')  # magic header
+                self.fileobj.write(b'\010')  # compression method
+                fname = self.name
+                if fname and fname.endswith(b'.gz'):
+                    fname = fname[:-3]
+                flags = 0
+                if fname:
+                    flags = gzip.FNAME  # pytype: disable=module-attr
+                self.fileobj.write(pycompat.bytechr(flags))
+                gzip.write32u(  # pytype: disable=module-attr
+                    self.fileobj, int(self.timestamp)
+                )
+                self.fileobj.write(b'\002')
+                self.fileobj.write(b'\377')
+                if fname:
+                    self.fileobj.write(fname + b'\000')
 
     def __init__(self, dest, mtime, kind=b''):
         self.mtime = mtime
         self.fileobj = None
 
         def taropen(mode, name=b'', fileobj=None):
             if kind == b'gz':
                 mode = mode[0:1]
                 if not fileobj:
                     fileobj = open(name, mode + b'b')
                 gzfileobj = self.GzipFileWithTime(
                     name,
                     pycompat.sysstr(mode + b'b'),
                     zlib.Z_BEST_COMPRESSION,
                     fileobj,
-                    timestamp=mtime,
+                    mtime=mtime,
                 )
                 self.fileobj = gzfileobj
                 return tarfile.TarFile.taropen(  # pytype: disable=attribute-error
                     name, pycompat.sysstr(mode), gzfileobj
                 )
             else:
                 return tarfile.open(name, pycompat.sysstr(mode + kind), fileobj)
 
         if isinstance(dest, bytes):
             self.z = taropen(b'w:', name=dest)
         else:
             self.z = taropen(b'w|', fileobj=dest)
 
     def addfile(self, name, mode, islink, data):
         name = pycompat.fsdecode(name)
         i = tarfile.TarInfo(name)
         i.mtime = self.mtime
         i.size = len(data)
         if islink:
             i.type = tarfile.SYMTYPE
             i.mode = 0o777
             i.linkname = pycompat.fsdecode(data)
             data = None
             i.size = 0
         else:
             i.mode = mode
             data = stringio(data)
         self.z.addfile(i, data)
 
     def done(self):
         self.z.close()
         if self.fileobj:
             self.fileobj.close()
 
 
 class zipit(object):
     '''write archive to zip file or stream.  can write uncompressed,
     or compressed with deflate.'''
 
     def __init__(self, dest, mtime, compress=True):
         if isinstance(dest, bytes):
             dest = pycompat.fsdecode(dest)
         self.z = zipfile.ZipFile(
             dest, 'w', compress and zipfile.ZIP_DEFLATED or zipfile.ZIP_STORED
         )
 
         # Python's zipfile module emits deprecation warnings if we try
         # to store files with a date before 1980.
         epoch = 315532800  # calendar.timegm((1980, 1, 1, 0, 0, 0, 1, 1, 0))
         if mtime < epoch:
             mtime = epoch
 
         self.mtime = mtime
         self.date_time = time.gmtime(mtime)[:6]
 
     def addfile(self, name, mode, islink, data):
         i = zipfile.ZipInfo(pycompat.fsdecode(name), self.date_time)
         i.compress_type = self.z.compression  # pytype: disable=attribute-error
         # unzip will not honor unix file modes unless file creator is
         # set to unix (id 3).
         i.create_system = 3
         ftype = _UNX_IFREG
         if islink:
             mode = 0o777
             ftype = _UNX_IFLNK
         i.external_attr = (mode | ftype) << 16
         # add "extended-timestamp" extra block, because zip archives
         # without this will be extracted with unexpected timestamp,
         # if TZ is not configured as GMT
         i.extra += struct.pack(
             b'<hhBl',
             0x5455,  # block type: "extended-timestamp"
             1 + 4,  # size of this block
             1,  # "modification time is present"
             int(self.mtime),
         )  # last modification (UTC)
         self.z.writestr(i, data)
 
     def done(self):
         self.z.close()
 
 
 class fileit(object):
     '''write archive as files in directory.'''
 
     def __init__(self, name, mtime):
         self.basedir = name
         self.opener = vfsmod.vfs(self.basedir)
         self.mtime = mtime
 
     def addfile(self, name, mode, islink, data):
         if islink:
             self.opener.symlink(data, name)
             return
         f = self.opener(name, b"w", atomictemp=False)
         f.write(data)
         f.close()
         destfile = os.path.join(self.basedir, name)
         os.chmod(destfile, mode)
         if self.mtime is not None:
             os.utime(destfile, (self.mtime, self.mtime))
 
     def done(self):
         pass
 
 
 archivers = {
     b'files': fileit,
     b'tar': tarit,
     b'tbz2': lambda name, mtime: tarit(name, mtime, b'bz2'),
     b'tgz': lambda name, mtime: tarit(name, mtime, b'gz'),
     b'txz': lambda name, mtime: tarit(name, mtime, b'xz'),
     b'uzip': lambda name, mtime: zipit(name, mtime, False),
     b'zip': zipit,
 }
 
 
 def archive(
     repo,
     dest,
     node,
     kind,
     decode=True,
     match=None,
     prefix=b'',
     mtime=None,
     subrepos=False,
 ):
     '''create archive of repo as it was at node.
 
     dest can be name of directory, name of archive file, or file
     object to write archive to.
 
     kind is type of archive to create.
 
     decode tells whether to put files through decode filters from
     hgrc.
 
     match is a matcher to filter names of files to write to archive.
 
     prefix is name of path to put before every archive member.
 
     mtime is the modified time, in seconds, or None to use the changeset time.
 
     subrepos tells whether to include subrepos.
     '''
 
     if kind == b'txz' and not pycompat.ispy3:
         raise error.Abort(_(b'xz compression is only available in Python 3'))
 
     if kind == b'files':
         if prefix:
             raise error.Abort(_(b'cannot give prefix when archiving to files'))
     else:
         prefix = tidyprefix(dest, kind, prefix)
 
     def write(name, mode, islink, getdata):
         data = getdata()
         if decode:
             data = repo.wwritedata(name, data)
         archiver.addfile(prefix + name, mode, islink, data)
 
     if kind not in archivers:
         raise error.Abort(_(b"unknown archive type '%s'") % kind)
 
     ctx = repo[node]
     archiver = archivers[kind](dest, mtime or ctx.date()[0])
 
     if not match:
         match = scmutil.matchall(repo)
 
     if repo.ui.configbool(b"ui", b"archivemeta"):
         name = b'.hg_archival.txt'
         if match(name):
             write(name, 0o644, False, lambda: buildmetadata(ctx))
 
     files = list(ctx.manifest().walk(match))
     total = len(files)
     if total:
         files.sort()
         scmutil.prefetchfiles(
             repo, [ctx.rev()], scmutil.matchfiles(repo, files)
         )
         progress = repo.ui.makeprogress(
             _(b'archiving'), unit=_(b'files'), total=total
         )
         progress.update(0)
         for f in files:
             ff = ctx.flags(f)
             write(f, b'x' in ff and 0o755 or 0o644, b'l' in ff, ctx[f].data)
             progress.increment(item=f)
         progress.complete()
 
     if subrepos:
         for subpath in sorted(ctx.substate):
             sub = ctx.workingsub(subpath)
             submatch = matchmod.subdirmatcher(subpath, match)
             subprefix = prefix + subpath + b'/'
             total += sub.archive(archiver, subprefix, submatch, decode)
 
     if total == 0:
         raise error.Abort(_(b'no files match the archive pattern'))
 
     archiver.done()
     return total

mercurial/cext/revlog.c

@@ -1,2914 +1,2917 b''
 /*
  parsers.c - efficient content parsing
 
  Copyright 2008 Matt Mackall <[e-mail]> and others
 
  This software may be used and distributed according to the terms of
  the GNU General Public License, incorporated herein by reference.
 */
 
 #define PY_SSIZE_T_CLEAN
 #include <Python.h>
 #include <assert.h>
 #include <ctype.h>
 #include <limits.h>
 #include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
 
 #include "bitmanipulation.h"
 #include "charencode.h"
 #include "revlog.h"
 #include "util.h"
 
 #ifdef IS_PY3K
 /* The mapping of Python types is meant to be temporary to get Python
  * 3 to compile. We should remove this once Python 3 support is fully
  * supported and proper types are used in the extensions themselves. */
 #define PyInt_Check PyLong_Check
 #define PyInt_FromLong PyLong_FromLong
 #define PyInt_FromSsize_t PyLong_FromSsize_t
 #define PyInt_AsLong PyLong_AsLong
 #endif
 
 typedef struct indexObjectStruct indexObject;
 
 typedef struct {
 	int children[16];
 } nodetreenode;
 
 typedef struct {
 	int abi_version;
 	Py_ssize_t (*index_length)(const indexObject *);
 	const char *(*index_node)(indexObject *, Py_ssize_t);
 	int (*index_parents)(PyObject *, int, int *);
 } Revlog_CAPI;
 
 /*
  * A base-16 trie for fast node->rev mapping.
  *
  * Positive value is index of the next node in the trie
  * Negative value is a leaf: -(rev + 2)
  * Zero is empty
  */
 typedef struct {
 	indexObject *index;
 	nodetreenode *nodes;
 	unsigned length;   /* # nodes in use */
 	unsigned capacity; /* # nodes allocated */
 	int depth;         /* maximum depth of tree */
 	int splits;        /* # splits performed */
 } nodetree;
 
 typedef struct {
 	PyObject_HEAD /* ; */
 	    nodetree nt;
 } nodetreeObject;
 
 /*
  * This class has two behaviors.
  *
  * When used in a list-like way (with integer keys), we decode an
  * entry in a RevlogNG index file on demand. We have limited support for
  * integer-keyed insert and delete, only at elements right before the
  * end.
  *
  * With string keys, we lazily perform a reverse mapping from node to
  * rev, using a base-16 trie.
  */
 struct indexObjectStruct {
 	PyObject_HEAD
 	    /* Type-specific fields go here. */
 	    PyObject *data;     /* raw bytes of index */
 	Py_buffer buf;          /* buffer of data */
 	PyObject **cache;       /* cached tuples */
 	const char **offsets;   /* populated on demand */
 	Py_ssize_t raw_length;  /* original number of elements */
 	Py_ssize_t length;      /* current number of elements */
 	PyObject *added;        /* populated on demand */
 	PyObject *headrevs;     /* cache, invalidated on changes */
 	PyObject *filteredrevs; /* filtered revs set */
 	nodetree nt;            /* base-16 trie */
 	int ntinitialized;      /* 0 or 1 */
 	int ntrev;              /* last rev scanned */
 	int ntlookups;          /* # lookups */
 	int ntmisses;           /* # lookups that miss the cache */
 	int inlined;
 };
 
 static Py_ssize_t index_length(const indexObject *self)
 {
 	if (self->added == NULL)
 		return self->length;
 	return self->length + PyList_GET_SIZE(self->added);
 }
 
 static PyObject *nullentry = NULL;
 static const char nullid[20] = {0};
 static const Py_ssize_t nullrev = -1;
 
 static Py_ssize_t inline_scan(indexObject *self, const char **offsets);
 
 #if LONG_MAX == 0x7fffffffL
 static const char *const tuple_format = PY23("Kiiiiiis#", "Kiiiiiiy#");
 #else
 static const char *const tuple_format = PY23("kiiiiiis#", "kiiiiiiy#");
 #endif
 
 /* A RevlogNG v1 index entry is 64 bytes long. */
 static const long v1_hdrsize = 64;
 
 static void raise_revlog_error(void)
 {
 	PyObject *mod = NULL, *dict = NULL, *errclass = NULL;
 
 	mod = PyImport_ImportModule("mercurial.error");
 	if (mod == NULL) {
 		goto cleanup;
 	}
 
 	dict = PyModule_GetDict(mod);
 	if (dict == NULL) {
 		goto cleanup;
 	}
 	Py_INCREF(dict);
 
 	errclass = PyDict_GetItemString(dict, "RevlogError");
 	if (errclass == NULL) {
 		PyErr_SetString(PyExc_SystemError,
 		                "could not find RevlogError");
 		goto cleanup;
 	}
 
 	/* value of exception is ignored by callers */
 	PyErr_SetString(errclass, "RevlogError");
 
 cleanup:
 	Py_XDECREF(dict);
 	Py_XDECREF(mod);
 }
 
 /*
  * Return a pointer to the beginning of a RevlogNG record.
  */
 static const char *index_deref(indexObject *self, Py_ssize_t pos)
 {
 	if (self->inlined && pos > 0) {
 		if (self->offsets == NULL) {
 			self->offsets = PyMem_Malloc(self->raw_length *
 			                             sizeof(*self->offsets));
 			if (self->offsets == NULL)
 				return (const char *)PyErr_NoMemory();
-			inline_scan(self, self->offsets);
+			Py_ssize_t ret = inline_scan(self, self->offsets);
+			if (ret == -1) {
+				return NULL;
+			};
 		}
 		return self->offsets[pos];
 	}
 
 	return (const char *)(self->buf.buf) + pos * v1_hdrsize;
 }
 
 /*
  * Get parents of the given rev.
  *
  * The specified rev must be valid and must not be nullrev. A returned
  * parent revision may be nullrev, but is guaranteed to be in valid range.
  */
 static inline int index_get_parents(indexObject *self, Py_ssize_t rev, int *ps,
                                     int maxrev)
 {
 	if (rev >= self->length) {
 		long tmp;
 		PyObject *tuple =
 		    PyList_GET_ITEM(self->added, rev - self->length);
 		if (!pylong_to_long(PyTuple_GET_ITEM(tuple, 5), &tmp)) {
 			return -1;
 		}
 		ps[0] = (int)tmp;
 		if (!pylong_to_long(PyTuple_GET_ITEM(tuple, 6), &tmp)) {
 			return -1;
 		}
 		ps[1] = (int)tmp;
 	} else {
 		const char *data = index_deref(self, rev);
 		ps[0] = getbe32(data + 24);
 		ps[1] = getbe32(data + 28);
 	}
 	/* If index file is corrupted, ps[] may point to invalid revisions. So
 	 * there is a risk of buffer overflow to trust them unconditionally. */
 	if (ps[0] < -1 || ps[0] > maxrev || ps[1] < -1 || ps[1] > maxrev) {
 		PyErr_SetString(PyExc_ValueError, "parent out of range");
 		return -1;
 	}
 	return 0;
 }
 
 /*
  * Get parents of the given rev.
  *
  * If the specified rev is out of range, IndexError will be raised. If the
  * revlog entry is corrupted, ValueError may be raised.
  *
  * Returns 0 on success or -1 on failure.
  */
 static int HgRevlogIndex_GetParents(PyObject *op, int rev, int *ps)
 {
 	int tiprev;
 	if (!op || !HgRevlogIndex_Check(op) || !ps) {
 		PyErr_BadInternalCall();
 		return -1;
 	}
 	tiprev = (int)index_length((indexObject *)op) - 1;
 	if (rev < -1 || rev > tiprev) {
 		PyErr_Format(PyExc_IndexError, "rev out of range: %d", rev);
 		return -1;
 	} else if (rev == -1) {
 		ps[0] = ps[1] = -1;
 		return 0;
 	} else {
 		return index_get_parents((indexObject *)op, rev, ps, tiprev);
 	}
 }
 
 static inline int64_t index_get_start(indexObject *self, Py_ssize_t rev)
 {
 	uint64_t offset;
 	if (rev == nullrev) {
 		return 0;
 	}
 	if (rev >= self->length) {
 		PyObject *tuple;
 		PyObject *pylong;
 		PY_LONG_LONG tmp;
 		tuple = PyList_GET_ITEM(self->added, rev - self->length);
 		pylong = PyTuple_GET_ITEM(tuple, 0);
 		tmp = PyLong_AsLongLong(pylong);
 		if (tmp == -1 && PyErr_Occurred()) {
 			return -1;
 		}
 		if (tmp < 0) {
 			PyErr_Format(PyExc_OverflowError,
 			             "revlog entry size out of bound (%lld)",
 			             (long long)tmp);
 			return -1;
 		}
 		offset = (uint64_t)tmp;
 	} else {
 		const char *data = index_deref(self, rev);
 		offset = getbe32(data + 4);
 		if (rev == 0) {
 			/* mask out version number for the first entry */
 			offset &= 0xFFFF;
 		} else {
 			uint32_t offset_high = getbe32(data);
 			offset |= ((uint64_t)offset_high) << 32;
 		}
 	}
 	return (int64_t)(offset >> 16);
 }
 
 static inline int index_get_length(indexObject *self, Py_ssize_t rev)
 {
 	if (rev == nullrev) {
 		return 0;
 	}
 	if (rev >= self->length) {
 		PyObject *tuple;
 		PyObject *pylong;
 		long ret;
 		tuple = PyList_GET_ITEM(self->added, rev - self->length);
 		pylong = PyTuple_GET_ITEM(tuple, 1);
 		ret = PyInt_AsLong(pylong);
 		if (ret == -1 && PyErr_Occurred()) {
 			return -1;
 		}
 		if (ret < 0 || ret > (long)INT_MAX) {
 			PyErr_Format(PyExc_OverflowError,
 			             "revlog entry size out of bound (%ld)",
 			             ret);
 			return -1;
 		}
 		return (int)ret;
 	} else {
 		const char *data = index_deref(self, rev);
 		int tmp = (int)getbe32(data + 8);
 		if (tmp < 0) {
 			PyErr_Format(PyExc_OverflowError,
 			             "revlog entry size out of bound (%d)",
 			             tmp);
 			return -1;
 		}
 		return tmp;
 	}
 }
 
 /*
  * RevlogNG format (all in big endian, data may be inlined):
  *    6 bytes: offset
  *    2 bytes: flags
  *    4 bytes: compressed length
  *    4 bytes: uncompressed length
  *    4 bytes: base revision
  *    4 bytes: link revision
  *    4 bytes: parent 1 revision
  *    4 bytes: parent 2 revision
  *   32 bytes: nodeid (only 20 bytes used)
  */
 static PyObject *index_get(indexObject *self, Py_ssize_t pos)
 {
 	uint64_t offset_flags;
 	int comp_len, uncomp_len, base_rev, link_rev, parent_1, parent_2;
 	const char *c_node_id;
 	const char *data;
 	Py_ssize_t length = index_length(self);
 	PyObject *entry;
 
 	if (pos == nullrev) {
 		Py_INCREF(nullentry);
 		return nullentry;
 	}
 
 	if (pos < 0 || pos >= length) {
 		PyErr_SetString(PyExc_IndexError, "revlog index out of range");
 		return NULL;
 	}
 
 	if (pos >= self->length) {
 		PyObject *obj;
 		obj = PyList_GET_ITEM(self->added, pos - self->length);
 		Py_INCREF(obj);
 		return obj;
 	}
 
 	if (self->cache) {
 		if (self->cache[pos]) {
 			Py_INCREF(self->cache[pos]);
 			return self->cache[pos];
 		}
 	} else {
 		self->cache = calloc(self->raw_length, sizeof(PyObject *));
 		if (self->cache == NULL)
 			return PyErr_NoMemory();
 	}
 
 	data = index_deref(self, pos);
 	if (data == NULL)
 		return NULL;
 
 	offset_flags = getbe32(data + 4);
 	if (pos == 0) /* mask out version number for the first entry */
 		offset_flags &= 0xFFFF;
 	else {
 		uint32_t offset_high = getbe32(data);
 		offset_flags |= ((uint64_t)offset_high) << 32;
 	}
 
 	comp_len = getbe32(data + 8);
 	uncomp_len = getbe32(data + 12);
 	base_rev = getbe32(data + 16);
 	link_rev = getbe32(data + 20);
 	parent_1 = getbe32(data + 24);
 	parent_2 = getbe32(data + 28);
 	c_node_id = data + 32;
 
 	entry = Py_BuildValue(tuple_format, offset_flags, comp_len, uncomp_len,
 	                      base_rev, link_rev, parent_1, parent_2, c_node_id,
 	                      (Py_ssize_t)20);
 
 	if (entry) {
 		PyObject_GC_UnTrack(entry);
 		Py_INCREF(entry);
 	}
 
 	self->cache[pos] = entry;
 
 	return entry;
 }
 
 /*
  * Return the 20-byte SHA of the node corresponding to the given rev.
  */
 static const char *index_node(indexObject *self, Py_ssize_t pos)
 {
 	Py_ssize_t length = index_length(self);
 	const char *data;
 
 	if (pos == nullrev)
 		return nullid;
 
 	if (pos >= length)
 		return NULL;
 
 	if (pos >= self->length) {
 		PyObject *tuple, *str;
 		tuple = PyList_GET_ITEM(self->added, pos - self->length);
 		str = PyTuple_GetItem(tuple, 7);
 		return str ? PyBytes_AS_STRING(str) : NULL;
 	}
 
 	data = index_deref(self, pos);
 	return data ? data + 32 : NULL;
 }
 
 /*
  * Return the 20-byte SHA of the node corresponding to the given rev. The
  * rev is assumed to be existing. If not, an exception is set.
  */
 static const char *index_node_existing(indexObject *self, Py_ssize_t pos)
 {
 	const char *node = index_node(self, pos);
 	if (node == NULL) {
 		PyErr_Format(PyExc_IndexError, "could not access rev %d",
 		             (int)pos);
 	}
 	return node;
 }
 
 static int nt_insert(nodetree *self, const char *node, int rev);
 
 static int node_check(PyObject *obj, char **node)
 {
 	Py_ssize_t nodelen;
 	if (PyBytes_AsStringAndSize(obj, node, &nodelen) == -1)
 		return -1;
 	if (nodelen == 20)
 		return 0;
 	PyErr_SetString(PyExc_ValueError, "20-byte hash required");
 	return -1;
 }
 
 static PyObject *index_append(indexObject *self, PyObject *obj)
 {
 	char *node;
 	Py_ssize_t len;
 
 	if (!PyTuple_Check(obj) || PyTuple_GET_SIZE(obj) != 8) {
 		PyErr_SetString(PyExc_TypeError, "8-tuple required");
 		return NULL;
 	}
 
 	if (node_check(PyTuple_GET_ITEM(obj, 7), &node) == -1)
 		return NULL;
 
 	len = index_length(self);
 
 	if (self->added == NULL) {
 		self->added = PyList_New(0);
 		if (self->added == NULL)
 			return NULL;
 	}
 
 	if (PyList_Append(self->added, obj) == -1)
 		return NULL;
 
 	if (self->ntinitialized)
 		nt_insert(&self->nt, node, (int)len);
 
 	Py_CLEAR(self->headrevs);
 	Py_RETURN_NONE;
 }
 
 static PyObject *index_stats(indexObject *self)
 {
 	PyObject *obj = PyDict_New();
 	PyObject *s = NULL;
 	PyObject *t = NULL;
 
 	if (obj == NULL)
 		return NULL;
 
 #define istat(__n, __d)                                                        \
 	do {                                                                   \
 		s = PyBytes_FromString(__d);                                   \
 		t = PyInt_FromSsize_t(self->__n);                              \
 		if (!s || !t)                                                  \
 			goto bail;                                             \
 		if (PyDict_SetItem(obj, s, t) == -1)                           \
 			goto bail;                                             \
 		Py_CLEAR(s);                                                   \
 		Py_CLEAR(t);                                                   \
 	} while (0)
 
 	if (self->added) {
 		Py_ssize_t len = PyList_GET_SIZE(self->added);
 		s = PyBytes_FromString("index entries added");
 		t = PyInt_FromSsize_t(len);
 		if (!s || !t)
 			goto bail;
 		if (PyDict_SetItem(obj, s, t) == -1)
 			goto bail;
 		Py_CLEAR(s);
 		Py_CLEAR(t);
 	}
 
 	if (self->raw_length != self->length)
 		istat(raw_length, "revs on disk");
 	istat(length, "revs in memory");
 	istat(ntlookups, "node trie lookups");
 	istat(ntmisses, "node trie misses");
 	istat(ntrev, "node trie last rev scanned");
 	if (self->ntinitialized) {
 		istat(nt.capacity, "node trie capacity");
 		istat(nt.depth, "node trie depth");
 		istat(nt.length, "node trie count");
 		istat(nt.splits, "node trie splits");
 	}
 
 #undef istat
 
 	return obj;
 
 bail:
 	Py_XDECREF(obj);
 	Py_XDECREF(s);
 	Py_XDECREF(t);
 	return NULL;
 }
 
 /*
  * When we cache a list, we want to be sure the caller can't mutate
  * the cached copy.
  */
 static PyObject *list_copy(PyObject *list)
 {
 	Py_ssize_t len = PyList_GET_SIZE(list);
 	PyObject *newlist = PyList_New(len);
 	Py_ssize_t i;
 
 	if (newlist == NULL)
 		return NULL;
 
 	for (i = 0; i < len; i++) {
 		PyObject *obj = PyList_GET_ITEM(list, i);
 		Py_INCREF(obj);
 		PyList_SET_ITEM(newlist, i, obj);
 	}
 
 	return newlist;
 }
 
 static int check_filter(PyObject *filter, Py_ssize_t arg)
 {
 	if (filter) {
 		PyObject *arglist, *result;
 		int isfiltered;
 
 		arglist = Py_BuildValue("(n)", arg);
 		if (!arglist) {
 			return -1;
 		}
 
 		result = PyEval_CallObject(filter, arglist);
 		Py_DECREF(arglist);
 		if (!result) {
 			return -1;
 		}
 
 		/* PyObject_IsTrue returns 1 if true, 0 if false, -1 if error,
 		 * same as this function, so we can just return it directly.*/
 		isfiltered = PyObject_IsTrue(result);
 		Py_DECREF(result);
 		return isfiltered;
 	} else {
 		return 0;
 	}
 }
 
 static Py_ssize_t add_roots_get_min(indexObject *self, PyObject *list,
                                     Py_ssize_t marker, char *phases)
 {
 	PyObject *iter = NULL;
 	PyObject *iter_item = NULL;
 	Py_ssize_t min_idx = index_length(self) + 2;
 	long iter_item_long;
 
 	if (PyList_GET_SIZE(list) != 0) {
 		iter = PyObject_GetIter(list);
 		if (iter == NULL)
 			return -2;
 		while ((iter_item = PyIter_Next(iter))) {
 			if (!pylong_to_long(iter_item, &iter_item_long)) {
 				Py_DECREF(iter_item);
 				return -2;
 			}
 			Py_DECREF(iter_item);
 			if (iter_item_long < min_idx)
 				min_idx = iter_item_long;
 			phases[iter_item_long] = (char)marker;
 		}
 		Py_DECREF(iter);
 	}
 
 	return min_idx;
 }
 
 static inline void set_phase_from_parents(char *phases, int parent_1,
                                           int parent_2, Py_ssize_t i)
 {
 	if (parent_1 >= 0 && phases[parent_1] > phases[i])
 		phases[i] = phases[parent_1];
 	if (parent_2 >= 0 && phases[parent_2] > phases[i])
 		phases[i] = phases[parent_2];
 }
 
 static PyObject *reachableroots2(indexObject *self, PyObject *args)
 {
 
 	/* Input */
 	long minroot;
 	PyObject *includepatharg = NULL;
 	int includepath = 0;
 	/* heads and roots are lists */
 	PyObject *heads = NULL;
 	PyObject *roots = NULL;
 	PyObject *reachable = NULL;
 
 	PyObject *val;
 	Py_ssize_t len = index_length(self);
 	long revnum;
 	Py_ssize_t k;
 	Py_ssize_t i;
 	Py_ssize_t l;
 	int r;
 	int parents[2];
 
 	/* Internal data structure:
 	 * tovisit: array of length len+1 (all revs + nullrev), filled upto
 	 * lentovisit
 	 *
 	 * revstates: array of length len+1 (all revs + nullrev) */
 	int *tovisit = NULL;
 	long lentovisit = 0;
 	enum { RS_SEEN = 1, RS_ROOT = 2, RS_REACHABLE = 4 };
 	char *revstates = NULL;
 
 	/* Get arguments */
 	if (!PyArg_ParseTuple(args, "lO!O!O!", &minroot, &PyList_Type, &heads,
 	                      &PyList_Type, &roots, &PyBool_Type,
 	                      &includepatharg))
 		goto bail;
 
 	if (includepatharg == Py_True)
 		includepath = 1;
 
 	/* Initialize return set */
 	reachable = PyList_New(0);
 	if (reachable == NULL)
 		goto bail;
 
 	/* Initialize internal datastructures */
 	tovisit = (int *)malloc((len + 1) * sizeof(int));
 	if (tovisit == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	revstates = (char *)calloc(len + 1, 1);
 	if (revstates == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	l = PyList_GET_SIZE(roots);
 	for (i = 0; i < l; i++) {
 		revnum = PyInt_AsLong(PyList_GET_ITEM(roots, i));
 		if (revnum == -1 && PyErr_Occurred())
 			goto bail;
 		/* If root is out of range, e.g. wdir(), it must be unreachable
 		 * from heads. So we can just ignore it. */
 		if (revnum + 1 < 0 || revnum + 1 >= len + 1)
 			continue;
 		revstates[revnum + 1] |= RS_ROOT;
 	}
 
 	/* Populate tovisit with all the heads */
 	l = PyList_GET_SIZE(heads);
 	for (i = 0; i < l; i++) {
 		revnum = PyInt_AsLong(PyList_GET_ITEM(heads, i));
 		if (revnum == -1 && PyErr_Occurred())
 			goto bail;
 		if (revnum + 1 < 0 || revnum + 1 >= len + 1) {
 			PyErr_SetString(PyExc_IndexError, "head out of range");
 			goto bail;
 		}
 		if (!(revstates[revnum + 1] & RS_SEEN)) {
 			tovisit[lentovisit++] = (int)revnum;
 			revstates[revnum + 1] |= RS_SEEN;
 		}
 	}
 
 	/* Visit the tovisit list and find the reachable roots */
 	k = 0;
 	while (k < lentovisit) {
 		/* Add the node to reachable if it is a root*/
 		revnum = tovisit[k++];
 		if (revstates[revnum + 1] & RS_ROOT) {
 			revstates[revnum + 1] |= RS_REACHABLE;
 			val = PyInt_FromLong(revnum);
 			if (val == NULL)
 				goto bail;
 			r = PyList_Append(reachable, val);
 			Py_DECREF(val);
 			if (r < 0)
 				goto bail;
 			if (includepath == 0)
 				continue;
 		}
 
 		/* Add its parents to the list of nodes to visit */
 		if (revnum == nullrev)
 			continue;
 		r = index_get_parents(self, revnum, parents, (int)len - 1);
 		if (r < 0)
 			goto bail;
 		for (i = 0; i < 2; i++) {
 			if (!(revstates[parents[i] + 1] & RS_SEEN) &&
 			    parents[i] >= minroot) {
 				tovisit[lentovisit++] = parents[i];
 				revstates[parents[i] + 1] |= RS_SEEN;
 			}
 		}
 	}
 
 	/* Find all the nodes in between the roots we found and the heads
 	 * and add them to the reachable set */
 	if (includepath == 1) {
 		long minidx = minroot;
 		if (minidx < 0)
 			minidx = 0;
 		for (i = minidx; i < len; i++) {
 			if (!(revstates[i + 1] & RS_SEEN))
 				continue;
 			r = index_get_parents(self, i, parents, (int)len - 1);
 			/* Corrupted index file, error is set from
 			 * index_get_parents */
 			if (r < 0)
 				goto bail;
 			if (((revstates[parents[0] + 1] |
 			      revstates[parents[1] + 1]) &
 			     RS_REACHABLE) &&
 			    !(revstates[i + 1] & RS_REACHABLE)) {
 				revstates[i + 1] |= RS_REACHABLE;
 				val = PyInt_FromSsize_t(i);
 				if (val == NULL)
 					goto bail;
 				r = PyList_Append(reachable, val);
 				Py_DECREF(val);
 				if (r < 0)
 					goto bail;
 			}
 		}
 	}
 
 	free(revstates);
 	free(tovisit);
 	return reachable;
 bail:
 	Py_XDECREF(reachable);
 	free(revstates);
 	free(tovisit);
 	return NULL;
 }
 
 static PyObject *compute_phases_map_sets(indexObject *self, PyObject *args)
 {
 	PyObject *roots = Py_None;
 	PyObject *ret = NULL;
 	PyObject *phasessize = NULL;
 	PyObject *phaseroots = NULL;
 	PyObject *phaseset = NULL;
 	PyObject *phasessetlist = NULL;
 	PyObject *rev = NULL;
 	Py_ssize_t len = index_length(self);
 	Py_ssize_t numphase = 0;
 	Py_ssize_t minrevallphases = 0;
 	Py_ssize_t minrevphase = 0;
 	Py_ssize_t i = 0;
 	char *phases = NULL;
 	long phase;
 
 	if (!PyArg_ParseTuple(args, "O", &roots))
 		goto done;
 	if (roots == NULL || !PyList_Check(roots)) {
 		PyErr_SetString(PyExc_TypeError, "roots must be a list");
 		goto done;
 	}
 
 	phases = calloc(
 	    len, 1); /* phase per rev: {0: public, 1: draft, 2: secret} */
 	if (phases == NULL) {
 		PyErr_NoMemory();
 		goto done;
 	}
 	/* Put the phase information of all the roots in phases */
 	numphase = PyList_GET_SIZE(roots) + 1;
 	minrevallphases = len + 1;
 	phasessetlist = PyList_New(numphase);
 	if (phasessetlist == NULL)
 		goto done;
 
 	PyList_SET_ITEM(phasessetlist, 0, Py_None);
 	Py_INCREF(Py_None);
 
 	for (i = 0; i < numphase - 1; i++) {
 		phaseroots = PyList_GET_ITEM(roots, i);
 		phaseset = PySet_New(NULL);
 		if (phaseset == NULL)
 			goto release;
 		PyList_SET_ITEM(phasessetlist, i + 1, phaseset);
 		if (!PyList_Check(phaseroots)) {
 			PyErr_SetString(PyExc_TypeError,
 			                "roots item must be a list");
 			goto release;
 		}
 		minrevphase =
 		    add_roots_get_min(self, phaseroots, i + 1, phases);
 		if (minrevphase == -2) /* Error from add_roots_get_min */
 			goto release;
 		minrevallphases = MIN(minrevallphases, minrevphase);
 	}
 	/* Propagate the phase information from the roots to the revs */
 	if (minrevallphases != -1) {
 		int parents[2];
 		for (i = minrevallphases; i < len; i++) {
 			if (index_get_parents(self, i, parents, (int)len - 1) <
 			    0)
 				goto release;
 			set_phase_from_parents(phases, parents[0], parents[1],
 			                       i);
 		}
 	}
 	/* Transform phase list to a python list */
 	phasessize = PyInt_FromSsize_t(len);
 	if (phasessize == NULL)
 		goto release;
 	for (i = 0; i < len; i++) {
 		phase = phases[i];
 		/* We only store the sets of phase for non public phase, the
 		 * public phase is computed as a difference */
 		if (phase != 0) {
 			phaseset = PyList_GET_ITEM(phasessetlist, phase);
 			rev = PyInt_FromSsize_t(i);
 			if (rev == NULL)
 				goto release;
 			PySet_Add(phaseset, rev);
 			Py_XDECREF(rev);
 		}
 	}
 	ret = PyTuple_Pack(2, phasessize, phasessetlist);
 
 release:
 	Py_XDECREF(phasessize);
 	Py_XDECREF(phasessetlist);
 done:
 	free(phases);
 	return ret;
 }
 
 static PyObject *index_headrevs(indexObject *self, PyObject *args)
 {
 	Py_ssize_t i, j, len;
 	char *nothead = NULL;
 	PyObject *heads = NULL;
 	PyObject *filter = NULL;
 	PyObject *filteredrevs = Py_None;
 
 	if (!PyArg_ParseTuple(args, "|O", &filteredrevs)) {
 		return NULL;
 	}
 
 	if (self->headrevs && filteredrevs == self->filteredrevs)
 		return list_copy(self->headrevs);
 
 	Py_DECREF(self->filteredrevs);
 	self->filteredrevs = filteredrevs;
 	Py_INCREF(filteredrevs);
 
 	if (filteredrevs != Py_None) {
 		filter = PyObject_GetAttrString(filteredrevs, "__contains__");
 		if (!filter) {
 			PyErr_SetString(
 			    PyExc_TypeError,
 			    "filteredrevs has no attribute __contains__");
 			goto bail;
 		}
 	}
 
 	len = index_length(self);
 	heads = PyList_New(0);
 	if (heads == NULL)
 		goto bail;
 	if (len == 0) {
 		PyObject *nullid = PyInt_FromLong(-1);
 		if (nullid == NULL || PyList_Append(heads, nullid) == -1) {
 			Py_XDECREF(nullid);
 			goto bail;
 		}
 		goto done;
 	}
 
 	nothead = calloc(len, 1);
 	if (nothead == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	for (i = len - 1; i >= 0; i--) {
 		int isfiltered;
 		int parents[2];
 
 		/* If nothead[i] == 1, it means we've seen an unfiltered child
 		 * of this node already, and therefore this node is not
 		 * filtered. So we can skip the expensive check_filter step.
 		 */
 		if (nothead[i] != 1) {
 			isfiltered = check_filter(filter, i);
 			if (isfiltered == -1) {
 				PyErr_SetString(PyExc_TypeError,
 				                "unable to check filter");
 				goto bail;
 			}
 
 			if (isfiltered) {
 				nothead[i] = 1;
 				continue;
 			}
 		}
 
 		if (index_get_parents(self, i, parents, (int)len - 1) < 0)
 			goto bail;
 		for (j = 0; j < 2; j++) {
 			if (parents[j] >= 0)
 				nothead[parents[j]] = 1;
 		}
 	}
 
 	for (i = 0; i < len; i++) {
 		PyObject *head;
 
 		if (nothead[i])
 			continue;
 		head = PyInt_FromSsize_t(i);
 		if (head == NULL || PyList_Append(heads, head) == -1) {
 			Py_XDECREF(head);
 			goto bail;
 		}
 	}
 
 done:
 	self->headrevs = heads;
 	Py_XDECREF(filter);
 	free(nothead);
 	return list_copy(self->headrevs);
 bail:
 	Py_XDECREF(filter);
 	Py_XDECREF(heads);
 	free(nothead);
 	return NULL;
 }
 
 /**
  * Obtain the base revision index entry.
  *
  * Callers must ensure that rev >= 0 or illegal memory access may occur.
  */
 static inline int index_baserev(indexObject *self, int rev)
 {
 	const char *data;
 	int result;
 
 	if (rev >= self->length) {
 		PyObject *tuple =
 		    PyList_GET_ITEM(self->added, rev - self->length);
 		long ret;
 		if (!pylong_to_long(PyTuple_GET_ITEM(tuple, 3), &ret)) {
 			return -2;
 		}
 		result = (int)ret;
 	} else {
 		data = index_deref(self, rev);
 		if (data == NULL) {
 			return -2;
 		}
 
 		result = getbe32(data + 16);
 	}
 	if (result > rev) {
 		PyErr_Format(
 		    PyExc_ValueError,
 		    "corrupted revlog, revision base above revision: %d, %d",
 		    rev, result);
 		return -2;
 	}
 	if (result < -1) {
 		PyErr_Format(
 		    PyExc_ValueError,
 		    "corrupted revlog, revision base out of range: %d, %d", rev,
 		    result);
 		return -2;
 	}
 	return result;
 }
 
 /**
  * Find if a revision is a snapshot or not
  *
  * Only relevant for sparse-revlog case.
  * Callers must ensure that rev is in a valid range.
  */
 static int index_issnapshotrev(indexObject *self, Py_ssize_t rev)
 {
 	int ps[2];
 	Py_ssize_t base;
 	while (rev >= 0) {
 		base = (Py_ssize_t)index_baserev(self, rev);
 		if (base == rev) {
 			base = -1;
 		}
 		if (base == -2) {
 			assert(PyErr_Occurred());
 			return -1;
 		}
 		if (base == -1) {
 			return 1;
 		}
 		if (index_get_parents(self, rev, ps, (int)rev) < 0) {
 			assert(PyErr_Occurred());
 			return -1;
 		};
 		if (base == ps[0] || base == ps[1]) {
 			return 0;
 		}
 		rev = base;
 	}
 	return rev == -1;
 }
 
 static PyObject *index_issnapshot(indexObject *self, PyObject *value)
 {
 	long rev;
 	int issnap;
 	Py_ssize_t length = index_length(self);
 
 	if (!pylong_to_long(value, &rev)) {
 		return NULL;
 	}
 	if (rev < -1 || rev >= length) {
 		PyErr_Format(PyExc_ValueError, "revlog index out of range: %ld",
 		             rev);
 		return NULL;
 	};
 	issnap = index_issnapshotrev(self, (Py_ssize_t)rev);
 	if (issnap < 0) {
 		return NULL;
 	};
 	return PyBool_FromLong((long)issnap);
 }
 
 static PyObject *index_findsnapshots(indexObject *self, PyObject *args)
 {
 	Py_ssize_t start_rev;
 	PyObject *cache;
 	Py_ssize_t base;
 	Py_ssize_t rev;
 	PyObject *key = NULL;
 	PyObject *value = NULL;
 	const Py_ssize_t length = index_length(self);
 	if (!PyArg_ParseTuple(args, "O!n", &PyDict_Type, &cache, &start_rev)) {
 		return NULL;
 	}
 	for (rev = start_rev; rev < length; rev++) {
 		int issnap;
 		PyObject *allvalues = NULL;
 		issnap = index_issnapshotrev(self, rev);
 		if (issnap < 0) {
 			goto bail;
 		}
 		if (issnap == 0) {
 			continue;
 		}
 		base = (Py_ssize_t)index_baserev(self, rev);
 		if (base == rev) {
 			base = -1;
 		}
 		if (base == -2) {
 			assert(PyErr_Occurred());
 			goto bail;
 		}
 		key = PyInt_FromSsize_t(base);
 		allvalues = PyDict_GetItem(cache, key);
 		if (allvalues == NULL && PyErr_Occurred()) {
 			goto bail;
 		}
 		if (allvalues == NULL) {
 			int r;
 			allvalues = PyList_New(0);
 			if (!allvalues) {
 				goto bail;
 			}
 			r = PyDict_SetItem(cache, key, allvalues);
 			Py_DECREF(allvalues);
 			if (r < 0) {
 				goto bail;
 			}
 		}
 		value = PyInt_FromSsize_t(rev);
 		if (PyList_Append(allvalues, value)) {
 			goto bail;
 		}
 		Py_CLEAR(key);
 		Py_CLEAR(value);
 	}
 	Py_RETURN_NONE;
 bail:
 	Py_XDECREF(key);
 	Py_XDECREF(value);
 	return NULL;
 }
 
 static PyObject *index_deltachain(indexObject *self, PyObject *args)
 {
 	int rev, generaldelta;
 	PyObject *stoparg;
 	int stoprev, iterrev, baserev = -1;
 	int stopped;
 	PyObject *chain = NULL, *result = NULL;
 	const Py_ssize_t length = index_length(self);
 
 	if (!PyArg_ParseTuple(args, "iOi", &rev, &stoparg, &generaldelta)) {
 		return NULL;
 	}
 
 	if (PyInt_Check(stoparg)) {
 		stoprev = (int)PyInt_AsLong(stoparg);
 		if (stoprev == -1 && PyErr_Occurred()) {
 			return NULL;
 		}
 	} else if (stoparg == Py_None) {
 		stoprev = -2;
 	} else {
 		PyErr_SetString(PyExc_ValueError,
 		                "stoprev must be integer or None");
 		return NULL;
 	}
 
 	if (rev < 0 || rev >= length) {
 		PyErr_SetString(PyExc_ValueError, "revlog index out of range");
 		return NULL;
 	}
 
 	chain = PyList_New(0);
 	if (chain == NULL) {
 		return NULL;
 	}
 
 	baserev = index_baserev(self, rev);
 
 	/* This should never happen. */
 	if (baserev <= -2) {
 		/* Error should be set by index_deref() */
 		assert(PyErr_Occurred());
 		goto bail;
 	}
 
 	iterrev = rev;
 
 	while (iterrev != baserev && iterrev != stoprev) {
 		PyObject *value = PyInt_FromLong(iterrev);
 		if (value == NULL) {
 			goto bail;
 		}
 		if (PyList_Append(chain, value)) {
 			Py_DECREF(value);
 			goto bail;
 		}
 		Py_DECREF(value);
 
 		if (generaldelta) {
 			iterrev = baserev;
 		} else {
 			iterrev--;
 		}
 
 		if (iterrev < 0) {
 			break;
 		}
 
 		if (iterrev >= length) {
 			PyErr_SetString(PyExc_IndexError,
 			                "revision outside index");
 			return NULL;
 		}
 
 		baserev = index_baserev(self, iterrev);
 
 		/* This should never happen. */
 		if (baserev <= -2) {
 			/* Error should be set by index_deref() */
 			assert(PyErr_Occurred());
 			goto bail;
 		}
 	}
 
 	if (iterrev == stoprev) {
 		stopped = 1;
 	} else {
 		PyObject *value = PyInt_FromLong(iterrev);
 		if (value == NULL) {
 			goto bail;
 		}
 		if (PyList_Append(chain, value)) {
 			Py_DECREF(value);
 			goto bail;
 		}
 		Py_DECREF(value);
 
 		stopped = 0;
 	}
 
 	if (PyList_Reverse(chain)) {
 		goto bail;
 	}
 
 	result = Py_BuildValue("OO", chain, stopped ? Py_True : Py_False);
 	Py_DECREF(chain);
 	return result;
 
 bail:
 	Py_DECREF(chain);
 	return NULL;
 }
 
 static inline int64_t
 index_segment_span(indexObject *self, Py_ssize_t start_rev, Py_ssize_t end_rev)
 {
 	int64_t start_offset;
 	int64_t end_offset;
 	int end_size;
 	start_offset = index_get_start(self, start_rev);
 	if (start_offset < 0) {
 		return -1;
 	}
 	end_offset = index_get_start(self, end_rev);
 	if (end_offset < 0) {
 		return -1;
 	}
 	end_size = index_get_length(self, end_rev);
 	if (end_size < 0) {
 		return -1;
 	}
 	if (end_offset < start_offset) {
 		PyErr_Format(PyExc_ValueError,
 		             "corrupted revlog index: inconsistent offset "
 		             "between revisions (%zd) and (%zd)",
 		             start_rev, end_rev);
 		return -1;
 	}
 	return (end_offset - start_offset) + (int64_t)end_size;
 }
 
 /* returns endidx so that revs[startidx:endidx] has no empty trailing revs */
 static Py_ssize_t trim_endidx(indexObject *self, const Py_ssize_t *revs,
                               Py_ssize_t startidx, Py_ssize_t endidx)
 {
 	int length;
 	while (endidx > 1 && endidx > startidx) {
 		length = index_get_length(self, revs[endidx - 1]);
 		if (length < 0) {
 			return -1;
 		}
 		if (length != 0) {
 			break;
 		}
 		endidx -= 1;
 	}
 	return endidx;
 }
 
 struct Gap {
 	int64_t size;
 	Py_ssize_t idx;
 };
 
 static int gap_compare(const void *left, const void *right)
 {
 	const struct Gap *l_left = ((const struct Gap *)left);
 	const struct Gap *l_right = ((const struct Gap *)right);
 	if (l_left->size < l_right->size) {
 		return -1;
 	} else if (l_left->size > l_right->size) {
 		return 1;
 	}
 	return 0;
 }
 static int Py_ssize_t_compare(const void *left, const void *right)
 {
 	const Py_ssize_t l_left = *(const Py_ssize_t *)left;
 	const Py_ssize_t l_right = *(const Py_ssize_t *)right;
 	if (l_left < l_right) {
 		return -1;
 	} else if (l_left > l_right) {
 		return 1;
 	}
 	return 0;
 }
 
 static PyObject *index_slicechunktodensity(indexObject *self, PyObject *args)
 {
 	/* method arguments */
 	PyObject *list_revs = NULL; /* revisions in the chain */
 	double targetdensity = 0;   /* min density to achieve */
 	Py_ssize_t mingapsize = 0;  /* threshold to ignore gaps */
 
 	/* other core variables */
 	Py_ssize_t idxlen = index_length(self);
 	Py_ssize_t i;            /* used for various iteration */
 	PyObject *result = NULL; /* the final return of the function */
 
 	/* generic information about the delta chain being slice */
 	Py_ssize_t num_revs = 0;    /* size of the full delta chain */
 	Py_ssize_t *revs = NULL;    /* native array of revision in the chain */
 	int64_t chainpayload = 0;   /* sum of all delta in the chain */
 	int64_t deltachainspan = 0; /* distance from first byte to last byte */
 
 	/* variable used for slicing the delta chain */
 	int64_t readdata = 0; /* amount of data currently planned to be read */
 	double density = 0;   /* ration of payload data compared to read ones */
 	int64_t previous_end;
 	struct Gap *gaps = NULL; /* array of notable gap in the chain */
 	Py_ssize_t num_gaps =
 	    0; /* total number of notable gap recorded so far */
 	Py_ssize_t *selected_indices = NULL; /* indices of gap skipped over */
 	Py_ssize_t num_selected = 0;         /* number of gaps skipped */
 	PyObject *chunk = NULL;              /* individual slice */
 	PyObject *allchunks = NULL;          /* all slices */
 	Py_ssize_t previdx;
 
 	/* parsing argument */
 	if (!PyArg_ParseTuple(args, "O!dn", &PyList_Type, &list_revs,
 	                      &targetdensity, &mingapsize)) {
 		goto bail;
 	}
 
 	/* If the delta chain contains a single element, we do not need slicing
 	 */
 	num_revs = PyList_GET_SIZE(list_revs);
 	if (num_revs <= 1) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 
 	/* Turn the python list into a native integer array (for efficiency) */
 	revs = (Py_ssize_t *)calloc(num_revs, sizeof(Py_ssize_t));
 	if (revs == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 	for (i = 0; i < num_revs; i++) {
 		Py_ssize_t revnum = PyInt_AsLong(PyList_GET_ITEM(list_revs, i));
 		if (revnum == -1 && PyErr_Occurred()) {
 			goto bail;
 		}
 		if (revnum < nullrev || revnum >= idxlen) {
 			PyErr_Format(PyExc_IndexError,
 			             "index out of range: %zd", revnum);
 			goto bail;
 		}
 		revs[i] = revnum;
 	}
 
 	/* Compute and check various property of the unsliced delta chain */
 	deltachainspan = index_segment_span(self, revs[0], revs[num_revs - 1]);
 	if (deltachainspan < 0) {
 		goto bail;
 	}
 
 	if (deltachainspan <= mingapsize) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 	chainpayload = 0;
 	for (i = 0; i < num_revs; i++) {
 		int tmp = index_get_length(self, revs[i]);
 		if (tmp < 0) {
 			goto bail;
 		}
 		chainpayload += tmp;
 	}
 
 	readdata = deltachainspan;
 	density = 1.0;
 
 	if (0 < deltachainspan) {
 		density = (double)chainpayload / (double)deltachainspan;
 	}
 
 	if (density >= targetdensity) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 
 	/* if chain is too sparse, look for relevant gaps */
 	gaps = (struct Gap *)calloc(num_revs, sizeof(struct Gap));
 	if (gaps == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	previous_end = -1;
 	for (i = 0; i < num_revs; i++) {
 		int64_t revstart;
 		int revsize;
 		revstart = index_get_start(self, revs[i]);
 		if (revstart < 0) {
 			goto bail;
 		};
 		revsize = index_get_length(self, revs[i]);
 		if (revsize < 0) {
 			goto bail;
 		};
 		if (revsize == 0) {
 			continue;
 		}
 		if (previous_end >= 0) {
 			int64_t gapsize = revstart - previous_end;
 			if (gapsize > mingapsize) {
 				gaps[num_gaps].size = gapsize;
 				gaps[num_gaps].idx = i;
 				num_gaps += 1;
 			}
 		}
 		previous_end = revstart + revsize;
 	}
 	if (num_gaps == 0) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 	qsort(gaps, num_gaps, sizeof(struct Gap), &gap_compare);
 
 	/* Slice the largest gap first, they improve the density the most */
 	selected_indices =
 	    (Py_ssize_t *)malloc((num_gaps + 1) * sizeof(Py_ssize_t));
 	if (selected_indices == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	for (i = num_gaps - 1; i >= 0; i--) {
 		selected_indices[num_selected] = gaps[i].idx;
 		readdata -= gaps[i].size;
 		num_selected += 1;
 		if (readdata <= 0) {
 			density = 1.0;
 		} else {
 			density = (double)chainpayload / (double)readdata;
 		}
 		if (density >= targetdensity) {
 			break;
 		}
 	}
 	qsort(selected_indices, num_selected, sizeof(Py_ssize_t),
 	      &Py_ssize_t_compare);
 
 	/* create the resulting slice */
 	allchunks = PyList_New(0);
 	if (allchunks == NULL) {
 		goto bail;
 	}
 	previdx = 0;
 	selected_indices[num_selected] = num_revs;
 	for (i = 0; i <= num_selected; i++) {
 		Py_ssize_t idx = selected_indices[i];
 		Py_ssize_t endidx = trim_endidx(self, revs, previdx, idx);
 		if (endidx < 0) {
 			goto bail;
 		}
 		if (previdx < endidx) {
 			chunk = PyList_GetSlice(list_revs, previdx, endidx);
 			if (chunk == NULL) {
 				goto bail;
 			}
 			if (PyList_Append(allchunks, chunk) == -1) {
 				goto bail;
 			}
 			Py_DECREF(chunk);
 			chunk = NULL;
 		}
 		previdx = idx;
 	}
 	result = allchunks;
 	goto done;
 
 bail:
 	Py_XDECREF(allchunks);
 	Py_XDECREF(chunk);
 done:
 	free(revs);
 	free(gaps);
 	free(selected_indices);
 	return result;
 }
 
 static inline int nt_level(const char *node, Py_ssize_t level)
 {
 	int v = node[level >> 1];
 	if (!(level & 1))
 		v >>= 4;
 	return v & 0xf;
 }
 
 /*
  * Return values:
  *
  *   -4: match is ambiguous (multiple candidates)
  *   -2: not found
  * rest: valid rev
  */
 static int nt_find(nodetree *self, const char *node, Py_ssize_t nodelen,
                    int hex)
 {
 	int (*getnybble)(const char *, Py_ssize_t) = hex ? hexdigit : nt_level;
 	int level, maxlevel, off;
 
 	if (nodelen == 20 && node[0] == '\0' && memcmp(node, nullid, 20) == 0)
 		return -1;
 
 	if (hex)
 		maxlevel = nodelen > 40 ? 40 : (int)nodelen;
 	else
 		maxlevel = nodelen > 20 ? 40 : ((int)nodelen * 2);
 
 	for (level = off = 0; level < maxlevel; level++) {
 		int k = getnybble(node, level);
 		nodetreenode *n = &self->nodes[off];
 		int v = n->children[k];
 
 		if (v < 0) {
 			const char *n;
 			Py_ssize_t i;
 
 			v = -(v + 2);
 			n = index_node(self->index, v);
 			if (n == NULL)
 				return -2;
 			for (i = level; i < maxlevel; i++)
 				if (getnybble(node, i) != nt_level(n, i))
 					return -2;
 			return v;
 		}
 		if (v == 0)
 			return -2;
 		off = v;
 	}
 	/* multiple matches against an ambiguous prefix */
 	return -4;
 }
 
 static int nt_new(nodetree *self)
 {
 	if (self->length == self->capacity) {
 		unsigned newcapacity;
 		nodetreenode *newnodes;
 		newcapacity = self->capacity * 2;
 		if (newcapacity >= INT_MAX / sizeof(nodetreenode)) {
 			PyErr_SetString(PyExc_MemoryError,
 			                "overflow in nt_new");
 			return -1;
 		}
 		newnodes =
 		    realloc(self->nodes, newcapacity * sizeof(nodetreenode));
 		if (newnodes == NULL) {
 			PyErr_SetString(PyExc_MemoryError, "out of memory");
 			return -1;
 		}
 		self->capacity = newcapacity;
 		self->nodes = newnodes;
 		memset(&self->nodes[self->length], 0,
 		       sizeof(nodetreenode) * (self->capacity - self->length));
 	}
 	return self->length++;
 }
 
 static int nt_insert(nodetree *self, const char *node, int rev)
 {
 	int level = 0;
 	int off = 0;
 
 	while (level < 40) {
 		int k = nt_level(node, level);
 		nodetreenode *n;
 		int v;
 
 		n = &self->nodes[off];
 		v = n->children[k];
 
 		if (v == 0) {
 			n->children[k] = -rev - 2;
 			return 0;
 		}
 		if (v < 0) {
 			const char *oldnode =
 			    index_node_existing(self->index, -(v + 2));
 			int noff;
 
 			if (oldnode == NULL)
 				return -1;
 			if (!memcmp(oldnode, node, 20)) {
 				n->children[k] = -rev - 2;
 				return 0;
 			}
 			noff = nt_new(self);
 			if (noff == -1)
 				return -1;
 			/* self->nodes may have been changed by realloc */
 			self->nodes[off].children[k] = noff;
 			off = noff;
 			n = &self->nodes[off];
 			n->children[nt_level(oldnode, ++level)] = v;
 			if (level > self->depth)
 				self->depth = level;
 			self->splits += 1;
 		} else {
 			level += 1;
 			off = v;
 		}
 	}
 
 	return -1;
 }
 
 static PyObject *ntobj_insert(nodetreeObject *self, PyObject *args)
 {
 	Py_ssize_t rev;
 	const char *node;
 	Py_ssize_t length;
 	if (!PyArg_ParseTuple(args, "n", &rev))
 		return NULL;
 	length = index_length(self->nt.index);
 	if (rev < 0 || rev >= length) {
 		PyErr_SetString(PyExc_ValueError, "revlog index out of range");
 		return NULL;
 	}
 	node = index_node_existing(self->nt.index, rev);
 	if (nt_insert(&self->nt, node, (int)rev) == -1)
 		return NULL;
 	Py_RETURN_NONE;
 }
 
 static int nt_delete_node(nodetree *self, const char *node)
 {
 	/* rev==-2 happens to get encoded as 0, which is interpreted as not set
 	 */
 	return nt_insert(self, node, -2);
 }
 
 static int nt_init(nodetree *self, indexObject *index, unsigned capacity)
 {
 	/* Initialize before overflow-checking to avoid nt_dealloc() crash. */
 	self->nodes = NULL;
 
 	self->index = index;
 	/* The input capacity is in terms of revisions, while the field is in
 	 * terms of nodetree nodes. */
 	self->capacity = (capacity < 4 ? 4 : capacity / 2);
 	self->depth = 0;
 	self->splits = 0;
 	if ((size_t)self->capacity > INT_MAX / sizeof(nodetreenode)) {
 		PyErr_SetString(PyExc_ValueError, "overflow in init_nt");
 		return -1;
 	}
 	self->nodes = calloc(self->capacity, sizeof(nodetreenode));
 	if (self->nodes == NULL) {
 		PyErr_NoMemory();
 		return -1;
 	}
 	self->length = 1;
 	return 0;
 }
 
 static int ntobj_init(nodetreeObject *self, PyObject *args)
 {
 	PyObject *index;
 	unsigned capacity;
 	if (!PyArg_ParseTuple(args, "O!I", &HgRevlogIndex_Type, &index,
 	                      &capacity))
 		return -1;
 	Py_INCREF(index);
 	return nt_init(&self->nt, (indexObject *)index, capacity);
 }
 
 static int nt_partialmatch(nodetree *self, const char *node, Py_ssize_t nodelen)
 {
 	return nt_find(self, node, nodelen, 1);
 }
 
 /*
  * Find the length of the shortest unique prefix of node.
  *
  * Return values:
  *
  *   -3: error (exception set)
  *   -2: not found (no exception set)
  * rest: length of shortest prefix
  */
 static int nt_shortest(nodetree *self, const char *node)
 {
 	int level, off;
 
 	for (level = off = 0; level < 40; level++) {
 		int k, v;
 		nodetreenode *n = &self->nodes[off];
 		k = nt_level(node, level);
 		v = n->children[k];
 		if (v < 0) {
 			const char *n;
 			v = -(v + 2);
 			n = index_node_existing(self->index, v);
 			if (n == NULL)
 				return -3;
 			if (memcmp(node, n, 20) != 0)
 				/*
 				 * Found a unique prefix, but it wasn't for the
 				 * requested node (i.e the requested node does
 				 * not exist).
 				 */
 				return -2;
 			return level + 1;
 		}
 		if (v == 0)
 			return -2;
 		off = v;
 	}
 	/*
 	 * The node was still not unique after 40 hex digits, so this won't
 	 * happen. Also, if we get here, then there's a programming error in
 	 * this file that made us insert a node longer than 40 hex digits.
 	 */
 	PyErr_SetString(PyExc_Exception, "broken node tree");
 	return -3;
 }
 
 static PyObject *ntobj_shortest(nodetreeObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int length;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(val, &node) == -1)
 		return NULL;
 
 	length = nt_shortest(&self->nt, node);
 	if (length == -3)
 		return NULL;
 	if (length == -2) {
 		raise_revlog_error();
 		return NULL;
 	}
 	return PyInt_FromLong(length);
 }
 
 static void nt_dealloc(nodetree *self)
 {
 	free(self->nodes);
 	self->nodes = NULL;
 }
 
 static void ntobj_dealloc(nodetreeObject *self)
 {
 	Py_XDECREF(self->nt.index);
 	nt_dealloc(&self->nt);
 	PyObject_Del(self);
 }
 
 static PyMethodDef ntobj_methods[] = {
     {"insert", (PyCFunction)ntobj_insert, METH_VARARGS,
      "insert an index entry"},
     {"shortest", (PyCFunction)ntobj_shortest, METH_VARARGS,
      "find length of shortest hex nodeid of a binary ID"},
     {NULL} /* Sentinel */
 };
 
 static PyTypeObject nodetreeType = {
     PyVarObject_HEAD_INIT(NULL, 0) /* header */
     "parsers.nodetree",            /* tp_name */
     sizeof(nodetreeObject),        /* tp_basicsize */
     0,                             /* tp_itemsize */
     (destructor)ntobj_dealloc,     /* tp_dealloc */
     0,                             /* tp_print */
     0,                             /* tp_getattr */
     0,                             /* tp_setattr */
     0,                             /* tp_compare */
     0,                             /* tp_repr */
     0,                             /* tp_as_number */
     0,                             /* tp_as_sequence */
     0,                             /* tp_as_mapping */
     0,                             /* tp_hash */
     0,                             /* tp_call */
     0,                             /* tp_str */
     0,                             /* tp_getattro */
     0,                             /* tp_setattro */
     0,                             /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,            /* tp_flags */
     "nodetree",                    /* tp_doc */
     0,                             /* tp_traverse */
     0,                             /* tp_clear */
     0,                             /* tp_richcompare */
     0,                             /* tp_weaklistoffset */
     0,                             /* tp_iter */
     0,                             /* tp_iternext */
     ntobj_methods,                 /* tp_methods */
     0,                             /* tp_members */
     0,                             /* tp_getset */
     0,                             /* tp_base */
     0,                             /* tp_dict */
     0,                             /* tp_descr_get */
     0,                             /* tp_descr_set */
     0,                             /* tp_dictoffset */
     (initproc)ntobj_init,          /* tp_init */
     0,                             /* tp_alloc */
 };
 
 static int index_init_nt(indexObject *self)
 {
 	if (!self->ntinitialized) {
 		if (nt_init(&self->nt, self, (int)self->raw_length) == -1) {
 			nt_dealloc(&self->nt);
 			return -1;
 		}
 		if (nt_insert(&self->nt, nullid, -1) == -1) {
 			nt_dealloc(&self->nt);
 			return -1;
 		}
 		self->ntinitialized = 1;
 		self->ntrev = (int)index_length(self);
 		self->ntlookups = 1;
 		self->ntmisses = 0;
 	}
 	return 0;
 }
 
 /*
  * Return values:
  *
  *   -3: error (exception set)
  *   -2: not found (no exception set)
  * rest: valid rev
  */
 static int index_find_node(indexObject *self, const char *node,
                            Py_ssize_t nodelen)
 {
 	int rev;
 
 	if (index_init_nt(self) == -1)
 		return -3;
 
 	self->ntlookups++;
 	rev = nt_find(&self->nt, node, nodelen, 0);
 	if (rev >= -1)
 		return rev;
 
 	/*
 	 * For the first handful of lookups, we scan the entire index,
 	 * and cache only the matching nodes. This optimizes for cases
 	 * like "hg tip", where only a few nodes are accessed.
 	 *
 	 * After that, we cache every node we visit, using a single
 	 * scan amortized over multiple lookups.  This gives the best
 	 * bulk performance, e.g. for "hg log".
 	 */
 	if (self->ntmisses++ < 4) {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -3;
 			if (memcmp(node, n, nodelen > 20 ? 20 : nodelen) == 0) {
 				if (nt_insert(&self->nt, n, rev) == -1)
 					return -3;
 				break;
 			}
 		}
 	} else {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -3;
 			if (nt_insert(&self->nt, n, rev) == -1) {
 				self->ntrev = rev + 1;
 				return -3;
 			}
 			if (memcmp(node, n, nodelen > 20 ? 20 : nodelen) == 0) {
 				break;
 			}
 		}
 		self->ntrev = rev;
 	}
 
 	if (rev >= 0)
 		return rev;
 	return -2;
 }
 
 static PyObject *index_getitem(indexObject *self, PyObject *value)
 {
 	char *node;
 	int rev;
 
 	if (PyInt_Check(value)) {
 		long idx;
 		if (!pylong_to_long(value, &idx)) {
 			return NULL;
 		}
 		return index_get(self, idx);
 	}
 
 	if (node_check(value, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node, 20);
 	if (rev >= -1)
 		return PyInt_FromLong(rev);
 	if (rev == -2)
 		raise_revlog_error();
 	return NULL;
 }
 
 /*
  * Fully populate the radix tree.
  */
 static int index_populate_nt(indexObject *self)
 {
 	int rev;
 	if (self->ntrev > 0) {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -1;
 			if (nt_insert(&self->nt, n, rev) == -1)
 				return -1;
 		}
 		self->ntrev = -1;
 	}
 	return 0;
 }
 
 static PyObject *index_partialmatch(indexObject *self, PyObject *args)
 {
 	const char *fullnode;
 	Py_ssize_t nodelen;
 	char *node;
 	int rev, i;
 
 	if (!PyArg_ParseTuple(args, PY23("s#", "y#"), &node, &nodelen))
 		return NULL;
 
 	if (nodelen < 1) {
 		PyErr_SetString(PyExc_ValueError, "key too short");
 		return NULL;
 	}
 
 	if (nodelen > 40) {
 		PyErr_SetString(PyExc_ValueError, "key too long");
 		return NULL;
 	}
 
 	for (i = 0; i < nodelen; i++)
 		hexdigit(node, i);
 	if (PyErr_Occurred()) {
 		/* input contains non-hex characters */
 		PyErr_Clear();
 		Py_RETURN_NONE;
 	}
 
 	if (index_init_nt(self) == -1)
 		return NULL;
 	if (index_populate_nt(self) == -1)
 		return NULL;
 	rev = nt_partialmatch(&self->nt, node, nodelen);
 
 	switch (rev) {
 	case -4:
 		raise_revlog_error();
 		return NULL;
 	case -2:
 		Py_RETURN_NONE;
 	case -1:
 		return PyBytes_FromStringAndSize(nullid, 20);
 	}
 
 	fullnode = index_node_existing(self, rev);
 	if (fullnode == NULL) {
 		return NULL;
 	}
 	return PyBytes_FromStringAndSize(fullnode, 20);
 }
 
 static PyObject *index_shortest(indexObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int length;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(val, &node) == -1)
 		return NULL;
 
 	self->ntlookups++;
 	if (index_init_nt(self) == -1)
 		return NULL;
 	if (index_populate_nt(self) == -1)
 		return NULL;
 	length = nt_shortest(&self->nt, node);
 	if (length == -3)
 		return NULL;
 	if (length == -2) {
 		raise_revlog_error();
 		return NULL;
 	}
 	return PyInt_FromLong(length);
 }
 
 static PyObject *index_m_get(indexObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int rev;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(val, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node, 20);
 	if (rev == -3)
 		return NULL;
 	if (rev == -2)
 		Py_RETURN_NONE;
 	return PyInt_FromLong(rev);
 }
 
 static int index_contains(indexObject *self, PyObject *value)
 {
 	char *node;
 
 	if (PyInt_Check(value)) {
 		long rev;
 		if (!pylong_to_long(value, &rev)) {
 			return -1;
 		}
 		return rev >= -1 && rev < index_length(self);
 	}
 
 	if (node_check(value, &node) == -1)
 		return -1;
 
 	switch (index_find_node(self, node, 20)) {
 	case -3:
 		return -1;
 	case -2:
 		return 0;
 	default:
 		return 1;
 	}
 }
 
 static PyObject *index_m_has_node(indexObject *self, PyObject *args)
 {
 	int ret = index_contains(self, args);
 	if (ret < 0)
 		return NULL;
 	return PyBool_FromLong((long)ret);
 }
 
 static PyObject *index_m_rev(indexObject *self, PyObject *val)
 {
 	char *node;
 	int rev;
 
 	if (node_check(val, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node, 20);
 	if (rev >= -1)
 		return PyInt_FromLong(rev);
 	if (rev == -2)
 		raise_revlog_error();
 	return NULL;
 }
 
 typedef uint64_t bitmask;
 
 /*
  * Given a disjoint set of revs, return all candidates for the
  * greatest common ancestor. In revset notation, this is the set
  * "heads(::a and ::b and ...)"
  */
 static PyObject *find_gca_candidates(indexObject *self, const int *revs,
                                      int revcount)
 {
 	const bitmask allseen = (1ull << revcount) - 1;
 	const bitmask poison = 1ull << revcount;
 	PyObject *gca = PyList_New(0);
 	int i, v, interesting;
 	int maxrev = -1;
 	bitmask sp;
 	bitmask *seen;
 
 	if (gca == NULL)
 		return PyErr_NoMemory();
 
 	for (i = 0; i < revcount; i++) {
 		if (revs[i] > maxrev)
 			maxrev = revs[i];
 	}
 
 	seen = calloc(sizeof(*seen), maxrev + 1);
 	if (seen == NULL) {
 		Py_DECREF(gca);
 		return PyErr_NoMemory();
 	}
 
 	for (i = 0; i < revcount; i++)
 		seen[revs[i]] = 1ull << i;
 
 	interesting = revcount;
 
 	for (v = maxrev; v >= 0 && interesting; v--) {
 		bitmask sv = seen[v];
 		int parents[2];
 
 		if (!sv)
 			continue;
 
 		if (sv < poison) {
 			interesting -= 1;
 			if (sv == allseen) {
 				PyObject *obj = PyInt_FromLong(v);
 				if (obj == NULL)
 					goto bail;
 				if (PyList_Append(gca, obj) == -1) {
 					Py_DECREF(obj);
 					goto bail;
 				}
 				sv |= poison;
 				for (i = 0; i < revcount; i++) {
 					if (revs[i] == v)
 						goto done;
 				}
 			}
 		}
 		if (index_get_parents(self, v, parents, maxrev) < 0)
 			goto bail;
 
 		for (i = 0; i < 2; i++) {
 			int p = parents[i];
 			if (p == -1)
 				continue;
 			sp = seen[p];
 			if (sv < poison) {
 				if (sp == 0) {
 					seen[p] = sv;
 					interesting++;
 				} else if (sp != sv)
 					seen[p] |= sv;
 			} else {
 				if (sp && sp < poison)
 					interesting--;
 				seen[p] = sv;
 			}
 		}
 	}
 
 done:
 	free(seen);
 	return gca;
 bail:
 	free(seen);
 	Py_XDECREF(gca);
 	return NULL;
 }
 
 /*
  * Given a disjoint set of revs, return the subset with the longest
  * path to the root.
  */
 static PyObject *find_deepest(indexObject *self, PyObject *revs)
 {
 	const Py_ssize_t revcount = PyList_GET_SIZE(revs);
 	static const Py_ssize_t capacity = 24;
 	int *depth, *interesting = NULL;
 	int i, j, v, ninteresting;
 	PyObject *dict = NULL, *keys = NULL;
 	long *seen = NULL;
 	int maxrev = -1;
 	long final;
 
 	if (revcount > capacity) {
 		PyErr_Format(PyExc_OverflowError,
 		             "bitset size (%ld) > capacity (%ld)",
 		             (long)revcount, (long)capacity);
 		return NULL;
 	}
 
 	for (i = 0; i < revcount; i++) {
 		int n = (int)PyInt_AsLong(PyList_GET_ITEM(revs, i));
 		if (n > maxrev)
 			maxrev = n;
 	}
 
 	depth = calloc(sizeof(*depth), maxrev + 1);
 	if (depth == NULL)
 		return PyErr_NoMemory();
 
 	seen = calloc(sizeof(*seen), maxrev + 1);
 	if (seen == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	interesting = calloc(sizeof(*interesting), ((size_t)1) << revcount);
 	if (interesting == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	if (PyList_Sort(revs) == -1)
 		goto bail;
 
 	for (i = 0; i < revcount; i++) {
 		int n = (int)PyInt_AsLong(PyList_GET_ITEM(revs, i));
 		long b = 1l << i;
 		depth[n] = 1;
 		seen[n] = b;
 		interesting[b] = 1;
 	}
 
 	/* invariant: ninteresting is the number of non-zero entries in
 	 * interesting. */
 	ninteresting = (int)revcount;
 
 	for (v = maxrev; v >= 0 && ninteresting > 1; v--) {
 		int dv = depth[v];
 		int parents[2];
 		long sv;
 
 		if (dv == 0)
 			continue;
 
 		sv = seen[v];
 		if (index_get_parents(self, v, parents, maxrev) < 0)
 			goto bail;
 
 		for (i = 0; i < 2; i++) {
 			int p = parents[i];
 			long sp;
 			int dp;
 
 			if (p == -1)
 				continue;
 
 			dp = depth[p];
 			sp = seen[p];
 			if (dp <= dv) {
 				depth[p] = dv + 1;
 				if (sp != sv) {
 					interesting[sv] += 1;
 					seen[p] = sv;
 					if (sp) {
 						interesting[sp] -= 1;
 						if (interesting[sp] == 0)
 							ninteresting -= 1;
 					}
 				}
 			} else if (dv == dp - 1) {
 				long nsp = sp | sv;
 				if (nsp == sp)
 					continue;
 				seen[p] = nsp;
 				interesting[sp] -= 1;
 				if (interesting[sp] == 0)
 					ninteresting -= 1;
 				if (interesting[nsp] == 0)
 					ninteresting += 1;
 				interesting[nsp] += 1;
 			}
 		}
 		interesting[sv] -= 1;
 		if (interesting[sv] == 0)
 			ninteresting -= 1;
 	}
 
 	final = 0;
 	j = ninteresting;
 	for (i = 0; i < (int)(2 << revcount) && j > 0; i++) {
 		if (interesting[i] == 0)
 			continue;
 		final |= i;
 		j -= 1;
 	}
 	if (final == 0) {
 		keys = PyList_New(0);
 		goto bail;
 	}
 
 	dict = PyDict_New();
 	if (dict == NULL)
 		goto bail;
 
 	for (i = 0; i < revcount; i++) {
 		PyObject *key;
 
 		if ((final & (1 << i)) == 0)
 			continue;
 
 		key = PyList_GET_ITEM(revs, i);
 		Py_INCREF(key);
 		Py_INCREF(Py_None);
 		if (PyDict_SetItem(dict, key, Py_None) == -1) {
 			Py_DECREF(key);
 			Py_DECREF(Py_None);
 			goto bail;
 		}
 	}
 
 	keys = PyDict_Keys(dict);
 
 bail:
 	free(depth);
 	free(seen);
 	free(interesting);
 	Py_XDECREF(dict);
 
 	return keys;
 }
 
 /*
  * Given a (possibly overlapping) set of revs, return all the
  * common ancestors heads: heads(::args[0] and ::a[1] and ...)
  */
 static PyObject *index_commonancestorsheads(indexObject *self, PyObject *args)
 {
 	PyObject *ret = NULL;
 	Py_ssize_t argcount, i, len;
 	bitmask repeat = 0;
 	int revcount = 0;
 	int *revs;
 
 	argcount = PySequence_Length(args);
 	revs = PyMem_Malloc(argcount * sizeof(*revs));
 	if (argcount > 0 && revs == NULL)
 		return PyErr_NoMemory();
 	len = index_length(self);
 
 	for (i = 0; i < argcount; i++) {
 		static const int capacity = 24;
 		PyObject *obj = PySequence_GetItem(args, i);
 		bitmask x;
 		long val;
 
 		if (!PyInt_Check(obj)) {
 			PyErr_SetString(PyExc_TypeError,
 			                "arguments must all be ints");
 			Py_DECREF(obj);
 			goto bail;
 		}
 		val = PyInt_AsLong(obj);
 		Py_DECREF(obj);
 		if (val == -1) {
 			ret = PyList_New(0);
 			goto done;
 		}
 		if (val < 0 || val >= len) {
 			PyErr_SetString(PyExc_IndexError, "index out of range");
 			goto bail;
 		}
 		/* this cheesy bloom filter lets us avoid some more
 		 * expensive duplicate checks in the common set-is-disjoint
 		 * case */
 		x = 1ull << (val & 0x3f);
 		if (repeat & x) {
 			int k;
 			for (k = 0; k < revcount; k++) {
 				if (val == revs[k])
 					goto duplicate;
 			}
 		} else
 			repeat |= x;
 		if (revcount >= capacity) {
 			PyErr_Format(PyExc_OverflowError,
 			             "bitset size (%d) > capacity (%d)",
 			             revcount, capacity);
 			goto bail;
 		}
 		revs[revcount++] = (int)val;
 	duplicate:;
 	}
 
 	if (revcount == 0) {
 		ret = PyList_New(0);
 		goto done;
 	}
 	if (revcount == 1) {
 		PyObject *obj;
 		ret = PyList_New(1);
 		if (ret == NULL)
 			goto bail;
 		obj = PyInt_FromLong(revs[0]);
 		if (obj == NULL)
 			goto bail;
 		PyList_SET_ITEM(ret, 0, obj);
 		goto done;
 	}
 
 	ret = find_gca_candidates(self, revs, revcount);
 	if (ret == NULL)
 		goto bail;
 
 done:
 	PyMem_Free(revs);
 	return ret;
 
 bail:
 	PyMem_Free(revs);
 	Py_XDECREF(ret);
 	return NULL;
 }
 
 /*
  * Given a (possibly overlapping) set of revs, return the greatest
  * common ancestors: those with the longest path to the root.
  */
 static PyObject *index_ancestors(indexObject *self, PyObject *args)
 {
 	PyObject *ret;
 	PyObject *gca = index_commonancestorsheads(self, args);
 	if (gca == NULL)
 		return NULL;
 
 	if (PyList_GET_SIZE(gca) <= 1) {
 		return gca;
 	}
 
 	ret = find_deepest(self, gca);
 	Py_DECREF(gca);
 	return ret;
 }
 
 /*
  * Invalidate any trie entries introduced by added revs.
  */
 static void index_invalidate_added(indexObject *self, Py_ssize_t start)
 {
 	Py_ssize_t i, len = PyList_GET_SIZE(self->added);
 
 	for (i = start; i < len; i++) {
 		PyObject *tuple = PyList_GET_ITEM(self->added, i);
 		PyObject *node = PyTuple_GET_ITEM(tuple, 7);
 
 		nt_delete_node(&self->nt, PyBytes_AS_STRING(node));
 	}
 
 	if (start == 0)
 		Py_CLEAR(self->added);
 }
 
 /*
  * Delete a numeric range of revs, which must be at the end of the
  * range.
  */
 static int index_slice_del(indexObject *self, PyObject *item)
 {
 	Py_ssize_t start, stop, step, slicelength;
 	Py_ssize_t length = index_length(self) + 1;
 	int ret = 0;
 
 /* Argument changed from PySliceObject* to PyObject* in Python 3. */
 #ifdef IS_PY3K
 	if (PySlice_GetIndicesEx(item, length, &start, &stop, &step,
 	                         &slicelength) < 0)
 #else
 	if (PySlice_GetIndicesEx((PySliceObject *)item, length, &start, &stop,
 	                         &step, &slicelength) < 0)
 #endif
 		return -1;
 
 	if (slicelength <= 0)
 		return 0;
 
 	if ((step < 0 && start < stop) || (step > 0 && start > stop))
 		stop = start;
 
 	if (step < 0) {
 		stop = start + 1;
 		start = stop + step * (slicelength - 1) - 1;
 		step = -step;
 	}
 
 	if (step != 1) {
 		PyErr_SetString(PyExc_ValueError,
 		                "revlog index delete requires step size of 1");
 		return -1;
 	}
 
 	if (stop != length - 1) {
 		PyErr_SetString(PyExc_IndexError,
 		                "revlog index deletion indices are invalid");
 		return -1;
 	}
 
 	if (start < self->length) {
 		if (self->ntinitialized) {
 			Py_ssize_t i;
 
 			for (i = start; i < self->length; i++) {
 				const char *node = index_node_existing(self, i);
 				if (node == NULL)
 					return -1;
 
 				nt_delete_node(&self->nt, node);
 			}
 			if (self->added)
 				index_invalidate_added(self, 0);
 			if (self->ntrev > start)
 				self->ntrev = (int)start;
 		} else if (self->added) {
 			Py_CLEAR(self->added);
 		}
 
 		self->length = start;
 		if (start < self->raw_length) {
 			if (self->cache) {
 				Py_ssize_t i;
 				for (i = start; i < self->raw_length; i++)
 					Py_CLEAR(self->cache[i]);
 			}
 			self->raw_length = start;
 		}
 		goto done;
 	}
 
 	if (self->ntinitialized) {
 		index_invalidate_added(self, start - self->length);
 		if (self->ntrev > start)
 			self->ntrev = (int)start;
 	}
 	if (self->added)
 		ret = PyList_SetSlice(self->added, start - self->length,
 		                      PyList_GET_SIZE(self->added), NULL);
 done:
 	Py_CLEAR(self->headrevs);
 	return ret;
 }
 
 /*
  * Supported ops:
  *
  * slice deletion
  * string assignment (extend node->rev mapping)
  * string deletion (shrink node->rev mapping)
  */
 static int index_assign_subscript(indexObject *self, PyObject *item,
                                   PyObject *value)
 {
 	char *node;
 	long rev;
 
 	if (PySlice_Check(item) && value == NULL)
 		return index_slice_del(self, item);
 
 	if (node_check(item, &node) == -1)
 		return -1;
 
 	if (value == NULL)
 		return self->ntinitialized ? nt_delete_node(&self->nt, node)
 		                           : 0;
 	rev = PyInt_AsLong(value);
 	if (rev > INT_MAX || rev < 0) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "rev out of range");
 		return -1;
 	}
 
 	if (index_init_nt(self) == -1)
 		return -1;
 	return nt_insert(&self->nt, node, (int)rev);
 }
 
 /*
  * Find all RevlogNG entries in an index that has inline data. Update
  * the optional "offsets" table with those entries.
  */
 static Py_ssize_t inline_scan(indexObject *self, const char **offsets)
 {
 	const char *data = (const char *)self->buf.buf;
 	Py_ssize_t pos = 0;
 	Py_ssize_t end = self->buf.len;
 	long incr = v1_hdrsize;
 	Py_ssize_t len = 0;
 
 	while (pos + v1_hdrsize <= end && pos >= 0) {
 		uint32_t comp_len;
 		/* 3rd element of header is length of compressed inline data */
 		comp_len = getbe32(data + pos + 8);
 		incr = v1_hdrsize + comp_len;
 		if (offsets)
 			offsets[len] = data + pos;
 		len++;
 		pos += incr;
 	}
 
 	if (pos != end) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 		return -1;
 	}
 
 	return len;
 }
 
 static int index_init(indexObject *self, PyObject *args)
 {
 	PyObject *data_obj, *inlined_obj;
 	Py_ssize_t size;
 
 	/* Initialize before argument-checking to avoid index_dealloc() crash.
 	 */
 	self->raw_length = 0;
 	self->added = NULL;
 	self->cache = NULL;
 	self->data = NULL;
 	memset(&self->buf, 0, sizeof(self->buf));
 	self->headrevs = NULL;
 	self->filteredrevs = Py_None;
 	Py_INCREF(Py_None);
 	self->ntinitialized = 0;
 	self->offsets = NULL;
 
 	if (!PyArg_ParseTuple(args, "OO", &data_obj, &inlined_obj))
 		return -1;
 	if (!PyObject_CheckBuffer(data_obj)) {
 		PyErr_SetString(PyExc_TypeError,
 		                "data does not support buffer interface");
 		return -1;
 	}
 
 	if (PyObject_GetBuffer(data_obj, &self->buf, PyBUF_SIMPLE) == -1)
 		return -1;
 	size = self->buf.len;
 
 	self->inlined = inlined_obj && PyObject_IsTrue(inlined_obj);
 	self->data = data_obj;
 
 	self->ntlookups = self->ntmisses = 0;
 	self->ntrev = -1;
 	Py_INCREF(self->data);
 
 	if (self->inlined) {
 		Py_ssize_t len = inline_scan(self, NULL);
 		if (len == -1)
 			goto bail;
 		self->raw_length = len;
 		self->length = len;
 	} else {
 		if (size % v1_hdrsize) {
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 			goto bail;
 		}
 		self->raw_length = size / v1_hdrsize;
 		self->length = self->raw_length;
 	}
 
 	return 0;
 bail:
 	return -1;
 }
 
 static PyObject *index_nodemap(indexObject *self)
 {
 	Py_INCREF(self);
 	return (PyObject *)self;
 }
 
 static void _index_clearcaches(indexObject *self)
 {
 	if (self->cache) {
 		Py_ssize_t i;
 
 		for (i = 0; i < self->raw_length; i++)
 			Py_CLEAR(self->cache[i]);
 		free(self->cache);
 		self->cache = NULL;
 	}
 	if (self->offsets) {
 		PyMem_Free((void *)self->offsets);
 		self->offsets = NULL;
 	}
 	if (self->ntinitialized) {
 		nt_dealloc(&self->nt);
 	}
 	self->ntinitialized = 0;
 	Py_CLEAR(self->headrevs);
 }
 
 static PyObject *index_clearcaches(indexObject *self)
 {
 	_index_clearcaches(self);
 	self->ntrev = -1;
 	self->ntlookups = self->ntmisses = 0;
 	Py_RETURN_NONE;
 }
 
 static void index_dealloc(indexObject *self)
 {
 	_index_clearcaches(self);
 	Py_XDECREF(self->filteredrevs);
 	if (self->buf.buf) {
 		PyBuffer_Release(&self->buf);
 		memset(&self->buf, 0, sizeof(self->buf));
 	}
 	Py_XDECREF(self->data);
 	Py_XDECREF(self->added);
 	PyObject_Del(self);
 }
 
 static PySequenceMethods index_sequence_methods = {
     (lenfunc)index_length,      /* sq_length */
     0,                          /* sq_concat */
     0,                          /* sq_repeat */
     (ssizeargfunc)index_get,    /* sq_item */
     0,                          /* sq_slice */
     0,                          /* sq_ass_item */
     0,                          /* sq_ass_slice */
     (objobjproc)index_contains, /* sq_contains */
 };
 
 static PyMappingMethods index_mapping_methods = {
     (lenfunc)index_length,                 /* mp_length */
     (binaryfunc)index_getitem,             /* mp_subscript */
     (objobjargproc)index_assign_subscript, /* mp_ass_subscript */
 };
 
 static PyMethodDef index_methods[] = {
     {"ancestors", (PyCFunction)index_ancestors, METH_VARARGS,
      "return the gca set of the given revs"},
     {"commonancestorsheads", (PyCFunction)index_commonancestorsheads,
      METH_VARARGS,
      "return the heads of the common ancestors of the given revs"},
     {"clearcaches", (PyCFunction)index_clearcaches, METH_NOARGS,
      "clear the index caches"},
     {"get", (PyCFunction)index_m_get, METH_VARARGS, "get an index entry"},
     {"get_rev", (PyCFunction)index_m_get, METH_VARARGS,
      "return `rev` associated with a node or None"},
     {"has_node", (PyCFunction)index_m_has_node, METH_O,
      "return True if the node exist in the index"},
     {"rev", (PyCFunction)index_m_rev, METH_O,
      "return `rev` associated with a node or raise RevlogError"},
     {"computephasesmapsets", (PyCFunction)compute_phases_map_sets, METH_VARARGS,
      "compute phases"},
     {"reachableroots2", (PyCFunction)reachableroots2, METH_VARARGS,
      "reachableroots"},
     {"headrevs", (PyCFunction)index_headrevs, METH_VARARGS,
      "get head revisions"}, /* Can do filtering since 3.2 */
     {"headrevsfiltered", (PyCFunction)index_headrevs, METH_VARARGS,
      "get filtered head revisions"}, /* Can always do filtering */
     {"issnapshot", (PyCFunction)index_issnapshot, METH_O,
      "True if the object is a snapshot"},
     {"findsnapshots", (PyCFunction)index_findsnapshots, METH_VARARGS,
      "Gather snapshot data in a cache dict"},
     {"deltachain", (PyCFunction)index_deltachain, METH_VARARGS,
      "determine revisions with deltas to reconstruct fulltext"},
     {"slicechunktodensity", (PyCFunction)index_slicechunktodensity,
      METH_VARARGS, "determine revisions with deltas to reconstruct fulltext"},
     {"append", (PyCFunction)index_append, METH_O, "append an index entry"},
     {"partialmatch", (PyCFunction)index_partialmatch, METH_VARARGS,
      "match a potentially ambiguous node ID"},
     {"shortest", (PyCFunction)index_shortest, METH_VARARGS,
      "find length of shortest hex nodeid of a binary ID"},
     {"stats", (PyCFunction)index_stats, METH_NOARGS, "stats for the index"},
     {NULL} /* Sentinel */
 };
 
 static PyGetSetDef index_getset[] = {
     {"nodemap", (getter)index_nodemap, NULL, "nodemap", NULL},
     {NULL} /* Sentinel */
 };
 
 PyTypeObject HgRevlogIndex_Type = {
     PyVarObject_HEAD_INIT(NULL, 0) /* header */
     "parsers.index",               /* tp_name */
     sizeof(indexObject),           /* tp_basicsize */
     0,                             /* tp_itemsize */
     (destructor)index_dealloc,     /* tp_dealloc */
     0,                             /* tp_print */
     0,                             /* tp_getattr */
     0,                             /* tp_setattr */
     0,                             /* tp_compare */
     0,                             /* tp_repr */
     0,                             /* tp_as_number */
     &index_sequence_methods,       /* tp_as_sequence */
     &index_mapping_methods,        /* tp_as_mapping */
     0,                             /* tp_hash */
     0,                             /* tp_call */
     0,                             /* tp_str */
     0,                             /* tp_getattro */
     0,                             /* tp_setattro */
     0,                             /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,            /* tp_flags */
     "revlog index",                /* tp_doc */
     0,                             /* tp_traverse */
     0,                             /* tp_clear */
     0,                             /* tp_richcompare */
     0,                             /* tp_weaklistoffset */
     0,                             /* tp_iter */
     0,                             /* tp_iternext */
     index_methods,                 /* tp_methods */
     0,                             /* tp_members */
     index_getset,                  /* tp_getset */
     0,                             /* tp_base */
     0,                             /* tp_dict */
     0,                             /* tp_descr_get */
     0,                             /* tp_descr_set */
     0,                             /* tp_dictoffset */
     (initproc)index_init,          /* tp_init */
     0,                             /* tp_alloc */
 };
 
 /*
  * returns a tuple of the form (index, index, cache) with elements as
  * follows:
  *
  * index: an index object that lazily parses RevlogNG records
  * cache: if data is inlined, a tuple (0, index_file_content), else None
  *        index_file_content could be a string, or a buffer
  *
  * added complications are for backwards compatibility
  */
 PyObject *parse_index2(PyObject *self, PyObject *args)
 {
 	PyObject *tuple = NULL, *cache = NULL;
 	indexObject *idx;
 	int ret;
 
 	idx = PyObject_New(indexObject, &HgRevlogIndex_Type);
 	if (idx == NULL)
 		goto bail;
 
 	ret = index_init(idx, args);
 	if (ret == -1)
 		goto bail;
 
 	if (idx->inlined) {
 		cache = Py_BuildValue("iO", 0, idx->data);
 		if (cache == NULL)
 			goto bail;
 	} else {
 		cache = Py_None;
 		Py_INCREF(cache);
 	}
 
 	tuple = Py_BuildValue("NN", idx, cache);
 	if (!tuple)
 		goto bail;
 	return tuple;
 
 bail:
 	Py_XDECREF(idx);
 	Py_XDECREF(cache);
 	Py_XDECREF(tuple);
 	return NULL;
 }
 
 static Revlog_CAPI CAPI = {
     /* increment the abi_version field upon each change in the Revlog_CAPI
        struct or in the ABI of the listed functions */
     2,
     index_length,
     index_node,
     HgRevlogIndex_GetParents,
 };
 
 void revlog_module_init(PyObject *mod)
 {
 	PyObject *caps = NULL;
 	HgRevlogIndex_Type.tp_new = PyType_GenericNew;
 	if (PyType_Ready(&HgRevlogIndex_Type) < 0)
 		return;
 	Py_INCREF(&HgRevlogIndex_Type);
 	PyModule_AddObject(mod, "index", (PyObject *)&HgRevlogIndex_Type);
 
 	nodetreeType.tp_new = PyType_GenericNew;
 	if (PyType_Ready(&nodetreeType) < 0)
 		return;
 	Py_INCREF(&nodetreeType);
 	PyModule_AddObject(mod, "nodetree", (PyObject *)&nodetreeType);
 
 	if (!nullentry) {
 		nullentry =
 		    Py_BuildValue(PY23("iiiiiiis#", "iiiiiiiy#"), 0, 0, 0, -1,
 		                  -1, -1, -1, nullid, (Py_ssize_t)20);
 	}
 	if (nullentry)
 		PyObject_GC_UnTrack(nullentry);
 
 	caps = PyCapsule_New(&CAPI, "mercurial.cext.parsers.revlog_CAPI", NULL);
 	if (caps != NULL)
 		PyModule_AddObject(mod, "revlog_CAPI", caps);
 }

rust/hg-core/src/lib.rs

@@ -1,191 +1,199 b''
 // Copyright 2018-2020 Georges Racinet <[e-mail]>
 //           and Mercurial contributors
 //
 // This software may be used and distributed according to the terms of the
 // GNU General Public License version 2 or any later version.
 mod ancestors;
 pub mod dagops;
 pub use ancestors::{AncestorsIterator, LazyAncestors, MissingAncestors};
 mod dirstate;
 pub mod discovery;
 pub mod testing; // unconditionally built, for use from integration tests
 pub use dirstate::{
     dirs_multiset::{DirsMultiset, DirsMultisetIter},
     dirstate_map::DirstateMap,
     parsers::{pack_dirstate, parse_dirstate, PARENT_SIZE},
     status::{
         status, BadMatch, BadType, DirstateStatus, StatusError, StatusOptions,
     },
     CopyMap, CopyMapIter, DirstateEntry, DirstateParents, EntryState,
     StateMap, StateMapIter,
 };
 mod filepatterns;
 pub mod matchers;
 pub mod revlog;
 pub use revlog::*;
 #[cfg(feature = "with-re2")]
 pub mod re2;
 pub mod utils;
 
+// Remove this to see (potential) non-artificial compile failures. MacOS
+// *should* compile, but fail to compile tests for example as of 2020-03-06
+#[cfg(not(target_os = "linux"))]
+compile_error!(
+    "`hg-core` has only been tested on Linux and will most \
+     likely not behave correctly on other platforms."
+);
+
 use crate::utils::hg_path::{HgPathBuf, HgPathError};
 pub use filepatterns::{
     parse_pattern_syntax, read_pattern_file, IgnorePattern,
     PatternFileWarning, PatternSyntax,
 };
 use std::collections::HashMap;
 use twox_hash::RandomXxHashBuilder64;
 
 /// This is a contract between the `micro-timer` crate and us, to expose
 /// the `log` crate as `crate::log`.
 use log;
 
 pub type LineNumber = usize;
 
 /// Rust's default hasher is too slow because it tries to prevent collision
 /// attacks. We are not concerned about those: if an ill-minded person has
 /// write access to your repository, you have other issues.
 pub type FastHashMap<K, V> = HashMap<K, V, RandomXxHashBuilder64>;
 
 #[derive(Clone, Debug, PartialEq)]
 pub enum DirstateParseError {
     TooLittleData,
     Overflow,
     CorruptedEntry(String),
     Damaged,
 }
 
 impl From<std::io::Error> for DirstateParseError {
     fn from(e: std::io::Error) -> Self {
         DirstateParseError::CorruptedEntry(e.to_string())
     }
 }
 
 impl ToString for DirstateParseError {
     fn to_string(&self) -> String {
         use crate::DirstateParseError::*;
         match self {
             TooLittleData => "Too little data for dirstate.".to_string(),
             Overflow => "Overflow in dirstate.".to_string(),
             CorruptedEntry(e) => format!("Corrupted entry: {:?}.", e),
             Damaged => "Dirstate appears to be damaged.".to_string(),
         }
     }
 }
 
 #[derive(Debug, PartialEq)]
 pub enum DirstatePackError {
     CorruptedEntry(String),
     CorruptedParent,
     BadSize(usize, usize),
 }
 
 impl From<std::io::Error> for DirstatePackError {
     fn from(e: std::io::Error) -> Self {
         DirstatePackError::CorruptedEntry(e.to_string())
     }
 }
 #[derive(Debug, PartialEq)]
 pub enum DirstateMapError {
     PathNotFound(HgPathBuf),
     EmptyPath,
     InvalidPath(HgPathError),
 }
 
 impl ToString for DirstateMapError {
     fn to_string(&self) -> String {
         match self {
             DirstateMapError::PathNotFound(_) => {
                 "expected a value, found none".to_string()
             }
             DirstateMapError::EmptyPath => "Overflow in dirstate.".to_string(),
             DirstateMapError::InvalidPath(e) => e.to_string(),
         }
     }
 }
 
 #[derive(Debug)]
 pub enum DirstateError {
     Parse(DirstateParseError),
     Pack(DirstatePackError),
     Map(DirstateMapError),
     IO(std::io::Error),
 }
 
 impl From<DirstateParseError> for DirstateError {
     fn from(e: DirstateParseError) -> Self {
         DirstateError::Parse(e)
     }
 }
 
 impl From<DirstatePackError> for DirstateError {
     fn from(e: DirstatePackError) -> Self {
         DirstateError::Pack(e)
     }
 }
 
 #[derive(Debug)]
 pub enum PatternError {
     Path(HgPathError),
     UnsupportedSyntax(String),
     UnsupportedSyntaxInFile(String, String, usize),
     TooLong(usize),
     IO(std::io::Error),
     /// Needed a pattern that can be turned into a regex but got one that
     /// can't. This should only happen through programmer error.
     NonRegexPattern(IgnorePattern),
     /// This is temporary, see `re2/mod.rs`.
     /// This will cause a fallback to Python.
     Re2NotInstalled,
 }
 
 impl ToString for PatternError {
     fn to_string(&self) -> String {
         match self {
             PatternError::UnsupportedSyntax(syntax) => {
                 format!("Unsupported syntax {}", syntax)
             }
             PatternError::UnsupportedSyntaxInFile(syntax, file_path, line) => {
                 format!(
                     "{}:{}: unsupported syntax {}",
                     file_path, line, syntax
                 )
             }
             PatternError::TooLong(size) => {
                 format!("matcher pattern is too long ({} bytes)", size)
             }
             PatternError::IO(e) => e.to_string(),
             PatternError::Path(e) => e.to_string(),
             PatternError::NonRegexPattern(pattern) => {
                 format!("'{:?}' cannot be turned into a regex", pattern)
             }
             PatternError::Re2NotInstalled => {
                 "Re2 is not installed, cannot use regex functionality."
                     .to_string()
             }
         }
     }
 }
 
 impl From<DirstateMapError> for DirstateError {
     fn from(e: DirstateMapError) -> Self {
         DirstateError::Map(e)
     }
 }
 
 impl From<std::io::Error> for DirstateError {
     fn from(e: std::io::Error) -> Self {
         DirstateError::IO(e)
     }
 }
 
 impl From<std::io::Error> for PatternError {
     fn from(e: std::io::Error) -> Self {
         PatternError::IO(e)
     }
 }
 
 impl From<HgPathError> for PatternError {
     fn from(e: HgPathError) -> Self {
         PatternError::Path(e)
     }
 }

tests/run-tests.py

@@ -1,3755 +1,3755 b''
 #!/usr/bin/env python
 #
 # run-tests.py - Run a set of tests on Mercurial
 #
 # Copyright 2006 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 # Modifying this script is tricky because it has many modes:
 #   - serial (default) vs parallel (-jN, N > 1)
 #   - no coverage (default) vs coverage (-c, -C, -s)
 #   - temp install (default) vs specific hg script (--with-hg, --local)
 #   - tests are a mix of shell scripts and Python scripts
 #
 # If you change this script, it is recommended that you ensure you
 # haven't broken it by running it in various modes with a representative
 # sample of test scripts.  For example:
 #
 #  1) serial, no coverage, temp install:
 #      ./run-tests.py test-s*
 #  2) serial, no coverage, local hg:
 #      ./run-tests.py --local test-s*
 #  3) serial, coverage, temp install:
 #      ./run-tests.py -c test-s*
 #  4) serial, coverage, local hg:
 #      ./run-tests.py -c --local test-s*      # unsupported
 #  5) parallel, no coverage, temp install:
 #      ./run-tests.py -j2 test-s*
 #  6) parallel, no coverage, local hg:
 #      ./run-tests.py -j2 --local test-s*
 #  7) parallel, coverage, temp install:
 #      ./run-tests.py -j2 -c test-s*          # currently broken
 #  8) parallel, coverage, local install:
 #      ./run-tests.py -j2 -c --local test-s*  # unsupported (and broken)
 #  9) parallel, custom tmp dir:
 #      ./run-tests.py -j2 --tmpdir /tmp/myhgtests
 #  10) parallel, pure, tests that call run-tests:
 #      ./run-tests.py --pure `grep -l run-tests.py *.t`
 #
 # (You could use any subset of the tests: test-s* happens to match
 # enough that it's worth doing parallel runs, few enough that it
 # completes fairly quickly, includes both shell and Python scripts, and
 # includes some scripts that run daemon processes.)
 
 from __future__ import absolute_import, print_function
 
 import argparse
 import collections
 import difflib
 import distutils.version as version
 import errno
 import json
 import multiprocessing
 import os
 import platform
 import random
 import re
 import shutil
 import signal
 import socket
 import subprocess
 import sys
 import sysconfig
 import tempfile
 import threading
 import time
 import unittest
 import uuid
 import xml.dom.minidom as minidom
 
 try:
     import Queue as queue
 except ImportError:
     import queue
 
 try:
     import shlex
 
     shellquote = shlex.quote
 except (ImportError, AttributeError):
     import pipes
 
     shellquote = pipes.quote
 
 processlock = threading.Lock()
 
 pygmentspresent = False
 # ANSI color is unsupported prior to Windows 10
 if os.name != 'nt':
     try:  # is pygments installed
         import pygments
         import pygments.lexers as lexers
         import pygments.lexer as lexer
         import pygments.formatters as formatters
         import pygments.token as token
         import pygments.style as style
 
         pygmentspresent = True
         difflexer = lexers.DiffLexer()
         terminal256formatter = formatters.Terminal256Formatter()
     except ImportError:
         pass
 
 if pygmentspresent:
 
     class TestRunnerStyle(style.Style):
         default_style = ""
         skipped = token.string_to_tokentype("Token.Generic.Skipped")
         failed = token.string_to_tokentype("Token.Generic.Failed")
         skippedname = token.string_to_tokentype("Token.Generic.SName")
         failedname = token.string_to_tokentype("Token.Generic.FName")
         styles = {
             skipped: '#e5e5e5',
             skippedname: '#00ffff',
             failed: '#7f0000',
             failedname: '#ff0000',
         }
 
     class TestRunnerLexer(lexer.RegexLexer):
         testpattern = r'[\w-]+\.(t|py)(#[a-zA-Z0-9_\-\.]+)?'
         tokens = {
             'root': [
                 (r'^Skipped', token.Generic.Skipped, 'skipped'),
                 (r'^Failed ', token.Generic.Failed, 'failed'),
                 (r'^ERROR: ', token.Generic.Failed, 'failed'),
             ],
             'skipped': [
                 (testpattern, token.Generic.SName),
                 (r':.*', token.Generic.Skipped),
             ],
             'failed': [
                 (testpattern, token.Generic.FName),
                 (r'(:| ).*', token.Generic.Failed),
             ],
         }
 
     runnerformatter = formatters.Terminal256Formatter(style=TestRunnerStyle)
     runnerlexer = TestRunnerLexer()
 
 origenviron = os.environ.copy()
 
 if sys.version_info > (3, 5, 0):
     PYTHON3 = True
     xrange = range  # we use xrange in one place, and we'd rather not use range
 
     def _sys2bytes(p):
         if p is None:
             return p
         return p.encode('utf-8')
 
     def _bytes2sys(p):
         if p is None:
             return p
         return p.decode('utf-8')
 
     osenvironb = getattr(os, 'environb', None)
     if osenvironb is None:
         # Windows lacks os.environb, for instance.  A proxy over the real thing
         # instead of a copy allows the environment to be updated via bytes on
         # all platforms.
         class environbytes(object):
             def __init__(self, strenv):
                 self.__len__ = strenv.__len__
                 self.clear = strenv.clear
                 self._strenv = strenv
 
             def __getitem__(self, k):
                 v = self._strenv.__getitem__(_bytes2sys(k))
                 return _sys2bytes(v)
 
             def __setitem__(self, k, v):
                 self._strenv.__setitem__(_bytes2sys(k), _bytes2sys(v))
 
             def __delitem__(self, k):
                 self._strenv.__delitem__(_bytes2sys(k))
 
             def __contains__(self, k):
                 return self._strenv.__contains__(_bytes2sys(k))
 
             def __iter__(self):
                 return iter([_sys2bytes(k) for k in iter(self._strenv)])
 
             def get(self, k, default=None):
                 v = self._strenv.get(_bytes2sys(k), _bytes2sys(default))
                 return _sys2bytes(v)
 
             def pop(self, k, default=None):
                 v = self._strenv.pop(_bytes2sys(k), _bytes2sys(default))
                 return _sys2bytes(v)
 
         osenvironb = environbytes(os.environ)
 
     getcwdb = getattr(os, 'getcwdb')
     if not getcwdb or os.name == 'nt':
         getcwdb = lambda: _sys2bytes(os.getcwd())
 
 elif sys.version_info >= (3, 0, 0):
     print(
         '%s is only supported on Python 3.5+ and 2.7, not %s'
         % (sys.argv[0], '.'.join(str(v) for v in sys.version_info[:3]))
     )
     sys.exit(70)  # EX_SOFTWARE from `man 3 sysexit`
 else:
     PYTHON3 = False
 
     # In python 2.x, path operations are generally done using
     # bytestrings by default, so we don't have to do any extra
     # fiddling there. We define the wrapper functions anyway just to
     # help keep code consistent between platforms.
     def _sys2bytes(p):
         return p
 
     _bytes2sys = _sys2bytes
     osenvironb = os.environ
     getcwdb = os.getcwd
 
 # For Windows support
 wifexited = getattr(os, "WIFEXITED", lambda x: False)
 
 # Whether to use IPv6
 def checksocketfamily(name, port=20058):
     """return true if we can listen on localhost using family=name
 
     name should be either 'AF_INET', or 'AF_INET6'.
     port being used is okay - EADDRINUSE is considered as successful.
     """
     family = getattr(socket, name, None)
     if family is None:
         return False
     try:
         s = socket.socket(family, socket.SOCK_STREAM)
         s.bind(('localhost', port))
         s.close()
         return True
     except socket.error as exc:
         if exc.errno == errno.EADDRINUSE:
             return True
         elif exc.errno in (errno.EADDRNOTAVAIL, errno.EPROTONOSUPPORT):
             return False
         else:
             raise
     else:
         return False
 
 
 # useipv6 will be set by parseargs
 useipv6 = None
 
 
 def checkportisavailable(port):
     """return true if a port seems free to bind on localhost"""
     if useipv6:
         family = socket.AF_INET6
     else:
         family = socket.AF_INET
     try:
         s = socket.socket(family, socket.SOCK_STREAM)
         s.bind(('localhost', port))
         s.close()
         return True
     except socket.error as exc:
         if exc.errno not in (
             errno.EADDRINUSE,
             errno.EADDRNOTAVAIL,
             errno.EPROTONOSUPPORT,
         ):
             raise
     return False
 
 
 closefds = os.name == 'posix'
 
 
 def Popen4(cmd, wd, timeout, env=None):
     processlock.acquire()
     p = subprocess.Popen(
         _bytes2sys(cmd),
         shell=True,
         bufsize=-1,
         cwd=_bytes2sys(wd),
         env=env,
         close_fds=closefds,
         stdin=subprocess.PIPE,
         stdout=subprocess.PIPE,
         stderr=subprocess.STDOUT,
     )
     processlock.release()
 
     p.fromchild = p.stdout
     p.tochild = p.stdin
     p.childerr = p.stderr
 
     p.timeout = False
     if timeout:
 
         def t():
             start = time.time()
             while time.time() - start < timeout and p.returncode is None:
                 time.sleep(0.1)
             p.timeout = True
             if p.returncode is None:
                 terminate(p)
 
         threading.Thread(target=t).start()
 
     return p
 
 
 if sys.executable:
     sysexecutable = sys.executable
 elif os.environ.get('PYTHONEXECUTABLE'):
     sysexecutable = os.environ['PYTHONEXECUTABLE']
 elif os.environ.get('PYTHON'):
     sysexecutable = os.environ['PYTHON']
 else:
     raise AssertionError('Could not find Python interpreter')
 
 PYTHON = _sys2bytes(sysexecutable.replace('\\', '/'))
 IMPL_PATH = b'PYTHONPATH'
 if 'java' in sys.platform:
     IMPL_PATH = b'JYTHONPATH'
 
 defaults = {
     'jobs': ('HGTEST_JOBS', multiprocessing.cpu_count()),
     'timeout': ('HGTEST_TIMEOUT', 180),
     'slowtimeout': ('HGTEST_SLOWTIMEOUT', 1500),
     'port': ('HGTEST_PORT', 20059),
     'shell': ('HGTEST_SHELL', 'sh'),
 }
 
 
 def canonpath(path):
     return os.path.realpath(os.path.expanduser(path))
 
 
 def parselistfiles(files, listtype, warn=True):
     entries = dict()
     for filename in files:
         try:
             path = os.path.expanduser(os.path.expandvars(filename))
             f = open(path, "rb")
         except IOError as err:
             if err.errno != errno.ENOENT:
                 raise
             if warn:
                 print("warning: no such %s file: %s" % (listtype, filename))
             continue
 
         for line in f.readlines():
             line = line.split(b'#', 1)[0].strip()
             if line:
                 entries[line] = filename
 
         f.close()
     return entries
 
 
 def parsettestcases(path):
     """read a .t test file, return a set of test case names
 
     If path does not exist, return an empty set.
     """
     cases = []
     try:
         with open(path, 'rb') as f:
             for l in f:
                 if l.startswith(b'#testcases '):
                     cases.append(sorted(l[11:].split()))
     except IOError as ex:
         if ex.errno != errno.ENOENT:
             raise
     return cases
 
 
 def getparser():
     """Obtain the OptionParser used by the CLI."""
     parser = argparse.ArgumentParser(usage='%(prog)s [options] [tests]')
 
     selection = parser.add_argument_group('Test Selection')
     selection.add_argument(
         '--allow-slow-tests',
         action='store_true',
         help='allow extremely slow tests',
     )
     selection.add_argument(
         "--blacklist",
         action="append",
         help="skip tests listed in the specified blacklist file",
     )
     selection.add_argument(
         "--changed",
         help="run tests that are changed in parent rev or working directory",
     )
     selection.add_argument(
         "-k", "--keywords", help="run tests matching keywords"
     )
     selection.add_argument(
         "-r", "--retest", action="store_true", help="retest failed tests"
     )
     selection.add_argument(
         "--test-list",
         action="append",
         help="read tests to run from the specified file",
     )
     selection.add_argument(
         "--whitelist",
         action="append",
         help="always run tests listed in the specified whitelist file",
     )
     selection.add_argument(
         'tests', metavar='TESTS', nargs='*', help='Tests to run'
     )
 
     harness = parser.add_argument_group('Test Harness Behavior')
     harness.add_argument(
         '--bisect-repo',
         metavar='bisect_repo',
         help=(
             "Path of a repo to bisect. Use together with " "--known-good-rev"
         ),
     )
     harness.add_argument(
         "-d",
         "--debug",
         action="store_true",
         help="debug mode: write output of test scripts to console"
         " rather than capturing and diffing it (disables timeout)",
     )
     harness.add_argument(
         "-f",
         "--first",
         action="store_true",
         help="exit on the first test failure",
     )
     harness.add_argument(
         "-i",
         "--interactive",
         action="store_true",
         help="prompt to accept changed output",
     )
     harness.add_argument(
         "-j",
         "--jobs",
         type=int,
         help="number of jobs to run in parallel"
         " (default: $%s or %d)" % defaults['jobs'],
     )
     harness.add_argument(
         "--keep-tmpdir",
         action="store_true",
         help="keep temporary directory after running tests",
     )
     harness.add_argument(
         '--known-good-rev',
         metavar="known_good_rev",
         help=(
             "Automatically bisect any failures using this "
             "revision as a known-good revision."
         ),
     )
     harness.add_argument(
         "--list-tests",
         action="store_true",
         help="list tests instead of running them",
     )
     harness.add_argument(
         "--loop", action="store_true", help="loop tests repeatedly"
     )
     harness.add_argument(
         '--random', action="store_true", help='run tests in random order'
     )
     harness.add_argument(
         '--order-by-runtime',
         action="store_true",
         help='run slowest tests first, according to .testtimes',
     )
     harness.add_argument(
         "-p",
         "--port",
         type=int,
         help="port on which servers should listen"
         " (default: $%s or %d)" % defaults['port'],
     )
     harness.add_argument(
         '--profile-runner',
         action='store_true',
         help='run statprof on run-tests',
     )
     harness.add_argument(
         "-R", "--restart", action="store_true", help="restart at last error"
     )
     harness.add_argument(
         "--runs-per-test",
         type=int,
         dest="runs_per_test",
         help="run each test N times (default=1)",
         default=1,
     )
     harness.add_argument(
         "--shell", help="shell to use (default: $%s or %s)" % defaults['shell']
     )
     harness.add_argument(
         '--showchannels', action='store_true', help='show scheduling channels'
     )
     harness.add_argument(
         "--slowtimeout",
         type=int,
         help="kill errant slow tests after SLOWTIMEOUT seconds"
         " (default: $%s or %d)" % defaults['slowtimeout'],
     )
     harness.add_argument(
         "-t",
         "--timeout",
         type=int,
         help="kill errant tests after TIMEOUT seconds"
         " (default: $%s or %d)" % defaults['timeout'],
     )
     harness.add_argument(
         "--tmpdir",
         help="run tests in the given temporary directory"
         " (implies --keep-tmpdir)",
     )
     harness.add_argument(
         "-v", "--verbose", action="store_true", help="output verbose messages"
     )
 
     hgconf = parser.add_argument_group('Mercurial Configuration')
     hgconf.add_argument(
         "--chg",
         action="store_true",
         help="install and use chg wrapper in place of hg",
     )
     hgconf.add_argument("--compiler", help="compiler to build with")
     hgconf.add_argument(
         '--extra-config-opt',
         action="append",
         default=[],
         help='set the given config opt in the test hgrc',
     )
     hgconf.add_argument(
         "-l",
         "--local",
         action="store_true",
         help="shortcut for --with-hg=<testdir>/../hg, "
         "and --with-chg=<testdir>/../contrib/chg/chg if --chg is set",
     )
     hgconf.add_argument(
         "--ipv6",
         action="store_true",
         help="prefer IPv6 to IPv4 for network related tests",
     )
     hgconf.add_argument(
         "--pure",
         action="store_true",
         help="use pure Python code instead of C extensions",
     )
     hgconf.add_argument(
         "--rust",
         action="store_true",
         help="use Rust code alongside C extensions",
     )
     hgconf.add_argument(
         "--no-rust",
         action="store_true",
         help="do not use Rust code even if compiled",
     )
     hgconf.add_argument(
         "--with-chg",
         metavar="CHG",
         help="use specified chg wrapper in place of hg",
     )
     hgconf.add_argument(
         "--with-hg",
         metavar="HG",
         help="test using specified hg script rather than a "
         "temporary installation",
     )
 
     reporting = parser.add_argument_group('Results Reporting')
     reporting.add_argument(
         "-C",
         "--annotate",
         action="store_true",
         help="output files annotated with coverage",
     )
     reporting.add_argument(
         "--color",
         choices=["always", "auto", "never"],
         default=os.environ.get('HGRUNTESTSCOLOR', 'auto'),
         help="colorisation: always|auto|never (default: auto)",
     )
     reporting.add_argument(
         "-c",
         "--cover",
         action="store_true",
         help="print a test coverage report",
     )
     reporting.add_argument(
         '--exceptions',
         action='store_true',
         help='log all exceptions and generate an exception report',
     )
     reporting.add_argument(
         "-H",
         "--htmlcov",
         action="store_true",
         help="create an HTML report of the coverage of the files",
     )
     reporting.add_argument(
         "--json",
         action="store_true",
         help="store test result data in 'report.json' file",
     )
     reporting.add_argument(
         "--outputdir",
         help="directory to write error logs to (default=test directory)",
     )
     reporting.add_argument(
         "-n", "--nodiff", action="store_true", help="skip showing test changes"
     )
     reporting.add_argument(
         "-S",
         "--noskips",
         action="store_true",
         help="don't report skip tests verbosely",
     )
     reporting.add_argument(
         "--time", action="store_true", help="time how long each test takes"
     )
     reporting.add_argument("--view", help="external diff viewer")
     reporting.add_argument(
         "--xunit", help="record xunit results at specified path"
     )
 
     for option, (envvar, default) in defaults.items():
         defaults[option] = type(default)(os.environ.get(envvar, default))
     parser.set_defaults(**defaults)
 
     return parser
 
 
 def parseargs(args, parser):
     """Parse arguments with our OptionParser and validate results."""
     options = parser.parse_args(args)
 
     # jython is always pure
     if 'java' in sys.platform or '__pypy__' in sys.modules:
         options.pure = True
 
     if platform.python_implementation() != 'CPython' and options.rust:
         parser.error('Rust extensions are only available with CPython')
 
     if options.pure and options.rust:
         parser.error('--rust cannot be used with --pure')
 
     if options.rust and options.no_rust:
         parser.error('--rust cannot be used with --no-rust')
 
     if options.local:
         if options.with_hg or options.with_chg:
             parser.error('--local cannot be used with --with-hg or --with-chg')
         testdir = os.path.dirname(_sys2bytes(canonpath(sys.argv[0])))
         reporootdir = os.path.dirname(testdir)
         pathandattrs = [(b'hg', 'with_hg')]
         if options.chg:
             pathandattrs.append((b'contrib/chg/chg', 'with_chg'))
         for relpath, attr in pathandattrs:
             binpath = os.path.join(reporootdir, relpath)
             if os.name != 'nt' and not os.access(binpath, os.X_OK):
                 parser.error(
                     '--local specified, but %r not found or '
                     'not executable' % binpath
                 )
             setattr(options, attr, _bytes2sys(binpath))
 
     if options.with_hg:
         options.with_hg = canonpath(_sys2bytes(options.with_hg))
         if not (
             os.path.isfile(options.with_hg)
             and os.access(options.with_hg, os.X_OK)
         ):
             parser.error('--with-hg must specify an executable hg script')
         if os.path.basename(options.with_hg) not in [b'hg', b'hg.exe']:
             sys.stderr.write('warning: --with-hg should specify an hg script\n')
             sys.stderr.flush()
 
     if (options.chg or options.with_chg) and os.name == 'nt':
         parser.error('chg does not work on %s' % os.name)
     if options.with_chg:
         options.chg = False  # no installation to temporary location
         options.with_chg = canonpath(_sys2bytes(options.with_chg))
         if not (
             os.path.isfile(options.with_chg)
             and os.access(options.with_chg, os.X_OK)
         ):
             parser.error('--with-chg must specify a chg executable')
     if options.chg and options.with_hg:
         # chg shares installation location with hg
         parser.error(
             '--chg does not work when --with-hg is specified '
             '(use --with-chg instead)'
         )
 
     if options.color == 'always' and not pygmentspresent:
         sys.stderr.write(
             'warning: --color=always ignored because '
             'pygments is not installed\n'
         )
 
     if options.bisect_repo and not options.known_good_rev:
         parser.error("--bisect-repo cannot be used without --known-good-rev")
 
     global useipv6
     if options.ipv6:
         useipv6 = checksocketfamily('AF_INET6')
     else:
         # only use IPv6 if IPv4 is unavailable and IPv6 is available
         useipv6 = (not checksocketfamily('AF_INET')) and checksocketfamily(
             'AF_INET6'
         )
 
     options.anycoverage = options.cover or options.annotate or options.htmlcov
     if options.anycoverage:
         try:
             import coverage
 
             covver = version.StrictVersion(coverage.__version__).version
             if covver < (3, 3):
                 parser.error('coverage options require coverage 3.3 or later')
         except ImportError:
             parser.error('coverage options now require the coverage package')
 
     if options.anycoverage and options.local:
         # this needs some path mangling somewhere, I guess
         parser.error(
             "sorry, coverage options do not work when --local " "is specified"
         )
 
     if options.anycoverage and options.with_hg:
         parser.error(
             "sorry, coverage options do not work when --with-hg " "is specified"
         )
 
     global verbose
     if options.verbose:
         verbose = ''
 
     if options.tmpdir:
         options.tmpdir = canonpath(options.tmpdir)
 
     if options.jobs < 1:
         parser.error('--jobs must be positive')
     if options.interactive and options.debug:
         parser.error("-i/--interactive and -d/--debug are incompatible")
     if options.debug:
         if options.timeout != defaults['timeout']:
             sys.stderr.write('warning: --timeout option ignored with --debug\n')
         if options.slowtimeout != defaults['slowtimeout']:
             sys.stderr.write(
                 'warning: --slowtimeout option ignored with --debug\n'
             )
         options.timeout = 0
         options.slowtimeout = 0
 
     if options.blacklist:
         options.blacklist = parselistfiles(options.blacklist, 'blacklist')
     if options.whitelist:
         options.whitelisted = parselistfiles(options.whitelist, 'whitelist')
     else:
         options.whitelisted = {}
 
     if options.showchannels:
         options.nodiff = True
 
     return options
 
 
 def rename(src, dst):
     """Like os.rename(), trade atomicity and opened files friendliness
     for existing destination support.
     """
     shutil.copy(src, dst)
     os.remove(src)
 
 
 def makecleanable(path):
     """Try to fix directory permission recursively so that the entire tree
     can be deleted"""
     for dirpath, dirnames, _filenames in os.walk(path, topdown=True):
         for d in dirnames:
             p = os.path.join(dirpath, d)
             try:
                 os.chmod(p, os.stat(p).st_mode & 0o777 | 0o700)  # chmod u+rwx
             except OSError:
                 pass
 
 
 _unified_diff = difflib.unified_diff
 if PYTHON3:
     import functools
 
     _unified_diff = functools.partial(difflib.diff_bytes, difflib.unified_diff)
 
 
 def getdiff(expected, output, ref, err):
     servefail = False
     lines = []
     for line in _unified_diff(expected, output, ref, err):
         if line.startswith(b'+++') or line.startswith(b'---'):
             line = line.replace(b'\\', b'/')
             if line.endswith(b' \n'):
                 line = line[:-2] + b'\n'
         lines.append(line)
         if not servefail and line.startswith(
             b'+  abort: child process failed to start'
         ):
             servefail = True
 
     return servefail, lines
 
 
 verbose = False
 
 
 def vlog(*msg):
     """Log only when in verbose mode."""
     if verbose is False:
         return
 
     return log(*msg)
 
 
 # Bytes that break XML even in a CDATA block: control characters 0-31
 # sans \t, \n and \r
 CDATA_EVIL = re.compile(br"[\000-\010\013\014\016-\037]")
 
 # Match feature conditionalized output lines in the form, capturing the feature
 # list in group 2, and the preceeding line output in group 1:
 #
 #   output..output (feature !)\n
 optline = re.compile(br'(.*) \((.+?) !\)\n$')
 
 
 def cdatasafe(data):
     """Make a string safe to include in a CDATA block.
 
     Certain control characters are illegal in a CDATA block, and
     there's no way to include a ]]> in a CDATA either. This function
     replaces illegal bytes with ? and adds a space between the ]] so
     that it won't break the CDATA block.
     """
     return CDATA_EVIL.sub(b'?', data).replace(b']]>', b'] ]>')
 
 
 def log(*msg):
     """Log something to stdout.
 
     Arguments are strings to print.
     """
     with iolock:
         if verbose:
             print(verbose, end=' ')
         for m in msg:
             print(m, end=' ')
         print()
         sys.stdout.flush()
 
 
 def highlightdiff(line, color):
     if not color:
         return line
     assert pygmentspresent
     return pygments.highlight(
         line.decode('latin1'), difflexer, terminal256formatter
     ).encode('latin1')
 
 
 def highlightmsg(msg, color):
     if not color:
         return msg
     assert pygmentspresent
     return pygments.highlight(msg, runnerlexer, runnerformatter)
 
 
 def terminate(proc):
     """Terminate subprocess"""
     vlog('# Terminating process %d' % proc.pid)
     try:
         proc.terminate()
     except OSError:
         pass
 
 
 def killdaemons(pidfile):
     import killdaemons as killmod
 
     return killmod.killdaemons(pidfile, tryhard=False, remove=True, logfn=vlog)
 
 
 class Test(unittest.TestCase):
     """Encapsulates a single, runnable test.
 
     While this class conforms to the unittest.TestCase API, it differs in that
     instances need to be instantiated manually. (Typically, unittest.TestCase
     classes are instantiated automatically by scanning modules.)
     """
 
     # Status code reserved for skipped tests (used by hghave).
     SKIPPED_STATUS = 80
 
     def __init__(
         self,
         path,
         outputdir,
         tmpdir,
         keeptmpdir=False,
         debug=False,
         first=False,
         timeout=None,
         startport=None,
         extraconfigopts=None,
         shell=None,
         hgcommand=None,
         slowtimeout=None,
         usechg=False,
         useipv6=False,
     ):
         """Create a test from parameters.
 
         path is the full path to the file defining the test.
 
         tmpdir is the main temporary directory to use for this test.
 
         keeptmpdir determines whether to keep the test's temporary directory
         after execution. It defaults to removal (False).
 
         debug mode will make the test execute verbosely, with unfiltered
         output.
 
         timeout controls the maximum run time of the test. It is ignored when
         debug is True. See slowtimeout for tests with #require slow.
 
         slowtimeout overrides timeout if the test has #require slow.
 
         startport controls the starting port number to use for this test. Each
         test will reserve 3 port numbers for execution. It is the caller's
         responsibility to allocate a non-overlapping port range to Test
         instances.
 
         extraconfigopts is an iterable of extra hgrc config options. Values
         must have the form "key=value" (something understood by hgrc). Values
         of the form "foo.key=value" will result in "[foo] key=value".
 
         shell is the shell to execute tests in.
         """
         if timeout is None:
             timeout = defaults['timeout']
         if startport is None:
             startport = defaults['port']
         if slowtimeout is None:
             slowtimeout = defaults['slowtimeout']
         self.path = path
         self.bname = os.path.basename(path)
         self.name = _bytes2sys(self.bname)
         self._testdir = os.path.dirname(path)
         self._outputdir = outputdir
         self._tmpname = os.path.basename(path)
         self.errpath = os.path.join(self._outputdir, b'%s.err' % self.bname)
 
         self._threadtmp = tmpdir
         self._keeptmpdir = keeptmpdir
         self._debug = debug
         self._first = first
         self._timeout = timeout
         self._slowtimeout = slowtimeout
         self._startport = startport
         self._extraconfigopts = extraconfigopts or []
         self._shell = _sys2bytes(shell)
         self._hgcommand = hgcommand or b'hg'
         self._usechg = usechg
         self._useipv6 = useipv6
 
         self._aborted = False
         self._daemonpids = []
         self._finished = None
         self._ret = None
         self._out = None
         self._skipped = None
         self._testtmp = None
         self._chgsockdir = None
 
         self._refout = self.readrefout()
 
     def readrefout(self):
         """read reference output"""
         # If we're not in --debug mode and reference output file exists,
         # check test output against it.
         if self._debug:
             return None  # to match "out is None"
         elif os.path.exists(self.refpath):
             with open(self.refpath, 'rb') as f:
                 return f.read().splitlines(True)
         else:
             return []
 
     # needed to get base class __repr__ running
     @property
     def _testMethodName(self):
         return self.name
 
     def __str__(self):
         return self.name
 
     def shortDescription(self):
         return self.name
 
     def setUp(self):
         """Tasks to perform before run()."""
         self._finished = False
         self._ret = None
         self._out = None
         self._skipped = None
 
         try:
             os.mkdir(self._threadtmp)
         except OSError as e:
             if e.errno != errno.EEXIST:
                 raise
 
         name = self._tmpname
         self._testtmp = os.path.join(self._threadtmp, name)
         os.mkdir(self._testtmp)
 
         # Remove any previous output files.
         if os.path.exists(self.errpath):
             try:
                 os.remove(self.errpath)
             except OSError as e:
                 # We might have raced another test to clean up a .err
                 # file, so ignore ENOENT when removing a previous .err
                 # file.
                 if e.errno != errno.ENOENT:
                     raise
 
         if self._usechg:
             self._chgsockdir = os.path.join(
                 self._threadtmp, b'%s.chgsock' % name
             )
             os.mkdir(self._chgsockdir)
 
     def run(self, result):
         """Run this test and report results against a TestResult instance."""
         # This function is extremely similar to unittest.TestCase.run(). Once
         # we require Python 2.7 (or at least its version of unittest), this
         # function can largely go away.
         self._result = result
         result.startTest(self)
         try:
             try:
                 self.setUp()
             except (KeyboardInterrupt, SystemExit):
                 self._aborted = True
                 raise
             except Exception:
                 result.addError(self, sys.exc_info())
                 return
 
             success = False
             try:
                 self.runTest()
             except KeyboardInterrupt:
                 self._aborted = True
                 raise
             except unittest.SkipTest as e:
                 result.addSkip(self, str(e))
                 # The base class will have already counted this as a
                 # test we "ran", but we want to exclude skipped tests
                 # from those we count towards those run.
                 result.testsRun -= 1
             except self.failureException as e:
                 # This differs from unittest in that we don't capture
                 # the stack trace. This is for historical reasons and
                 # this decision could be revisited in the future,
                 # especially for PythonTest instances.
                 if result.addFailure(self, str(e)):
                     success = True
             except Exception:
                 result.addError(self, sys.exc_info())
             else:
                 success = True
 
             try:
                 self.tearDown()
             except (KeyboardInterrupt, SystemExit):
                 self._aborted = True
                 raise
             except Exception:
                 result.addError(self, sys.exc_info())
                 success = False
 
             if success:
                 result.addSuccess(self)
         finally:
             result.stopTest(self, interrupted=self._aborted)
 
     def runTest(self):
         """Run this test instance.
 
         This will return a tuple describing the result of the test.
         """
         env = self._getenv()
         self._genrestoreenv(env)
         self._daemonpids.append(env['DAEMON_PIDS'])
         self._createhgrc(env['HGRCPATH'])
 
         vlog('# Test', self.name)
 
         ret, out = self._run(env)
         self._finished = True
         self._ret = ret
         self._out = out
 
         def describe(ret):
             if ret < 0:
                 return 'killed by signal: %d' % -ret
             return 'returned error code %d' % ret
 
         self._skipped = False
 
         if ret == self.SKIPPED_STATUS:
             if out is None:  # Debug mode, nothing to parse.
                 missing = ['unknown']
                 failed = None
             else:
                 missing, failed = TTest.parsehghaveoutput(out)
 
             if not missing:
                 missing = ['skipped']
 
             if failed:
                 self.fail('hg have failed checking for %s' % failed[-1])
             else:
                 self._skipped = True
                 raise unittest.SkipTest(missing[-1])
         elif ret == 'timeout':
             self.fail('timed out')
         elif ret is False:
             self.fail('no result code from test')
         elif out != self._refout:
             # Diff generation may rely on written .err file.
             if (
                 (ret != 0 or out != self._refout)
                 and not self._skipped
                 and not self._debug
             ):
                 with open(self.errpath, 'wb') as f:
                     for line in out:
                         f.write(line)
 
             # The result object handles diff calculation for us.
             with firstlock:
                 if self._result.addOutputMismatch(self, ret, out, self._refout):
                     # change was accepted, skip failing
                     return
                 if self._first:
                     global firsterror
                     firsterror = True
 
             if ret:
                 msg = 'output changed and ' + describe(ret)
             else:
                 msg = 'output changed'
 
             self.fail(msg)
         elif ret:
             self.fail(describe(ret))
 
     def tearDown(self):
         """Tasks to perform after run()."""
         for entry in self._daemonpids:
             killdaemons(entry)
         self._daemonpids = []
 
         if self._keeptmpdir:
             log(
                 '\nKeeping testtmp dir: %s\nKeeping threadtmp dir: %s'
                 % (_bytes2sys(self._testtmp), _bytes2sys(self._threadtmp),)
             )
         else:
             try:
                 shutil.rmtree(self._testtmp)
             except OSError:
                 # unreadable directory may be left in $TESTTMP; fix permission
                 # and try again
                 makecleanable(self._testtmp)
                 shutil.rmtree(self._testtmp, True)
             shutil.rmtree(self._threadtmp, True)
 
         if self._usechg:
             # chgservers will stop automatically after they find the socket
             # files are deleted
             shutil.rmtree(self._chgsockdir, True)
 
         if (
             (self._ret != 0 or self._out != self._refout)
             and not self._skipped
             and not self._debug
             and self._out
         ):
             with open(self.errpath, 'wb') as f:
                 for line in self._out:
                     f.write(line)
 
         vlog("# Ret was:", self._ret, '(%s)' % self.name)
 
     def _run(self, env):
         # This should be implemented in child classes to run tests.
         raise unittest.SkipTest('unknown test type')
 
     def abort(self):
         """Terminate execution of this test."""
         self._aborted = True
 
     def _portmap(self, i):
         offset = b'' if i == 0 else b'%d' % i
         return (br':%d\b' % (self._startport + i), b':$HGPORT%s' % offset)
 
     def _getreplacements(self):
         """Obtain a mapping of text replacements to apply to test output.
 
         Test output needs to be normalized so it can be compared to expected
         output. This function defines how some of that normalization will
         occur.
         """
         r = [
             # This list should be parallel to defineport in _getenv
             self._portmap(0),
             self._portmap(1),
             self._portmap(2),
             (br'([^0-9])%s' % re.escape(self._localip()), br'\1$LOCALIP'),
             (br'\bHG_TXNID=TXN:[a-f0-9]{40}\b', br'HG_TXNID=TXN:$ID$'),
         ]
         r.append((self._escapepath(self._testtmp), b'$TESTTMP'))
 
         replacementfile = os.path.join(self._testdir, b'common-pattern.py')
 
         if os.path.exists(replacementfile):
             data = {}
             with open(replacementfile, mode='rb') as source:
                 # the intermediate 'compile' step help with debugging
                 code = compile(source.read(), replacementfile, 'exec')
                 exec(code, data)
                 for value in data.get('substitutions', ()):
                     if len(value) != 2:
                         msg = 'malformatted substitution in %s: %r'
                         msg %= (replacementfile, value)
                         raise ValueError(msg)
                     r.append(value)
         return r
 
     def _escapepath(self, p):
         if os.name == 'nt':
             return b''.join(
                 c.isalpha()
                 and b'[%s%s]' % (c.lower(), c.upper())
                 or c in b'/\\'
                 and br'[/\\]'
                 or c.isdigit()
                 and c
                 or b'\\' + c
                 for c in [p[i : i + 1] for i in range(len(p))]
             )
         else:
             return re.escape(p)
 
     def _localip(self):
         if self._useipv6:
             return b'::1'
         else:
             return b'127.0.0.1'
 
     def _genrestoreenv(self, testenv):
         """Generate a script that can be used by tests to restore the original
         environment."""
         # Put the restoreenv script inside self._threadtmp
         scriptpath = os.path.join(self._threadtmp, b'restoreenv.sh')
         testenv['HGTEST_RESTOREENV'] = _bytes2sys(scriptpath)
 
         # Only restore environment variable names that the shell allows
         # us to export.
         name_regex = re.compile('^[a-zA-Z][a-zA-Z0-9_]*$')
 
         # Do not restore these variables; otherwise tests would fail.
         reqnames = {'PYTHON', 'TESTDIR', 'TESTTMP'}
 
         with open(scriptpath, 'w') as envf:
             for name, value in origenviron.items():
                 if not name_regex.match(name):
                     # Skip environment variables with unusual names not
                     # allowed by most shells.
                     continue
                 if name in reqnames:
                     continue
                 envf.write('%s=%s\n' % (name, shellquote(value)))
 
             for name in testenv:
                 if name in origenviron or name in reqnames:
                     continue
                 envf.write('unset %s\n' % (name,))
 
     def _getenv(self):
         """Obtain environment variables to use during test execution."""
 
         def defineport(i):
             offset = '' if i == 0 else '%s' % i
             env["HGPORT%s" % offset] = '%s' % (self._startport + i)
 
         env = os.environ.copy()
         env['PYTHONUSERBASE'] = sysconfig.get_config_var('userbase') or ''
         env['HGEMITWARNINGS'] = '1'
         env['TESTTMP'] = _bytes2sys(self._testtmp)
         env['TESTNAME'] = self.name
         env['HOME'] = _bytes2sys(self._testtmp)
         # This number should match portneeded in _getport
         for port in xrange(3):
             # This list should be parallel to _portmap in _getreplacements
             defineport(port)
         env["HGRCPATH"] = _bytes2sys(os.path.join(self._threadtmp, b'.hgrc'))
         env["DAEMON_PIDS"] = _bytes2sys(
             os.path.join(self._threadtmp, b'daemon.pids')
         )
         env["HGEDITOR"] = (
             '"' + sysexecutable + '"' + ' -c "import sys; sys.exit(0)"'
         )
         env["HGUSER"] = "test"
         env["HGENCODING"] = "ascii"
         env["HGENCODINGMODE"] = "strict"
         env["HGHOSTNAME"] = "test-hostname"
         env['HGIPV6'] = str(int(self._useipv6))
         # See contrib/catapipe.py for how to use this functionality.
         if 'HGTESTCATAPULTSERVERPIPE' not in env:
             # If we don't have HGTESTCATAPULTSERVERPIPE explicitly set, pull the
             # non-test one in as a default, otherwise set to devnull
             env['HGTESTCATAPULTSERVERPIPE'] = env.get(
                 'HGCATAPULTSERVERPIPE', os.devnull
             )
 
         extraextensions = []
         for opt in self._extraconfigopts:
             section, key = _sys2bytes(opt).split(b'.', 1)
             if section != 'extensions':
                 continue
             name = key.split(b'=', 1)[0]
             extraextensions.append(name)
 
         if extraextensions:
             env['HGTESTEXTRAEXTENSIONS'] = b' '.join(extraextensions)
 
         # LOCALIP could be ::1 or 127.0.0.1. Useful for tests that require raw
         # IP addresses.
         env['LOCALIP'] = _bytes2sys(self._localip())
 
         # This has the same effect as Py_LegacyWindowsStdioFlag in exewrapper.c,
         # but this is needed for testing python instances like dummyssh,
         # dummysmtpd.py, and dumbhttp.py.
         if PYTHON3 and os.name == 'nt':
             env['PYTHONLEGACYWINDOWSSTDIO'] = '1'
 
         # Modified HOME in test environment can confuse Rust tools. So set
         # CARGO_HOME and RUSTUP_HOME automatically if a Rust toolchain is
         # present and these variables aren't already defined.
         cargo_home_path = os.path.expanduser('~/.cargo')
         rustup_home_path = os.path.expanduser('~/.rustup')
 
         if os.path.exists(cargo_home_path) and b'CARGO_HOME' not in osenvironb:
             env['CARGO_HOME'] = cargo_home_path
         if (
             os.path.exists(rustup_home_path)
             and b'RUSTUP_HOME' not in osenvironb
         ):
             env['RUSTUP_HOME'] = rustup_home_path
 
         # Reset some environment variables to well-known values so that
         # the tests produce repeatable output.
         env['LANG'] = env['LC_ALL'] = env['LANGUAGE'] = 'C'
         env['TZ'] = 'GMT'
         env["EMAIL"] = "Foo Bar <[e-mail]>"
         env['COLUMNS'] = '80'
         env['TERM'] = 'xterm'
 
         dropped = [
             'CDPATH',
             'CHGDEBUG',
             'EDITOR',
             'GREP_OPTIONS',
             'HG',
             'HGMERGE',
             'HGPLAIN',
             'HGPLAINEXCEPT',
             'HGPROF',
             'http_proxy',
             'no_proxy',
             'NO_PROXY',
             'PAGER',
             'VISUAL',
         ]
 
         for k in dropped:
             if k in env:
                 del env[k]
 
         # unset env related to hooks
         for k in list(env):
             if k.startswith('HG_'):
                 del env[k]
 
         if self._usechg:
             env['CHGSOCKNAME'] = os.path.join(self._chgsockdir, b'server')
 
         return env
 
     def _createhgrc(self, path):
         """Create an hgrc file for this test."""
         with open(path, 'wb') as hgrc:
             hgrc.write(b'[ui]\n')
             hgrc.write(b'slash = True\n')
             hgrc.write(b'interactive = False\n')
             hgrc.write(b'merge = internal:merge\n')
             hgrc.write(b'mergemarkers = detailed\n')
             hgrc.write(b'promptecho = True\n')
             hgrc.write(b'[defaults]\n')
             hgrc.write(b'[devel]\n')
             hgrc.write(b'all-warnings = true\n')
             hgrc.write(b'default-date = 0 0\n')
             hgrc.write(b'[largefiles]\n')
             hgrc.write(
                 b'usercache = %s\n'
                 % (os.path.join(self._testtmp, b'.cache/largefiles'))
             )
             hgrc.write(b'[lfs]\n')
             hgrc.write(
                 b'usercache = %s\n'
                 % (os.path.join(self._testtmp, b'.cache/lfs'))
             )
             hgrc.write(b'[web]\n')
             hgrc.write(b'address = localhost\n')
             hgrc.write(b'ipv6 = %r\n' % self._useipv6)
             hgrc.write(b'server-header = testing stub value\n')
 
             for opt in self._extraconfigopts:
                 section, key = _sys2bytes(opt).split(b'.', 1)
                 assert b'=' in key, (
                     'extra config opt %s must ' 'have an = for assignment' % opt
                 )
                 hgrc.write(b'[%s]\n%s\n' % (section, key))
 
     def fail(self, msg):
         # unittest differentiates between errored and failed.
         # Failed is denoted by AssertionError (by default at least).
         raise AssertionError(msg)
 
     def _runcommand(self, cmd, env, normalizenewlines=False):
         """Run command in a sub-process, capturing the output (stdout and
         stderr).
 
         Return a tuple (exitcode, output). output is None in debug mode.
         """
         if self._debug:
             proc = subprocess.Popen(
                 _bytes2sys(cmd),
                 shell=True,
                 cwd=_bytes2sys(self._testtmp),
                 env=env,
             )
             ret = proc.wait()
             return (ret, None)
 
         proc = Popen4(cmd, self._testtmp, self._timeout, env)
 
         def cleanup():
             terminate(proc)
             ret = proc.wait()
             if ret == 0:
                 ret = signal.SIGTERM << 8
             killdaemons(env['DAEMON_PIDS'])
             return ret
 
         proc.tochild.close()
 
         try:
             output = proc.fromchild.read()
         except KeyboardInterrupt:
             vlog('# Handling keyboard interrupt')
             cleanup()
             raise
 
         ret = proc.wait()
         if wifexited(ret):
             ret = os.WEXITSTATUS(ret)
 
         if proc.timeout:
             ret = 'timeout'
 
         if ret:
             killdaemons(env['DAEMON_PIDS'])
 
         for s, r in self._getreplacements():
             output = re.sub(s, r, output)
 
         if normalizenewlines:
             output = output.replace(b'\r\n', b'\n')
 
         return ret, output.splitlines(True)
 
 
 class PythonTest(Test):
     """A Python-based test."""
 
     @property
     def refpath(self):
         return os.path.join(self._testdir, b'%s.out' % self.bname)
 
     def _run(self, env):
         # Quote the python(3) executable for Windows
         cmd = b'"%s" "%s"' % (PYTHON, self.path)
         vlog("# Running", cmd.decode("utf-8"))
         normalizenewlines = os.name == 'nt'
         result = self._runcommand(cmd, env, normalizenewlines=normalizenewlines)
         if self._aborted:
             raise KeyboardInterrupt()
 
         return result
 
 
 # Some glob patterns apply only in some circumstances, so the script
 # might want to remove (glob) annotations that otherwise should be
 # retained.
 checkcodeglobpats = [
     # On Windows it looks like \ doesn't require a (glob), but we know
     # better.
     re.compile(br'^pushing to \$TESTTMP/.*[^)]$'),
     re.compile(br'^moving \S+/.*[^)]$'),
     re.compile(br'^pulling from \$TESTTMP/.*[^)]$'),
     # Not all platforms have 127.0.0.1 as loopback (though most do),
     # so we always glob that too.
     re.compile(br'.*\$LOCALIP.*$'),
 ]
 
 bchr = chr
 if PYTHON3:
     bchr = lambda x: bytes([x])
 
 WARN_UNDEFINED = 1
 WARN_YES = 2
 WARN_NO = 3
 
 MARK_OPTIONAL = b" (?)\n"
 
 
 def isoptional(line):
     return line.endswith(MARK_OPTIONAL)
 
 
 class TTest(Test):
     """A "t test" is a test backed by a .t file."""
 
     SKIPPED_PREFIX = b'skipped: '
     FAILED_PREFIX = b'hghave check failed: '
     NEEDESCAPE = re.compile(br'[\x00-\x08\x0b-\x1f\x7f-\xff]').search
 
     ESCAPESUB = re.compile(br'[\x00-\x08\x0b-\x1f\\\x7f-\xff]').sub
     ESCAPEMAP = {bchr(i): br'\x%02x' % i for i in range(256)}
     ESCAPEMAP.update({b'\\': b'\\\\', b'\r': br'\r'})
 
     def __init__(self, path, *args, **kwds):
         # accept an extra "case" parameter
         case = kwds.pop('case', [])
         self._case = case
         self._allcases = {x for y in parsettestcases(path) for x in y}
         super(TTest, self).__init__(path, *args, **kwds)
         if case:
             casepath = b'#'.join(case)
             self.name = '%s#%s' % (self.name, _bytes2sys(casepath))
             self.errpath = b'%s#%s.err' % (self.errpath[:-4], casepath)
             self._tmpname += b'-%s' % casepath
         self._have = {}
 
     @property
     def refpath(self):
         return os.path.join(self._testdir, self.bname)
 
     def _run(self, env):
         with open(self.path, 'rb') as f:
             lines = f.readlines()
 
         # .t file is both reference output and the test input, keep reference
         # output updated with the the test input. This avoids some race
         # conditions where the reference output does not match the actual test.
         if self._refout is not None:
             self._refout = lines
 
         salt, script, after, expected = self._parsetest(lines)
 
         # Write out the generated script.
         fname = b'%s.sh' % self._testtmp
         with open(fname, 'wb') as f:
             for l in script:
                 f.write(l)
 
         cmd = b'%s "%s"' % (self._shell, fname)
         vlog("# Running", cmd.decode("utf-8"))
 
         exitcode, output = self._runcommand(cmd, env)
 
         if self._aborted:
             raise KeyboardInterrupt()
 
         # Do not merge output if skipped. Return hghave message instead.
         # Similarly, with --debug, output is None.
         if exitcode == self.SKIPPED_STATUS or output is None:
             return exitcode, output
 
         return self._processoutput(exitcode, output, salt, after, expected)
 
     def _hghave(self, reqs):
         allreqs = b' '.join(reqs)
 
         self._detectslow(reqs)
 
         if allreqs in self._have:
             return self._have.get(allreqs)
 
         # TODO do something smarter when all other uses of hghave are gone.
-        runtestdir = os.path.abspath(os.path.dirname(_sys2bytes(__file__)))
+        runtestdir = osenvironb[b'RUNTESTDIR']
         tdir = runtestdir.replace(b'\\', b'/')
         proc = Popen4(
             b'%s -c "%s/hghave %s"' % (self._shell, tdir, allreqs),
             self._testtmp,
             0,
             self._getenv(),
         )
         stdout, stderr = proc.communicate()
         ret = proc.wait()
         if wifexited(ret):
             ret = os.WEXITSTATUS(ret)
         if ret == 2:
             print(stdout.decode('utf-8'))
             sys.exit(1)
 
         if ret != 0:
             self._have[allreqs] = (False, stdout)
             return False, stdout
 
         self._have[allreqs] = (True, None)
         return True, None
 
     def _detectslow(self, reqs):
         """update the timeout of slow test when appropriate"""
         if b'slow' in reqs:
             self._timeout = self._slowtimeout
 
     def _iftest(self, args):
         # implements "#if"
         reqs = []
         for arg in args:
             if arg.startswith(b'no-') and arg[3:] in self._allcases:
                 if arg[3:] in self._case:
                     return False
             elif arg in self._allcases:
                 if arg not in self._case:
                     return False
             else:
                 reqs.append(arg)
         self._detectslow(reqs)
         return self._hghave(reqs)[0]
 
     def _parsetest(self, lines):
         # We generate a shell script which outputs unique markers to line
         # up script results with our source. These markers include input
         # line number and the last return code.
         salt = b"SALT%d" % time.time()
 
         def addsalt(line, inpython):
             if inpython:
                 script.append(b'%s %d 0\n' % (salt, line))
             else:
                 script.append(b'echo %s %d $?\n' % (salt, line))
 
         activetrace = []
         session = str(uuid.uuid4())
         if PYTHON3:
             session = session.encode('ascii')
         hgcatapult = os.getenv('HGTESTCATAPULTSERVERPIPE') or os.getenv(
             'HGCATAPULTSERVERPIPE'
         )
 
         def toggletrace(cmd=None):
             if not hgcatapult or hgcatapult == os.devnull:
                 return
 
             if activetrace:
                 script.append(
                     b'echo END %s %s >> "$HGTESTCATAPULTSERVERPIPE"\n'
                     % (session, activetrace[0])
                 )
             if cmd is None:
                 return
 
             if isinstance(cmd, str):
                 quoted = shellquote(cmd.strip())
             else:
                 quoted = shellquote(cmd.strip().decode('utf8')).encode('utf8')
             quoted = quoted.replace(b'\\', b'\\\\')
             script.append(
                 b'echo START %s %s >> "$HGTESTCATAPULTSERVERPIPE"\n'
                 % (session, quoted)
             )
             activetrace[0:] = [quoted]
 
         script = []
 
         # After we run the shell script, we re-unify the script output
         # with non-active parts of the source, with synchronization by our
         # SALT line number markers. The after table contains the non-active
         # components, ordered by line number.
         after = {}
 
         # Expected shell script output.
         expected = {}
 
         pos = prepos = -1
 
         # True or False when in a true or false conditional section
         skipping = None
 
         # We keep track of whether or not we're in a Python block so we
         # can generate the surrounding doctest magic.
         inpython = False
 
         if self._debug:
             script.append(b'set -x\n')
         if self._hgcommand != b'hg':
             script.append(b'alias hg="%s"\n' % self._hgcommand)
         if os.getenv('MSYSTEM'):
             script.append(b'alias pwd="pwd -W"\n')
 
         if hgcatapult and hgcatapult != os.devnull:
             if PYTHON3:
                 hgcatapult = hgcatapult.encode('utf8')
                 cataname = self.name.encode('utf8')
             else:
                 cataname = self.name
 
             # Kludge: use a while loop to keep the pipe from getting
             # closed by our echo commands. The still-running file gets
             # reaped at the end of the script, which causes the while
             # loop to exit and closes the pipe. Sigh.
             script.append(
                 b'rtendtracing() {\n'
                 b'  echo END %(session)s %(name)s >> %(catapult)s\n'
                 b'  rm -f "$TESTTMP/.still-running"\n'
                 b'}\n'
                 b'trap "rtendtracing" 0\n'
                 b'touch "$TESTTMP/.still-running"\n'
                 b'while [ -f "$TESTTMP/.still-running" ]; do sleep 1; done '
                 b'> %(catapult)s &\n'
                 b'HGCATAPULTSESSION=%(session)s ; export HGCATAPULTSESSION\n'
                 b'echo START %(session)s %(name)s >> %(catapult)s\n'
                 % {
                     b'name': cataname,
                     b'session': session,
                     b'catapult': hgcatapult,
                 }
             )
 
         if self._case:
             casestr = b'#'.join(self._case)
             if isinstance(casestr, str):
                 quoted = shellquote(casestr)
             else:
                 quoted = shellquote(casestr.decode('utf8')).encode('utf8')
             script.append(b'TESTCASE=%s\n' % quoted)
             script.append(b'export TESTCASE\n')
 
         n = 0
         for n, l in enumerate(lines):
             if not l.endswith(b'\n'):
                 l += b'\n'
             if l.startswith(b'#require'):
                 lsplit = l.split()
                 if len(lsplit) < 2 or lsplit[0] != b'#require':
                     after.setdefault(pos, []).append(
                         b'  !!! invalid #require\n'
                     )
                 if not skipping:
                     haveresult, message = self._hghave(lsplit[1:])
                     if not haveresult:
                         script = [b'echo "%s"\nexit 80\n' % message]
                         break
                 after.setdefault(pos, []).append(l)
             elif l.startswith(b'#if'):
                 lsplit = l.split()
                 if len(lsplit) < 2 or lsplit[0] != b'#if':
                     after.setdefault(pos, []).append(b'  !!! invalid #if\n')
                 if skipping is not None:
                     after.setdefault(pos, []).append(b'  !!! nested #if\n')
                 skipping = not self._iftest(lsplit[1:])
                 after.setdefault(pos, []).append(l)
             elif l.startswith(b'#else'):
                 if skipping is None:
                     after.setdefault(pos, []).append(b'  !!! missing #if\n')
                 skipping = not skipping
                 after.setdefault(pos, []).append(l)
             elif l.startswith(b'#endif'):
                 if skipping is None:
                     after.setdefault(pos, []).append(b'  !!! missing #if\n')
                 skipping = None
                 after.setdefault(pos, []).append(l)
             elif skipping:
                 after.setdefault(pos, []).append(l)
             elif l.startswith(b'  >>> '):  # python inlines
                 after.setdefault(pos, []).append(l)
                 prepos = pos
                 pos = n
                 if not inpython:
                     # We've just entered a Python block. Add the header.
                     inpython = True
                     addsalt(prepos, False)  # Make sure we report the exit code.
                     script.append(b'"%s" -m heredoctest <<EOF\n' % PYTHON)
                 addsalt(n, True)
                 script.append(l[2:])
             elif l.startswith(b'  ... '):  # python inlines
                 after.setdefault(prepos, []).append(l)
                 script.append(l[2:])
             elif l.startswith(b'  $ '):  # commands
                 if inpython:
                     script.append(b'EOF\n')
                     inpython = False
                 after.setdefault(pos, []).append(l)
                 prepos = pos
                 pos = n
                 addsalt(n, False)
                 rawcmd = l[4:]
                 cmd = rawcmd.split()
                 toggletrace(rawcmd)
                 if len(cmd) == 2 and cmd[0] == b'cd':
                     rawcmd = b'cd %s || exit 1\n' % cmd[1]
                 script.append(rawcmd)
             elif l.startswith(b'  > '):  # continuations
                 after.setdefault(prepos, []).append(l)
                 script.append(l[4:])
             elif l.startswith(b'  '):  # results
                 # Queue up a list of expected results.
                 expected.setdefault(pos, []).append(l[2:])
             else:
                 if inpython:
                     script.append(b'EOF\n')
                     inpython = False
                 # Non-command/result. Queue up for merged output.
                 after.setdefault(pos, []).append(l)
 
         if inpython:
             script.append(b'EOF\n')
         if skipping is not None:
             after.setdefault(pos, []).append(b'  !!! missing #endif\n')
         addsalt(n + 1, False)
         # Need to end any current per-command trace
         if activetrace:
             toggletrace()
         return salt, script, after, expected
 
     def _processoutput(self, exitcode, output, salt, after, expected):
         # Merge the script output back into a unified test.
         warnonly = WARN_UNDEFINED  # 1: not yet; 2: yes; 3: for sure not
         if exitcode != 0:
             warnonly = WARN_NO
 
         pos = -1
         postout = []
         for out_rawline in output:
             out_line, cmd_line = out_rawline, None
             if salt in out_rawline:
                 out_line, cmd_line = out_rawline.split(salt, 1)
 
             pos, postout, warnonly = self._process_out_line(
                 out_line, pos, postout, expected, warnonly
             )
             pos, postout = self._process_cmd_line(cmd_line, pos, postout, after)
 
         if pos in after:
             postout += after.pop(pos)
 
         if warnonly == WARN_YES:
             exitcode = False  # Set exitcode to warned.
 
         return exitcode, postout
 
     def _process_out_line(self, out_line, pos, postout, expected, warnonly):
         while out_line:
             if not out_line.endswith(b'\n'):
                 out_line += b' (no-eol)\n'
 
             # Find the expected output at the current position.
             els = [None]
             if expected.get(pos, None):
                 els = expected[pos]
 
             optional = []
             for i, el in enumerate(els):
                 r = False
                 if el:
                     r, exact = self.linematch(el, out_line)
                 if isinstance(r, str):
                     if r == '-glob':
                         out_line = ''.join(el.rsplit(' (glob)', 1))
                         r = ''  # Warn only this line.
                     elif r == "retry":
                         postout.append(b'  ' + el)
                     else:
                         log('\ninfo, unknown linematch result: %r\n' % r)
                         r = False
                 if r:
                     els.pop(i)
                     break
                 if el:
                     if isoptional(el):
                         optional.append(i)
                     else:
                         m = optline.match(el)
                         if m:
                             conditions = [c for c in m.group(2).split(b' ')]
 
                             if not self._iftest(conditions):
                                 optional.append(i)
                     if exact:
                         # Don't allow line to be matches against a later
                         # line in the output
                         els.pop(i)
                         break
 
             if r:
                 if r == "retry":
                     continue
                 # clean up any optional leftovers
                 for i in optional:
                     postout.append(b'  ' + els[i])
                 for i in reversed(optional):
                     del els[i]
                 postout.append(b'  ' + el)
             else:
                 if self.NEEDESCAPE(out_line):
                     out_line = TTest._stringescape(
                         b'%s (esc)\n' % out_line.rstrip(b'\n')
                     )
                 postout.append(b'  ' + out_line)  # Let diff deal with it.
                 if r != '':  # If line failed.
                     warnonly = WARN_NO
                 elif warnonly == WARN_UNDEFINED:
                     warnonly = WARN_YES
             break
         else:
             # clean up any optional leftovers
             while expected.get(pos, None):
                 el = expected[pos].pop(0)
                 if el:
                     if not isoptional(el):
                         m = optline.match(el)
                         if m:
                             conditions = [c for c in m.group(2).split(b' ')]
 
                             if self._iftest(conditions):
                                 # Don't append as optional line
                                 continue
                         else:
                             continue
                 postout.append(b'  ' + el)
         return pos, postout, warnonly
 
     def _process_cmd_line(self, cmd_line, pos, postout, after):
         """process a "command" part of a line from unified test output"""
         if cmd_line:
             # Add on last return code.
             ret = int(cmd_line.split()[1])
             if ret != 0:
                 postout.append(b'  [%d]\n' % ret)
             if pos in after:
                 # Merge in non-active test bits.
                 postout += after.pop(pos)
             pos = int(cmd_line.split()[0])
         return pos, postout
 
     @staticmethod
     def rematch(el, l):
         try:
             # parse any flags at the beginning of the regex. Only 'i' is
             # supported right now, but this should be easy to extend.
             flags, el = re.match(br'^(\(\?i\))?(.*)', el).groups()[0:2]
             flags = flags or b''
             el = flags + b'(?:' + el + b')'
             # use \Z to ensure that the regex matches to the end of the string
             if os.name == 'nt':
                 return re.match(el + br'\r?\n\Z', l)
             return re.match(el + br'\n\Z', l)
         except re.error:
             # el is an invalid regex
             return False
 
     @staticmethod
     def globmatch(el, l):
         # The only supported special characters are * and ? plus / which also
         # matches \ on windows. Escaping of these characters is supported.
         if el + b'\n' == l:
             if os.altsep:
                 # matching on "/" is not needed for this line
                 for pat in checkcodeglobpats:
                     if pat.match(el):
                         return True
                 return b'-glob'
             return True
         el = el.replace(b'$LOCALIP', b'*')
         i, n = 0, len(el)
         res = b''
         while i < n:
             c = el[i : i + 1]
             i += 1
             if c == b'\\' and i < n and el[i : i + 1] in b'*?\\/':
                 res += el[i - 1 : i + 1]
                 i += 1
             elif c == b'*':
                 res += b'.*'
             elif c == b'?':
                 res += b'.'
             elif c == b'/' and os.altsep:
                 res += b'[/\\\\]'
             else:
                 res += re.escape(c)
         return TTest.rematch(res, l)
 
     def linematch(self, el, l):
         if el == l:  # perfect match (fast)
             return True, True
         retry = False
         if isoptional(el):
             retry = "retry"
             el = el[: -len(MARK_OPTIONAL)] + b"\n"
         else:
             m = optline.match(el)
             if m:
                 conditions = [c for c in m.group(2).split(b' ')]
 
                 el = m.group(1) + b"\n"
                 if not self._iftest(conditions):
                     # listed feature missing, should not match
                     return "retry", False
 
         if el.endswith(b" (esc)\n"):
             if PYTHON3:
                 el = el[:-7].decode('unicode_escape') + '\n'
                 el = el.encode('utf-8')
             else:
                 el = el[:-7].decode('string-escape') + '\n'
         if el == l or os.name == 'nt' and el[:-1] + b'\r\n' == l:
             return True, True
         if el.endswith(b" (re)\n"):
             return (TTest.rematch(el[:-6], l) or retry), False
         if el.endswith(b" (glob)\n"):
             # ignore '(glob)' added to l by 'replacements'
             if l.endswith(b" (glob)\n"):
                 l = l[:-8] + b"\n"
             return (TTest.globmatch(el[:-8], l) or retry), False
         if os.altsep:
             _l = l.replace(b'\\', b'/')
             if el == _l or os.name == 'nt' and el[:-1] + b'\r\n' == _l:
                 return True, True
         return retry, True
 
     @staticmethod
     def parsehghaveoutput(lines):
         '''Parse hghave log lines.
 
         Return tuple of lists (missing, failed):
           * the missing/unknown features
           * the features for which existence check failed'''
         missing = []
         failed = []
         for line in lines:
             if line.startswith(TTest.SKIPPED_PREFIX):
                 line = line.splitlines()[0]
                 missing.append(_bytes2sys(line[len(TTest.SKIPPED_PREFIX) :]))
             elif line.startswith(TTest.FAILED_PREFIX):
                 line = line.splitlines()[0]
                 failed.append(_bytes2sys(line[len(TTest.FAILED_PREFIX) :]))
 
         return missing, failed
 
     @staticmethod
     def _escapef(m):
         return TTest.ESCAPEMAP[m.group(0)]
 
     @staticmethod
     def _stringescape(s):
         return TTest.ESCAPESUB(TTest._escapef, s)
 
 
 iolock = threading.RLock()
 firstlock = threading.RLock()
 firsterror = False
 
 
 class TestResult(unittest._TextTestResult):
     """Holds results when executing via unittest."""
 
     # Don't worry too much about accessing the non-public _TextTestResult.
     # It is relatively common in Python testing tools.
     def __init__(self, options, *args, **kwargs):
         super(TestResult, self).__init__(*args, **kwargs)
 
         self._options = options
 
         # unittest.TestResult didn't have skipped until 2.7. We need to
         # polyfill it.
         self.skipped = []
 
         # We have a custom "ignored" result that isn't present in any Python
         # unittest implementation. It is very similar to skipped. It may make
         # sense to map it into skip some day.
         self.ignored = []
 
         self.times = []
         self._firststarttime = None
         # Data stored for the benefit of generating xunit reports.
         self.successes = []
         self.faildata = {}
 
         if options.color == 'auto':
             self.color = pygmentspresent and self.stream.isatty()
         elif options.color == 'never':
             self.color = False
         else:  # 'always', for testing purposes
             self.color = pygmentspresent
 
     def onStart(self, test):
         """ Can be overriden by custom TestResult
         """
 
     def onEnd(self):
         """ Can be overriden by custom TestResult
         """
 
     def addFailure(self, test, reason):
         self.failures.append((test, reason))
 
         if self._options.first:
             self.stop()
         else:
             with iolock:
                 if reason == "timed out":
                     self.stream.write('t')
                 else:
                     if not self._options.nodiff:
                         self.stream.write('\n')
                         # Exclude the '\n' from highlighting to lex correctly
                         formatted = 'ERROR: %s output changed\n' % test
                         self.stream.write(highlightmsg(formatted, self.color))
                     self.stream.write('!')
 
                 self.stream.flush()
 
     def addSuccess(self, test):
         with iolock:
             super(TestResult, self).addSuccess(test)
         self.successes.append(test)
 
     def addError(self, test, err):
         super(TestResult, self).addError(test, err)
         if self._options.first:
             self.stop()
 
     # Polyfill.
     def addSkip(self, test, reason):
         self.skipped.append((test, reason))
         with iolock:
             if self.showAll:
                 self.stream.writeln('skipped %s' % reason)
             else:
                 self.stream.write('s')
                 self.stream.flush()
 
     def addIgnore(self, test, reason):
         self.ignored.append((test, reason))
         with iolock:
             if self.showAll:
                 self.stream.writeln('ignored %s' % reason)
             else:
                 if reason not in ('not retesting', "doesn't match keyword"):
                     self.stream.write('i')
                 else:
                     self.testsRun += 1
                 self.stream.flush()
 
     def addOutputMismatch(self, test, ret, got, expected):
         """Record a mismatch in test output for a particular test."""
         if self.shouldStop or firsterror:
             # don't print, some other test case already failed and
             # printed, we're just stale and probably failed due to our
             # temp dir getting cleaned up.
             return
 
         accepted = False
         lines = []
 
         with iolock:
             if self._options.nodiff:
                 pass
             elif self._options.view:
                 v = self._options.view
                 subprocess.call(
                     r'"%s" "%s" "%s"'
                     % (v, _bytes2sys(test.refpath), _bytes2sys(test.errpath)),
                     shell=True,
                 )
             else:
                 servefail, lines = getdiff(
                     expected, got, test.refpath, test.errpath
                 )
                 self.stream.write('\n')
                 for line in lines:
                     line = highlightdiff(line, self.color)
                     if PYTHON3:
                         self.stream.flush()
                         self.stream.buffer.write(line)
                         self.stream.buffer.flush()
                     else:
                         self.stream.write(line)
                         self.stream.flush()
 
                 if servefail:
                     raise test.failureException(
                         'server failed to start (HGPORT=%s)' % test._startport
                     )
 
             # handle interactive prompt without releasing iolock
             if self._options.interactive:
                 if test.readrefout() != expected:
                     self.stream.write(
                         'Reference output has changed (run again to prompt '
                         'changes)'
                     )
                 else:
                     self.stream.write('Accept this change? [n] ')
                     self.stream.flush()
                     answer = sys.stdin.readline().strip()
                     if answer.lower() in ('y', 'yes'):
                         if test.path.endswith(b'.t'):
                             rename(test.errpath, test.path)
                         else:
                             rename(test.errpath, '%s.out' % test.path)
                         accepted = True
             if not accepted:
                 self.faildata[test.name] = b''.join(lines)
 
         return accepted
 
     def startTest(self, test):
         super(TestResult, self).startTest(test)
 
         # os.times module computes the user time and system time spent by
         # child's processes along with real elapsed time taken by a process.
         # This module has one limitation. It can only work for Linux user
         # and not for Windows. Hence why we fall back to another function
         # for wall time calculations.
         test.started_times = os.times()
         # TODO use a monotonic clock once support for Python 2.7 is dropped.
         test.started_time = time.time()
         if self._firststarttime is None:  # thread racy but irrelevant
             self._firststarttime = test.started_time
 
     def stopTest(self, test, interrupted=False):
         super(TestResult, self).stopTest(test)
 
         test.stopped_times = os.times()
         stopped_time = time.time()
 
         starttime = test.started_times
         endtime = test.stopped_times
         origin = self._firststarttime
         self.times.append(
             (
                 test.name,
                 endtime[2] - starttime[2],  # user space CPU time
                 endtime[3] - starttime[3],  # sys  space CPU time
                 stopped_time - test.started_time,  # real time
                 test.started_time - origin,  # start date in run context
                 stopped_time - origin,  # end date in run context
             )
         )
 
         if interrupted:
             with iolock:
                 self.stream.writeln(
                     'INTERRUPTED: %s (after %d seconds)'
                     % (test.name, self.times[-1][3])
                 )
 
 
 def getTestResult():
     """
     Returns the relevant test result
     """
     if "CUSTOM_TEST_RESULT" in os.environ:
         testresultmodule = __import__(os.environ["CUSTOM_TEST_RESULT"])
         return testresultmodule.TestResult
     else:
         return TestResult
 
 
 class TestSuite(unittest.TestSuite):
     """Custom unittest TestSuite that knows how to execute Mercurial tests."""
 
     def __init__(
         self,
         testdir,
         jobs=1,
         whitelist=None,
         blacklist=None,
         retest=False,
         keywords=None,
         loop=False,
         runs_per_test=1,
         loadtest=None,
         showchannels=False,
         *args,
         **kwargs
     ):
         """Create a new instance that can run tests with a configuration.
 
         testdir specifies the directory where tests are executed from. This
         is typically the ``tests`` directory from Mercurial's source
         repository.
 
         jobs specifies the number of jobs to run concurrently. Each test
         executes on its own thread. Tests actually spawn new processes, so
         state mutation should not be an issue.
 
         If there is only one job, it will use the main thread.
 
         whitelist and blacklist denote tests that have been whitelisted and
         blacklisted, respectively. These arguments don't belong in TestSuite.
         Instead, whitelist and blacklist should be handled by the thing that
         populates the TestSuite with tests. They are present to preserve
         backwards compatible behavior which reports skipped tests as part
         of the results.
 
         retest denotes whether to retest failed tests. This arguably belongs
         outside of TestSuite.
 
         keywords denotes key words that will be used to filter which tests
         to execute. This arguably belongs outside of TestSuite.
 
         loop denotes whether to loop over tests forever.
         """
         super(TestSuite, self).__init__(*args, **kwargs)
 
         self._jobs = jobs
         self._whitelist = whitelist
         self._blacklist = blacklist
         self._retest = retest
         self._keywords = keywords
         self._loop = loop
         self._runs_per_test = runs_per_test
         self._loadtest = loadtest
         self._showchannels = showchannels
 
     def run(self, result):
         # We have a number of filters that need to be applied. We do this
         # here instead of inside Test because it makes the running logic for
         # Test simpler.
         tests = []
         num_tests = [0]
         for test in self._tests:
 
             def get():
                 num_tests[0] += 1
                 if getattr(test, 'should_reload', False):
                     return self._loadtest(test, num_tests[0])
                 return test
 
             if not os.path.exists(test.path):
                 result.addSkip(test, "Doesn't exist")
                 continue
 
             if not (self._whitelist and test.bname in self._whitelist):
                 if self._blacklist and test.bname in self._blacklist:
                     result.addSkip(test, 'blacklisted')
                     continue
 
                 if self._retest and not os.path.exists(test.errpath):
                     result.addIgnore(test, 'not retesting')
                     continue
 
                 if self._keywords:
                     with open(test.path, 'rb') as f:
                         t = f.read().lower() + test.bname.lower()
                     ignored = False
                     for k in self._keywords.lower().split():
                         if k not in t:
                             result.addIgnore(test, "doesn't match keyword")
                             ignored = True
                             break
 
                     if ignored:
                         continue
             for _ in xrange(self._runs_per_test):
                 tests.append(get())
 
         runtests = list(tests)
         done = queue.Queue()
         running = 0
 
         channels = [""] * self._jobs
 
         def job(test, result):
             for n, v in enumerate(channels):
                 if not v:
                     channel = n
                     break
             else:
                 raise ValueError('Could not find output channel')
             channels[channel] = "=" + test.name[5:].split(".")[0]
             try:
                 test(result)
                 done.put(None)
             except KeyboardInterrupt:
                 pass
             except:  # re-raises
                 done.put(('!', test, 'run-test raised an error, see traceback'))
                 raise
             finally:
                 try:
                     channels[channel] = ''
                 except IndexError:
                     pass
 
         def stat():
             count = 0
             while channels:
                 d = '\n%03s  ' % count
                 for n, v in enumerate(channels):
                     if v:
                         d += v[0]
                         channels[n] = v[1:] or '.'
                     else:
                         d += ' '
                     d += ' '
                 with iolock:
                     sys.stdout.write(d + '  ')
                     sys.stdout.flush()
                 for x in xrange(10):
                     if channels:
                         time.sleep(0.1)
                 count += 1
 
         stoppedearly = False
 
         if self._showchannels:
             statthread = threading.Thread(target=stat, name="stat")
             statthread.start()
 
         try:
             while tests or running:
                 if not done.empty() or running == self._jobs or not tests:
                     try:
                         done.get(True, 1)
                         running -= 1
                         if result and result.shouldStop:
                             stoppedearly = True
                             break
                     except queue.Empty:
                         continue
                 if tests and not running == self._jobs:
                     test = tests.pop(0)
                     if self._loop:
                         if getattr(test, 'should_reload', False):
                             num_tests[0] += 1
                             tests.append(self._loadtest(test, num_tests[0]))
                         else:
                             tests.append(test)
                     if self._jobs == 1:
                         job(test, result)
                     else:
                         t = threading.Thread(
                             target=job, name=test.name, args=(test, result)
                         )
                         t.start()
                     running += 1
 
             # If we stop early we still need to wait on started tests to
             # finish. Otherwise, there is a race between the test completing
             # and the test's cleanup code running. This could result in the
             # test reporting incorrect.
             if stoppedearly:
                 while running:
                     try:
                         done.get(True, 1)
                         running -= 1
                     except queue.Empty:
                         continue
         except KeyboardInterrupt:
             for test in runtests:
                 test.abort()
 
         channels = []
 
         return result
 
 
 # Save the most recent 5 wall-clock runtimes of each test to a
 # human-readable text file named .testtimes. Tests are sorted
 # alphabetically, while times for each test are listed from oldest to
 # newest.
 
 
 def loadtimes(outputdir):
     times = []
     try:
         with open(os.path.join(outputdir, b'.testtimes')) as fp:
             for line in fp:
                 m = re.match('(.*?) ([0-9. ]+)', line)
                 times.append(
                     (m.group(1), [float(t) for t in m.group(2).split()])
                 )
     except IOError as err:
         if err.errno != errno.ENOENT:
             raise
     return times
 
 
 def savetimes(outputdir, result):
     saved = dict(loadtimes(outputdir))
     maxruns = 5
     skipped = {str(t[0]) for t in result.skipped}
     for tdata in result.times:
         test, real = tdata[0], tdata[3]
         if test not in skipped:
             ts = saved.setdefault(test, [])
             ts.append(real)
             ts[:] = ts[-maxruns:]
 
     fd, tmpname = tempfile.mkstemp(
         prefix=b'.testtimes', dir=outputdir, text=True
     )
     with os.fdopen(fd, 'w') as fp:
         for name, ts in sorted(saved.items()):
             fp.write('%s %s\n' % (name, ' '.join(['%.3f' % (t,) for t in ts])))
     timepath = os.path.join(outputdir, b'.testtimes')
     try:
         os.unlink(timepath)
     except OSError:
         pass
     try:
         os.rename(tmpname, timepath)
     except OSError:
         pass
 
 
 class TextTestRunner(unittest.TextTestRunner):
     """Custom unittest test runner that uses appropriate settings."""
 
     def __init__(self, runner, *args, **kwargs):
         super(TextTestRunner, self).__init__(*args, **kwargs)
 
         self._runner = runner
 
         self._result = getTestResult()(
             self._runner.options, self.stream, self.descriptions, self.verbosity
         )
 
     def listtests(self, test):
         test = sorted(test, key=lambda t: t.name)
 
         self._result.onStart(test)
 
         for t in test:
             print(t.name)
             self._result.addSuccess(t)
 
         if self._runner.options.xunit:
             with open(self._runner.options.xunit, "wb") as xuf:
                 self._writexunit(self._result, xuf)
 
         if self._runner.options.json:
             jsonpath = os.path.join(self._runner._outputdir, b'report.json')
             with open(jsonpath, 'w') as fp:
                 self._writejson(self._result, fp)
 
         return self._result
 
     def run(self, test):
         self._result.onStart(test)
         test(self._result)
 
         failed = len(self._result.failures)
         skipped = len(self._result.skipped)
         ignored = len(self._result.ignored)
 
         with iolock:
             self.stream.writeln('')
 
             if not self._runner.options.noskips:
                 for test, msg in sorted(
                     self._result.skipped, key=lambda s: s[0].name
                 ):
                     formatted = 'Skipped %s: %s\n' % (test.name, msg)
                     msg = highlightmsg(formatted, self._result.color)
                     self.stream.write(msg)
             for test, msg in sorted(
                 self._result.failures, key=lambda f: f[0].name
             ):
                 formatted = 'Failed %s: %s\n' % (test.name, msg)
                 self.stream.write(highlightmsg(formatted, self._result.color))
             for test, msg in sorted(
                 self._result.errors, key=lambda e: e[0].name
             ):
                 self.stream.writeln('Errored %s: %s' % (test.name, msg))
 
             if self._runner.options.xunit:
                 with open(self._runner.options.xunit, "wb") as xuf:
                     self._writexunit(self._result, xuf)
 
             if self._runner.options.json:
                 jsonpath = os.path.join(self._runner._outputdir, b'report.json')
                 with open(jsonpath, 'w') as fp:
                     self._writejson(self._result, fp)
 
             self._runner._checkhglib('Tested')
 
             savetimes(self._runner._outputdir, self._result)
 
             if failed and self._runner.options.known_good_rev:
                 self._bisecttests(t for t, m in self._result.failures)
             self.stream.writeln(
                 '# Ran %d tests, %d skipped, %d failed.'
                 % (self._result.testsRun, skipped + ignored, failed)
             )
             if failed:
                 self.stream.writeln(
                     'python hash seed: %s' % os.environ['PYTHONHASHSEED']
                 )
             if self._runner.options.time:
                 self.printtimes(self._result.times)
 
             if self._runner.options.exceptions:
                 exceptions = aggregateexceptions(
                     os.path.join(self._runner._outputdir, b'exceptions')
                 )
 
                 self.stream.writeln('Exceptions Report:')
                 self.stream.writeln(
                     '%d total from %d frames'
                     % (exceptions['total'], len(exceptions['exceptioncounts']))
                 )
                 combined = exceptions['combined']
                 for key in sorted(combined, key=combined.get, reverse=True):
                     frame, line, exc = key
                     totalcount, testcount, leastcount, leasttest = combined[key]
 
                     self.stream.writeln(
                         '%d (%d tests)\t%s: %s (%s - %d total)'
                         % (
                             totalcount,
                             testcount,
                             frame,
                             exc,
                             leasttest,
                             leastcount,
                         )
                     )
 
             self.stream.flush()
 
         return self._result
 
     def _bisecttests(self, tests):
         bisectcmd = ['hg', 'bisect']
         bisectrepo = self._runner.options.bisect_repo
         if bisectrepo:
             bisectcmd.extend(['-R', os.path.abspath(bisectrepo)])
 
         def pread(args):
             env = os.environ.copy()
             env['HGPLAIN'] = '1'
             p = subprocess.Popen(
                 args, stderr=subprocess.STDOUT, stdout=subprocess.PIPE, env=env
             )
             data = p.stdout.read()
             p.wait()
             return data
 
         for test in tests:
             pread(bisectcmd + ['--reset']),
             pread(bisectcmd + ['--bad', '.'])
             pread(bisectcmd + ['--good', self._runner.options.known_good_rev])
             # TODO: we probably need to forward more options
             # that alter hg's behavior inside the tests.
             opts = ''
             withhg = self._runner.options.with_hg
             if withhg:
                 opts += ' --with-hg=%s ' % shellquote(_bytes2sys(withhg))
             rtc = '%s %s %s %s' % (sysexecutable, sys.argv[0], opts, test)
             data = pread(bisectcmd + ['--command', rtc])
             m = re.search(
                 (
                     br'\nThe first (?P<goodbad>bad|good) revision '
                     br'is:\nchangeset: +\d+:(?P<node>[a-f0-9]+)\n.*\n'
                     br'summary: +(?P<summary>[^\n]+)\n'
                 ),
                 data,
                 (re.MULTILINE | re.DOTALL),
             )
             if m is None:
                 self.stream.writeln(
                     'Failed to identify failure point for %s' % test
                 )
                 continue
             dat = m.groupdict()
             verb = 'broken' if dat['goodbad'] == b'bad' else 'fixed'
             self.stream.writeln(
                 '%s %s by %s (%s)'
                 % (
                     test,
                     verb,
                     dat['node'].decode('ascii'),
                     dat['summary'].decode('utf8', 'ignore'),
                 )
             )
 
     def printtimes(self, times):
         # iolock held by run
         self.stream.writeln('# Producing time report')
         times.sort(key=lambda t: (t[3]))
         cols = '%7.3f %7.3f %7.3f %7.3f %7.3f   %s'
         self.stream.writeln(
             '%-7s %-7s %-7s %-7s %-7s   %s'
             % ('start', 'end', 'cuser', 'csys', 'real', 'Test')
         )
         for tdata in times:
             test = tdata[0]
             cuser, csys, real, start, end = tdata[1:6]
             self.stream.writeln(cols % (start, end, cuser, csys, real, test))
 
     @staticmethod
     def _writexunit(result, outf):
         # See http://llg.cubic.org/docs/junit/ for a reference.
         timesd = {t[0]: t[3] for t in result.times}
         doc = minidom.Document()
         s = doc.createElement('testsuite')
         s.setAttribute('errors', "0")  # TODO
         s.setAttribute('failures', str(len(result.failures)))
         s.setAttribute('name', 'run-tests')
         s.setAttribute(
             'skipped', str(len(result.skipped) + len(result.ignored))
         )
         s.setAttribute('tests', str(result.testsRun))
         doc.appendChild(s)
         for tc in result.successes:
             t = doc.createElement('testcase')
             t.setAttribute('name', tc.name)
             tctime = timesd.get(tc.name)
             if tctime is not None:
                 t.setAttribute('time', '%.3f' % tctime)
             s.appendChild(t)
         for tc, err in sorted(result.faildata.items()):
             t = doc.createElement('testcase')
             t.setAttribute('name', tc)
             tctime = timesd.get(tc)
             if tctime is not None:
                 t.setAttribute('time', '%.3f' % tctime)
             # createCDATASection expects a unicode or it will
             # convert using default conversion rules, which will
             # fail if string isn't ASCII.
             err = cdatasafe(err).decode('utf-8', 'replace')
             cd = doc.createCDATASection(err)
             # Use 'failure' here instead of 'error' to match errors = 0,
             # failures = len(result.failures) in the testsuite element.
             failelem = doc.createElement('failure')
             failelem.setAttribute('message', 'output changed')
             failelem.setAttribute('type', 'output-mismatch')
             failelem.appendChild(cd)
             t.appendChild(failelem)
             s.appendChild(t)
         for tc, message in result.skipped:
             # According to the schema, 'skipped' has no attributes. So store
             # the skip message as a text node instead.
             t = doc.createElement('testcase')
             t.setAttribute('name', tc.name)
             binmessage = message.encode('utf-8')
             message = cdatasafe(binmessage).decode('utf-8', 'replace')
             cd = doc.createCDATASection(message)
             skipelem = doc.createElement('skipped')
             skipelem.appendChild(cd)
             t.appendChild(skipelem)
             s.appendChild(t)
         outf.write(doc.toprettyxml(indent='  ', encoding='utf-8'))
 
     @staticmethod
     def _writejson(result, outf):
         timesd = {}
         for tdata in result.times:
             test = tdata[0]
             timesd[test] = tdata[1:]
 
         outcome = {}
         groups = [
             ('success', ((tc, None) for tc in result.successes)),
             ('failure', result.failures),
             ('skip', result.skipped),
         ]
         for res, testcases in groups:
             for tc, __ in testcases:
                 if tc.name in timesd:
                     diff = result.faildata.get(tc.name, b'')
                     try:
                         diff = diff.decode('unicode_escape')
                     except UnicodeDecodeError as e:
                         diff = '%r decoding diff, sorry' % e
                     tres = {
                         'result': res,
                         'time': ('%0.3f' % timesd[tc.name][2]),
                         'cuser': ('%0.3f' % timesd[tc.name][0]),
                         'csys': ('%0.3f' % timesd[tc.name][1]),
                         'start': ('%0.3f' % timesd[tc.name][3]),
                         'end': ('%0.3f' % timesd[tc.name][4]),
                         'diff': diff,
                     }
                 else:
                     # blacklisted test
                     tres = {'result': res}
 
                 outcome[tc.name] = tres
         jsonout = json.dumps(
             outcome, sort_keys=True, indent=4, separators=(',', ': ')
         )
         outf.writelines(("testreport =", jsonout))
 
 
 def sorttests(testdescs, previoustimes, shuffle=False):
     """Do an in-place sort of tests."""
     if shuffle:
         random.shuffle(testdescs)
         return
 
     if previoustimes:
 
         def sortkey(f):
             f = f['path']
             if f in previoustimes:
                 # Use most recent time as estimate
                 return -(previoustimes[f][-1])
             else:
                 # Default to a rather arbitrary value of 1 second for new tests
                 return -1.0
 
     else:
         # keywords for slow tests
         slow = {
             b'svn': 10,
             b'cvs': 10,
             b'hghave': 10,
             b'largefiles-update': 10,
             b'run-tests': 10,
             b'corruption': 10,
             b'race': 10,
             b'i18n': 10,
             b'check': 100,
             b'gendoc': 100,
             b'contrib-perf': 200,
             b'merge-combination': 100,
         }
         perf = {}
 
         def sortkey(f):
             # run largest tests first, as they tend to take the longest
             f = f['path']
             try:
                 return perf[f]
             except KeyError:
                 try:
                     val = -os.stat(f).st_size
                 except OSError as e:
                     if e.errno != errno.ENOENT:
                         raise
                     perf[f] = -1e9  # file does not exist, tell early
                     return -1e9
                 for kw, mul in slow.items():
                     if kw in f:
                         val *= mul
                 if f.endswith(b'.py'):
                     val /= 10.0
                 perf[f] = val / 1000.0
                 return perf[f]
 
     testdescs.sort(key=sortkey)
 
 
 class TestRunner(object):
     """Holds context for executing tests.
 
     Tests rely on a lot of state. This object holds it for them.
     """
 
     # Programs required to run tests.
     REQUIREDTOOLS = [
         b'diff',
         b'grep',
         b'unzip',
         b'gunzip',
         b'bunzip2',
         b'sed',
     ]
 
     # Maps file extensions to test class.
     TESTTYPES = [
         (b'.py', PythonTest),
         (b'.t', TTest),
     ]
 
     def __init__(self):
         self.options = None
         self._hgroot = None
         self._testdir = None
         self._outputdir = None
         self._hgtmp = None
         self._installdir = None
         self._bindir = None
         self._tmpbinddir = None
         self._pythondir = None
         self._coveragefile = None
         self._createdfiles = []
         self._hgcommand = None
         self._hgpath = None
         self._portoffset = 0
         self._ports = {}
 
     def run(self, args, parser=None):
         """Run the test suite."""
         oldmask = os.umask(0o22)
         try:
             parser = parser or getparser()
             options = parseargs(args, parser)
             tests = [_sys2bytes(a) for a in options.tests]
             if options.test_list is not None:
                 for listfile in options.test_list:
                     with open(listfile, 'rb') as f:
                         tests.extend(t for t in f.read().splitlines() if t)
             self.options = options
 
             self._checktools()
             testdescs = self.findtests(tests)
             if options.profile_runner:
                 import statprof
 
                 statprof.start()
             result = self._run(testdescs)
             if options.profile_runner:
                 statprof.stop()
                 statprof.display()
             return result
 
         finally:
             os.umask(oldmask)
 
     def _run(self, testdescs):
         testdir = getcwdb()
         self._testdir = osenvironb[b'TESTDIR'] = getcwdb()
         # assume all tests in same folder for now
         if testdescs:
             pathname = os.path.dirname(testdescs[0]['path'])
             if pathname:
                 testdir = os.path.join(testdir, pathname)
         self._testdir = osenvironb[b'TESTDIR'] = testdir
         if self.options.outputdir:
             self._outputdir = canonpath(_sys2bytes(self.options.outputdir))
         else:
             self._outputdir = getcwdb()
             if testdescs and pathname:
                 self._outputdir = os.path.join(self._outputdir, pathname)
         previoustimes = {}
         if self.options.order_by_runtime:
             previoustimes = dict(loadtimes(self._outputdir))
         sorttests(testdescs, previoustimes, shuffle=self.options.random)
 
         if 'PYTHONHASHSEED' not in os.environ:
             # use a random python hash seed all the time
             # we do the randomness ourself to know what seed is used
             os.environ['PYTHONHASHSEED'] = str(random.getrandbits(32))
 
         # Rayon (Rust crate for multi-threading) will use all logical CPU cores
         # by default, causing thrashing on high-cpu-count systems.
         # Setting its limit to 3 during tests should still let us uncover
         # multi-threading bugs while keeping the thrashing reasonable.
         os.environ.setdefault("RAYON_NUM_THREADS", "3")
 
         if self.options.tmpdir:
             self.options.keep_tmpdir = True
             tmpdir = _sys2bytes(self.options.tmpdir)
             if os.path.exists(tmpdir):
                 # Meaning of tmpdir has changed since 1.3: we used to create
                 # HGTMP inside tmpdir; now HGTMP is tmpdir.  So fail if
                 # tmpdir already exists.
                 print("error: temp dir %r already exists" % tmpdir)
                 return 1
 
             os.makedirs(tmpdir)
         else:
             d = None
             if os.name == 'nt':
                 # without this, we get the default temp dir location, but
                 # in all lowercase, which causes troubles with paths (issue3490)
                 d = osenvironb.get(b'TMP', None)
             tmpdir = tempfile.mkdtemp(b'', b'hgtests.', d)
 
         self._hgtmp = osenvironb[b'HGTMP'] = os.path.realpath(tmpdir)
 
         if self.options.with_hg:
             self._installdir = None
             whg = self.options.with_hg
             self._bindir = os.path.dirname(os.path.realpath(whg))
             assert isinstance(self._bindir, bytes)
             self._hgcommand = os.path.basename(whg)
             self._tmpbindir = os.path.join(self._hgtmp, b'install', b'bin')
             os.makedirs(self._tmpbindir)
 
             normbin = os.path.normpath(os.path.abspath(whg))
             normbin = normbin.replace(_sys2bytes(os.sep), b'/')
 
             # Other Python scripts in the test harness need to
             # `import mercurial`. If `hg` is a Python script, we assume
             # the Mercurial modules are relative to its path and tell the tests
             # to load Python modules from its directory.
             with open(whg, 'rb') as fh:
                 initial = fh.read(1024)
 
             if re.match(b'#!.*python', initial):
                 self._pythondir = self._bindir
             # If it looks like our in-repo Rust binary, use the source root.
             # This is a bit hacky. But rhg is still not supported outside the
             # source directory. So until it is, do the simple thing.
             elif re.search(b'/rust/target/[^/]+/hg', normbin):
                 self._pythondir = os.path.dirname(self._testdir)
             # Fall back to the legacy behavior.
             else:
                 self._pythondir = self._bindir
 
         else:
             self._installdir = os.path.join(self._hgtmp, b"install")
             self._bindir = os.path.join(self._installdir, b"bin")
             self._hgcommand = b'hg'
             self._tmpbindir = self._bindir
             self._pythondir = os.path.join(self._installdir, b"lib", b"python")
 
         # Force the use of hg.exe instead of relying on MSYS to recognize hg is
         # a python script and feed it to python.exe.  Legacy stdio is force
         # enabled by hg.exe, and this is a more realistic way to launch hg
         # anyway.
         if os.name == 'nt' and not self._hgcommand.endswith(b'.exe'):
             self._hgcommand += b'.exe'
 
         # set CHGHG, then replace "hg" command by "chg"
         chgbindir = self._bindir
         if self.options.chg or self.options.with_chg:
             osenvironb[b'CHGHG'] = os.path.join(self._bindir, self._hgcommand)
         else:
             osenvironb.pop(b'CHGHG', None)  # drop flag for hghave
         if self.options.chg:
             self._hgcommand = b'chg'
         elif self.options.with_chg:
             chgbindir = os.path.dirname(os.path.realpath(self.options.with_chg))
             self._hgcommand = os.path.basename(self.options.with_chg)
 
         osenvironb[b"BINDIR"] = self._bindir
         osenvironb[b"PYTHON"] = PYTHON
 
         fileb = _sys2bytes(__file__)
         runtestdir = os.path.abspath(os.path.dirname(fileb))
         osenvironb[b'RUNTESTDIR'] = runtestdir
         if PYTHON3:
             sepb = _sys2bytes(os.pathsep)
         else:
             sepb = os.pathsep
         path = [self._bindir, runtestdir] + osenvironb[b"PATH"].split(sepb)
         if os.path.islink(__file__):
             # test helper will likely be at the end of the symlink
             realfile = os.path.realpath(fileb)
             realdir = os.path.abspath(os.path.dirname(realfile))
             path.insert(2, realdir)
         if chgbindir != self._bindir:
             path.insert(1, chgbindir)
         if self._testdir != runtestdir:
             path = [self._testdir] + path
         if self._tmpbindir != self._bindir:
             path = [self._tmpbindir] + path
         osenvironb[b"PATH"] = sepb.join(path)
 
         # Include TESTDIR in PYTHONPATH so that out-of-tree extensions
         # can run .../tests/run-tests.py test-foo where test-foo
         # adds an extension to HGRC. Also include run-test.py directory to
         # import modules like heredoctest.
         pypath = [self._pythondir, self._testdir, runtestdir]
         # We have to augment PYTHONPATH, rather than simply replacing
         # it, in case external libraries are only available via current
         # PYTHONPATH.  (In particular, the Subversion bindings on OS X
         # are in /opt/subversion.)
         oldpypath = osenvironb.get(IMPL_PATH)
         if oldpypath:
             pypath.append(oldpypath)
         osenvironb[IMPL_PATH] = sepb.join(pypath)
 
         if self.options.pure:
             os.environ["HGTEST_RUN_TESTS_PURE"] = "--pure"
             os.environ["HGMODULEPOLICY"] = "py"
         if self.options.rust:
             os.environ["HGMODULEPOLICY"] = "rust+c"
         if self.options.no_rust:
             current_policy = os.environ.get("HGMODULEPOLICY", "")
             if current_policy.startswith("rust+"):
                 os.environ["HGMODULEPOLICY"] = current_policy[len("rust+") :]
             os.environ.pop("HGWITHRUSTEXT", None)
 
         if self.options.allow_slow_tests:
             os.environ["HGTEST_SLOW"] = "slow"
         elif 'HGTEST_SLOW' in os.environ:
             del os.environ['HGTEST_SLOW']
 
         self._coveragefile = os.path.join(self._testdir, b'.coverage')
 
         if self.options.exceptions:
             exceptionsdir = os.path.join(self._outputdir, b'exceptions')
             try:
                 os.makedirs(exceptionsdir)
             except OSError as e:
                 if e.errno != errno.EEXIST:
                     raise
 
             # Remove all existing exception reports.
             for f in os.listdir(exceptionsdir):
                 os.unlink(os.path.join(exceptionsdir, f))
 
             osenvironb[b'HGEXCEPTIONSDIR'] = exceptionsdir
             logexceptions = os.path.join(self._testdir, b'logexceptions.py')
             self.options.extra_config_opt.append(
                 'extensions.logexceptions=%s' % logexceptions.decode('utf-8')
             )
 
         vlog("# Using TESTDIR", _bytes2sys(self._testdir))
         vlog("# Using RUNTESTDIR", _bytes2sys(osenvironb[b'RUNTESTDIR']))
         vlog("# Using HGTMP", _bytes2sys(self._hgtmp))
         vlog("# Using PATH", os.environ["PATH"])
         vlog(
             "# Using", _bytes2sys(IMPL_PATH), _bytes2sys(osenvironb[IMPL_PATH]),
         )
         vlog("# Writing to directory", _bytes2sys(self._outputdir))
 
         try:
             return self._runtests(testdescs) or 0
         finally:
             time.sleep(0.1)
             self._cleanup()
 
     def findtests(self, args):
         """Finds possible test files from arguments.
 
         If you wish to inject custom tests into the test harness, this would
         be a good function to monkeypatch or override in a derived class.
         """
         if not args:
             if self.options.changed:
                 proc = Popen4(
                     b'hg st --rev "%s" -man0 .'
                     % _sys2bytes(self.options.changed),
                     None,
                     0,
                 )
                 stdout, stderr = proc.communicate()
                 args = stdout.strip(b'\0').split(b'\0')
             else:
                 args = os.listdir(b'.')
 
         expanded_args = []
         for arg in args:
             if os.path.isdir(arg):
                 if not arg.endswith(b'/'):
                     arg += b'/'
                 expanded_args.extend([arg + a for a in os.listdir(arg)])
             else:
                 expanded_args.append(arg)
         args = expanded_args
 
         testcasepattern = re.compile(br'([\w-]+\.t|py)(?:#([a-zA-Z0-9_\-.#]+))')
         tests = []
         for t in args:
             case = []
 
             if not (
                 os.path.basename(t).startswith(b'test-')
                 and (t.endswith(b'.py') or t.endswith(b'.t'))
             ):
 
                 m = testcasepattern.match(os.path.basename(t))
                 if m is not None:
                     t_basename, casestr = m.groups()
                     t = os.path.join(os.path.dirname(t), t_basename)
                     if casestr:
                         case = casestr.split(b'#')
                 else:
                     continue
 
             if t.endswith(b'.t'):
                 # .t file may contain multiple test cases
                 casedimensions = parsettestcases(t)
                 if casedimensions:
                     cases = []
 
                     def addcases(case, casedimensions):
                         if not casedimensions:
                             cases.append(case)
                         else:
                             for c in casedimensions[0]:
                                 addcases(case + [c], casedimensions[1:])
 
                     addcases([], casedimensions)
                     if case and case in cases:
                         cases = [case]
                     elif case:
                         # Ignore invalid cases
                         cases = []
                     else:
                         pass
                     tests += [{'path': t, 'case': c} for c in sorted(cases)]
                 else:
                     tests.append({'path': t})
             else:
                 tests.append({'path': t})
         return tests
 
     def _runtests(self, testdescs):
         def _reloadtest(test, i):
             # convert a test back to its description dict
             desc = {'path': test.path}
             case = getattr(test, '_case', [])
             if case:
                 desc['case'] = case
             return self._gettest(desc, i)
 
         try:
             if self.options.restart:
                 orig = list(testdescs)
                 while testdescs:
                     desc = testdescs[0]
                     # desc['path'] is a relative path
                     if 'case' in desc:
                         casestr = b'#'.join(desc['case'])
                         errpath = b'%s#%s.err' % (desc['path'], casestr)
                     else:
                         errpath = b'%s.err' % desc['path']
                     errpath = os.path.join(self._outputdir, errpath)
                     if os.path.exists(errpath):
                         break
                     testdescs.pop(0)
                 if not testdescs:
                     print("running all tests")
                     testdescs = orig
 
             tests = [self._gettest(d, i) for i, d in enumerate(testdescs)]
             num_tests = len(tests) * self.options.runs_per_test
 
             jobs = min(num_tests, self.options.jobs)
 
             failed = False
             kws = self.options.keywords
             if kws is not None and PYTHON3:
                 kws = kws.encode('utf-8')
 
             suite = TestSuite(
                 self._testdir,
                 jobs=jobs,
                 whitelist=self.options.whitelisted,
                 blacklist=self.options.blacklist,
                 retest=self.options.retest,
                 keywords=kws,
                 loop=self.options.loop,
                 runs_per_test=self.options.runs_per_test,
                 showchannels=self.options.showchannels,
                 tests=tests,
                 loadtest=_reloadtest,
             )
             verbosity = 1
             if self.options.list_tests:
                 verbosity = 0
             elif self.options.verbose:
                 verbosity = 2
             runner = TextTestRunner(self, verbosity=verbosity)
 
             if self.options.list_tests:
                 result = runner.listtests(suite)
             else:
                 if self._installdir:
                     self._installhg()
                     self._checkhglib("Testing")
                 else:
                     self._usecorrectpython()
                 if self.options.chg:
                     assert self._installdir
                     self._installchg()
 
                 log(
                     'running %d tests using %d parallel processes'
                     % (num_tests, jobs)
                 )
 
                 result = runner.run(suite)
 
             if result.failures or result.errors:
                 failed = True
 
             result.onEnd()
 
             if self.options.anycoverage:
                 self._outputcoverage()
         except KeyboardInterrupt:
             failed = True
             print("\ninterrupted!")
 
         if failed:
             return 1
 
     def _getport(self, count):
         port = self._ports.get(count)  # do we have a cached entry?
         if port is None:
             portneeded = 3
             # above 100 tries we just give up and let test reports failure
             for tries in xrange(100):
                 allfree = True
                 port = self.options.port + self._portoffset
                 for idx in xrange(portneeded):
                     if not checkportisavailable(port + idx):
                         allfree = False
                         break
                 self._portoffset += portneeded
                 if allfree:
                     break
             self._ports[count] = port
         return port
 
     def _gettest(self, testdesc, count):
         """Obtain a Test by looking at its filename.
 
         Returns a Test instance. The Test may not be runnable if it doesn't
         map to a known type.
         """
         path = testdesc['path']
         lctest = path.lower()
         testcls = Test
 
         for ext, cls in self.TESTTYPES:
             if lctest.endswith(ext):
                 testcls = cls
                 break
 
         refpath = os.path.join(getcwdb(), path)
         tmpdir = os.path.join(self._hgtmp, b'child%d' % count)
 
         # extra keyword parameters. 'case' is used by .t tests
         kwds = {k: testdesc[k] for k in ['case'] if k in testdesc}
 
         t = testcls(
             refpath,
             self._outputdir,
             tmpdir,
             keeptmpdir=self.options.keep_tmpdir,
             debug=self.options.debug,
             first=self.options.first,
             timeout=self.options.timeout,
             startport=self._getport(count),
             extraconfigopts=self.options.extra_config_opt,
             shell=self.options.shell,
             hgcommand=self._hgcommand,
             usechg=bool(self.options.with_chg or self.options.chg),
             useipv6=useipv6,
             **kwds
         )
         t.should_reload = True
         return t
 
     def _cleanup(self):
         """Clean up state from this test invocation."""
         if self.options.keep_tmpdir:
             return
 
         vlog("# Cleaning up HGTMP", _bytes2sys(self._hgtmp))
         shutil.rmtree(self._hgtmp, True)
         for f in self._createdfiles:
             try:
                 os.remove(f)
             except OSError:
                 pass
 
     def _usecorrectpython(self):
         """Configure the environment to use the appropriate Python in tests."""
         # Tests must use the same interpreter as us or bad things will happen.
         pyexename = sys.platform == 'win32' and b'python.exe' or b'python'
 
         # os.symlink() is a thing with py3 on Windows, but it requires
         # Administrator rights.
         if getattr(os, 'symlink', None) and os.name != 'nt':
             vlog(
                 "# Making python executable in test path a symlink to '%s'"
                 % sysexecutable
             )
             mypython = os.path.join(self._tmpbindir, pyexename)
             try:
                 if os.readlink(mypython) == sysexecutable:
                     return
                 os.unlink(mypython)
             except OSError as err:
                 if err.errno != errno.ENOENT:
                     raise
             if self._findprogram(pyexename) != sysexecutable:
                 try:
                     os.symlink(sysexecutable, mypython)
                     self._createdfiles.append(mypython)
                 except OSError as err:
                     # child processes may race, which is harmless
                     if err.errno != errno.EEXIST:
                         raise
         else:
             exedir, exename = os.path.split(sysexecutable)
             vlog(
                 "# Modifying search path to find %s as %s in '%s'"
                 % (exename, pyexename, exedir)
             )
             path = os.environ['PATH'].split(os.pathsep)
             while exedir in path:
                 path.remove(exedir)
             os.environ['PATH'] = os.pathsep.join([exedir] + path)
             if not self._findprogram(pyexename):
                 print("WARNING: Cannot find %s in search path" % pyexename)
 
     def _installhg(self):
         """Install hg into the test environment.
 
         This will also configure hg with the appropriate testing settings.
         """
         vlog("# Performing temporary installation of HG")
         installerrs = os.path.join(self._hgtmp, b"install.err")
         compiler = ''
         if self.options.compiler:
             compiler = '--compiler ' + self.options.compiler
         setup_opts = b""
         if self.options.pure:
             setup_opts = b"--pure"
         elif self.options.rust:
             setup_opts = b"--rust"
         elif self.options.no_rust:
             setup_opts = b"--no-rust"
 
         # Run installer in hg root
         script = os.path.realpath(sys.argv[0])
         exe = sysexecutable
         if PYTHON3:
             compiler = _sys2bytes(compiler)
             script = _sys2bytes(script)
             exe = _sys2bytes(exe)
         hgroot = os.path.dirname(os.path.dirname(script))
         self._hgroot = hgroot
         os.chdir(hgroot)
         nohome = b'--home=""'
         if os.name == 'nt':
             # The --home="" trick works only on OS where os.sep == '/'
             # because of a distutils convert_path() fast-path. Avoid it at
             # least on Windows for now, deal with .pydistutils.cfg bugs
             # when they happen.
             nohome = b''
         cmd = (
             b'"%(exe)s" setup.py %(setup_opts)s clean --all'
             b' build %(compiler)s --build-base="%(base)s"'
             b' install --force --prefix="%(prefix)s"'
             b' --install-lib="%(libdir)s"'
             b' --install-scripts="%(bindir)s" %(nohome)s >%(logfile)s 2>&1'
             % {
                 b'exe': exe,
                 b'setup_opts': setup_opts,
                 b'compiler': compiler,
                 b'base': os.path.join(self._hgtmp, b"build"),
                 b'prefix': self._installdir,
                 b'libdir': self._pythondir,
                 b'bindir': self._bindir,
                 b'nohome': nohome,
                 b'logfile': installerrs,
             }
         )
 
         # setuptools requires install directories to exist.
         def makedirs(p):
             try:
                 os.makedirs(p)
             except OSError as e:
                 if e.errno != errno.EEXIST:
                     raise
 
         makedirs(self._pythondir)
         makedirs(self._bindir)
 
         vlog("# Running", cmd.decode("utf-8"))
         if subprocess.call(_bytes2sys(cmd), shell=True) == 0:
             if not self.options.verbose:
                 try:
                     os.remove(installerrs)
                 except OSError as e:
                     if e.errno != errno.ENOENT:
                         raise
         else:
             with open(installerrs, 'rb') as f:
                 for line in f:
                     if PYTHON3:
                         sys.stdout.buffer.write(line)
                     else:
                         sys.stdout.write(line)
             sys.exit(1)
         os.chdir(self._testdir)
 
         self._usecorrectpython()
 
         hgbat = os.path.join(self._bindir, b'hg.bat')
         if os.path.isfile(hgbat):
             # hg.bat expects to be put in bin/scripts while run-tests.py
             # installation layout put it in bin/ directly. Fix it
             with open(hgbat, 'rb') as f:
                 data = f.read()
             if br'"%~dp0..\python" "%~dp0hg" %*' in data:
                 data = data.replace(
                     br'"%~dp0..\python" "%~dp0hg" %*',
                     b'"%~dp0python" "%~dp0hg" %*',
                 )
                 with open(hgbat, 'wb') as f:
                     f.write(data)
             else:
                 print('WARNING: cannot fix hg.bat reference to python.exe')
 
         if self.options.anycoverage:
             custom = os.path.join(
                 osenvironb[b'RUNTESTDIR'], b'sitecustomize.py'
             )
             target = os.path.join(self._pythondir, b'sitecustomize.py')
             vlog('# Installing coverage trigger to %s' % target)
             shutil.copyfile(custom, target)
             rc = os.path.join(self._testdir, b'.coveragerc')
             vlog('# Installing coverage rc to %s' % rc)
             osenvironb[b'COVERAGE_PROCESS_START'] = rc
             covdir = os.path.join(self._installdir, b'..', b'coverage')
             try:
                 os.mkdir(covdir)
             except OSError as e:
                 if e.errno != errno.EEXIST:
                     raise
 
             osenvironb[b'COVERAGE_DIR'] = covdir
 
     def _checkhglib(self, verb):
         """Ensure that the 'mercurial' package imported by python is
         the one we expect it to be.  If not, print a warning to stderr."""
         if (self._bindir == self._pythondir) and (
             self._bindir != self._tmpbindir
         ):
             # The pythondir has been inferred from --with-hg flag.
             # We cannot expect anything sensible here.
             return
         expecthg = os.path.join(self._pythondir, b'mercurial')
         actualhg = self._gethgpath()
         if os.path.abspath(actualhg) != os.path.abspath(expecthg):
             sys.stderr.write(
                 'warning: %s with unexpected mercurial lib: %s\n'
                 '         (expected %s)\n' % (verb, actualhg, expecthg)
             )
 
     def _gethgpath(self):
         """Return the path to the mercurial package that is actually found by
         the current Python interpreter."""
         if self._hgpath is not None:
             return self._hgpath
 
         cmd = b'"%s" -c "import mercurial; print (mercurial.__path__[0])"'
         cmd = cmd % PYTHON
         if PYTHON3:
             cmd = _bytes2sys(cmd)
 
         p = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True)
         out, err = p.communicate()
 
         self._hgpath = out.strip()
 
         return self._hgpath
 
     def _installchg(self):
         """Install chg into the test environment"""
         vlog('# Performing temporary installation of CHG')
         assert os.path.dirname(self._bindir) == self._installdir
         assert self._hgroot, 'must be called after _installhg()'
         cmd = b'"%(make)s" clean install PREFIX="%(prefix)s"' % {
             b'make': b'make',  # TODO: switch by option or environment?
             b'prefix': self._installdir,
         }
         cwd = os.path.join(self._hgroot, b'contrib', b'chg')
         vlog("# Running", cmd)
         proc = subprocess.Popen(
             cmd,
             shell=True,
             cwd=cwd,
             stdin=subprocess.PIPE,
             stdout=subprocess.PIPE,
             stderr=subprocess.STDOUT,
         )
         out, _err = proc.communicate()
         if proc.returncode != 0:
             if PYTHON3:
                 sys.stdout.buffer.write(out)
             else:
                 sys.stdout.write(out)
             sys.exit(1)
 
     def _outputcoverage(self):
         """Produce code coverage output."""
         import coverage
 
         coverage = coverage.coverage
 
         vlog('# Producing coverage report')
         # chdir is the easiest way to get short, relative paths in the
         # output.
         os.chdir(self._hgroot)
         covdir = os.path.join(_bytes2sys(self._installdir), '..', 'coverage')
         cov = coverage(data_file=os.path.join(covdir, 'cov'))
 
         # Map install directory paths back to source directory.
         cov.config.paths['srcdir'] = ['.', _bytes2sys(self._pythondir)]
 
         cov.combine()
 
         omit = [
             _bytes2sys(os.path.join(x, b'*'))
             for x in [self._bindir, self._testdir]
         ]
         cov.report(ignore_errors=True, omit=omit)
 
         if self.options.htmlcov:
             htmldir = os.path.join(_bytes2sys(self._outputdir), 'htmlcov')
             cov.html_report(directory=htmldir, omit=omit)
         if self.options.annotate:
             adir = os.path.join(_bytes2sys(self._outputdir), 'annotated')
             if not os.path.isdir(adir):
                 os.mkdir(adir)
             cov.annotate(directory=adir, omit=omit)
 
     def _findprogram(self, program):
         """Search PATH for a executable program"""
         dpb = _sys2bytes(os.defpath)
         sepb = _sys2bytes(os.pathsep)
         for p in osenvironb.get(b'PATH', dpb).split(sepb):
             name = os.path.join(p, program)
             if os.name == 'nt' or os.access(name, os.X_OK):
                 return name
         return None
 
     def _checktools(self):
         """Ensure tools required to run tests are present."""
         for p in self.REQUIREDTOOLS:
             if os.name == 'nt' and not p.endswith(b'.exe'):
                 p += b'.exe'
             found = self._findprogram(p)
             p = p.decode("utf-8")
             if found:
                 vlog("# Found prerequisite", p, "at", _bytes2sys(found))
             else:
                 print("WARNING: Did not find prerequisite tool: %s " % p)
 
 
 def aggregateexceptions(path):
     exceptioncounts = collections.Counter()
     testsbyfailure = collections.defaultdict(set)
     failuresbytest = collections.defaultdict(set)
 
     for f in os.listdir(path):
         with open(os.path.join(path, f), 'rb') as fh:
             data = fh.read().split(b'\0')
             if len(data) != 5:
                 continue
 
             exc, mainframe, hgframe, hgline, testname = data
             exc = exc.decode('utf-8')
             mainframe = mainframe.decode('utf-8')
             hgframe = hgframe.decode('utf-8')
             hgline = hgline.decode('utf-8')
             testname = testname.decode('utf-8')
 
             key = (hgframe, hgline, exc)
             exceptioncounts[key] += 1
             testsbyfailure[key].add(testname)
             failuresbytest[testname].add(key)
 
     # Find test having fewest failures for each failure.
     leastfailing = {}
     for key, tests in testsbyfailure.items():
         fewesttest = None
         fewestcount = 99999999
         for test in sorted(tests):
             if len(failuresbytest[test]) < fewestcount:
                 fewesttest = test
                 fewestcount = len(failuresbytest[test])
 
         leastfailing[key] = (fewestcount, fewesttest)
 
     # Create a combined counter so we can sort by total occurrences and
     # impacted tests.
     combined = {}
     for key in exceptioncounts:
         combined[key] = (
             exceptioncounts[key],
             len(testsbyfailure[key]),
             leastfailing[key][0],
             leastfailing[key][1],
         )
 
     return {
         'exceptioncounts': exceptioncounts,
         'total': sum(exceptioncounts.values()),
         'combined': combined,
         'leastfailing': leastfailing,
         'byfailure': testsbyfailure,
         'bytest': failuresbytest,
     }
 
 
 if __name__ == '__main__':
     runner = TestRunner()
 
     try:
         import msvcrt
 
         msvcrt.setmode(sys.stdin.fileno(), os.O_BINARY)
         msvcrt.setmode(sys.stdout.fileno(), os.O_BINARY)
         msvcrt.setmode(sys.stderr.fileno(), os.O_BINARY)
     except ImportError:
         pass
 
     sys.exit(runner.run(sys.argv[1:]))