Create an atomic opener that does not automatically rename on close

The revlog.checkinlinesize() uses an atomic opener to replace the
index file after converting it from inline to traditional .i and .d
files.  If this operation is interrupted, the atomic file class can
overwrite a valid file with a partially written one.

This patch introduces an atomic opener that does not automatically
replace the destination file with the tempfile.  This way
an interrupted checkinlinesize() call turns into a noop.

mercurial/revlog.py

@@ -1,1064 +1,1066 b''
 """
 revlog.py - storage back-end for mercurial
 
 This provides efficient delta storage with O(1) retrieve and append
 and O(changes) merge between branches
 
 Copyright 2005 Matt Mackall <[e-mail]>
 
 This software may be used and distributed according to the terms
 of the GNU General Public License, incorporated herein by reference.
 """
 
 from node import *
 from i18n import gettext as _
 from demandload import demandload
 demandload(globals(), "binascii changegroup errno heapq mdiff os")
 demandload(globals(), "sha struct zlib")
 
 # revlog version strings
 REVLOGV0 = 0
 REVLOGNG = 1
 
 # revlog flags
 REVLOGNGINLINEDATA = (1 << 16)
 
 def flagstr(flag):
     if flag == "inline":
         return REVLOGNGINLINEDATA
     raise RevlogError(_("unknown revlog flag %s" % flag))
 
 def hash(text, p1, p2):
     """generate a hash from the given text and its parent hashes
 
     This hash combines both the current file contents and its history
     in a manner that makes it easy to distinguish nodes with the same
     content in the revision graph.
     """
     l = [p1, p2]
     l.sort()
     s = sha.new(l[0])
     s.update(l[1])
     s.update(text)
     return s.digest()
 
 def compress(text):
     """ generate a possibly-compressed representation of text """
     if not text: return ("", text)
     if len(text) < 44:
         if text[0] == '\0': return ("", text)
         return ('u', text)
     bin = zlib.compress(text)
     if len(bin) > len(text):
         if text[0] == '\0': return ("", text)
         return ('u', text)
     return ("", bin)
 
 def decompress(bin):
     """ decompress the given input """
     if not bin: return bin
     t = bin[0]
     if t == '\0': return bin
     if t == 'x': return zlib.decompress(bin)
     if t == 'u': return bin[1:]
     raise RevlogError(_("unknown compression type %r") % t)
 
 indexformatv0 = ">4l20s20s20s"
 # index ng:
 # 6 bytes offset
 # 2 bytes flags
 # 4 bytes compressed length
 # 4 bytes uncompressed length
 # 4 bytes: base rev
 # 4 bytes link rev
 # 4 bytes parent 1 rev
 # 4 bytes parent 2 rev
 # 32 bytes: nodeid
 indexformatng = ">Qiiiiii20s12x"
 versionformat = ">i"
 
 class lazyparser(object):
     """
     this class avoids the need to parse the entirety of large indices
 
     By default we parse and load 1000 entries at a time.
 
     If no position is specified, we load the whole index, and replace
     the lazy objects in revlog with the underlying objects for
     efficiency in cases where we look at most of the nodes.
     """
     def __init__(self, data, revlog, indexformat):
         self.data = data
         self.s = struct.calcsize(indexformat)
         self.indexformat = indexformat
         self.l = len(data)/self.s
         self.index = [None] * self.l
         self.map = {nullid: -1}
         self.all = 0
         self.revlog = revlog
 
     def load(self, pos=None):
         if self.all: return
         if pos is not None:
             block = pos / 1000
             i = block * 1000
             end = min(self.l, i + 1000)
         else:
             self.all = 1
             i = 0
             end = self.l
             self.revlog.index = self.index
             self.revlog.nodemap = self.map
 
         while i < end:
             if not self.index[i]:
                 d = self.data[i * self.s: (i + 1) * self.s]
                 e = struct.unpack(self.indexformat, d)
                 self.index[i] = e
                 self.map[e[-1]] = i
             i += 1
 
 class lazyindex(object):
     """a lazy version of the index array"""
     def __init__(self, parser):
         self.p = parser
     def __len__(self):
         return len(self.p.index)
     def load(self, pos):
         if pos < 0:
             pos += len(self.p.index)
         self.p.load(pos)
         return self.p.index[pos]
     def __getitem__(self, pos):
         return self.p.index[pos] or self.load(pos)
     def __setitem__(self, pos, item):
         self.p.index[pos] = item
     def __delitem__(self, pos):
         del self.p.index[pos]
     def append(self, e):
         self.p.index.append(e)
 
 class lazymap(object):
     """a lazy version of the node map"""
     def __init__(self, parser):
         self.p = parser
     def load(self, key):
         if self.p.all: return
         n = self.p.data.find(key)
         if n < 0:
             raise KeyError(key)
         pos = n / self.p.s
         self.p.load(pos)
     def __contains__(self, key):
         self.p.load()
         return key in self.p.map
     def __iter__(self):
         yield nullid
         for i in xrange(self.p.l):
             try:
                 yield self.p.index[i][-1]
             except:
                 self.p.load(i)
                 yield self.p.index[i][-1]
     def __getitem__(self, key):
         try:
             return self.p.map[key]
         except KeyError:
             try:
                 self.load(key)
                 return self.p.map[key]
             except KeyError:
                 raise KeyError("node " + hex(key))
     def __setitem__(self, key, val):
         self.p.map[key] = val
     def __delitem__(self, key):
         del self.p.map[key]
 
 class RevlogError(Exception): pass
 
 class revlog(object):
     """
     the underlying revision storage object
 
     A revlog consists of two parts, an index and the revision data.
 
     The index is a file with a fixed record size containing
     information on each revision, includings its nodeid (hash), the
     nodeids of its parents, the position and offset of its data within
     the data file, and the revision it's based on. Finally, each entry
     contains a linkrev entry that can serve as a pointer to external
     data.
 
     The revision data itself is a linear collection of data chunks.
     Each chunk represents a revision and is usually represented as a
     delta against the previous chunk. To bound lookup time, runs of
     deltas are limited to about 2 times the length of the original
     version data. This makes retrieval of a version proportional to
     its size, or O(1) relative to the number of revisions.
 
     Both pieces of the revlog are written to in an append-only
     fashion, which means we never need to rewrite a file to insert or
     remove data, and can use some simple techniques to avoid the need
     for locking while reading.
     """
     def __init__(self, opener, indexfile, datafile, defversion=0):
         """
         create a revlog object
 
         opener is a function that abstracts the file opening operation
         and can be used to implement COW semantics or the like.
         """
         self.indexfile = indexfile
         self.datafile = datafile
         self.opener = opener
 
         self.indexstat = None
         self.cache = None
         self.chunkcache = None
         self.defversion = defversion
         self.load()
 
     def load(self):
         v = self.defversion
         try:
             f = self.opener(self.indexfile)
             i = f.read()
         except IOError, inst:
             if inst.errno != errno.ENOENT:
                 raise
             i = ""
         else:
             try:
                 st = os.fstat(f.fileno())
             except AttributeError, inst:
                 st = None
             else:
                 oldst = self.indexstat
                 if (oldst and st.st_dev == oldst.st_dev
                     and st.st_ino == oldst.st_ino
                     and st.st_mtime == oldst.st_mtime
                     and st.st_ctime == oldst.st_ctime):
                     return
                 self.indexstat = st
                 if len(i) > 0:
                     v = struct.unpack(versionformat, i[:4])[0]
         flags = v & ~0xFFFF
         fmt = v & 0xFFFF
         if fmt == 0:
             if flags:
                 raise RevlogError(_("index %s invalid flags %x for format v0" %
                                    (self.indexfile, flags)))
         elif fmt == REVLOGNG:
             if flags & ~REVLOGNGINLINEDATA:
                 raise RevlogError(_("index %s invalid flags %x for revlogng" %
                                    (self.indexfile, flags)))
         else:
             raise RevlogError(_("index %s invalid format %d" %
                                (self.indexfile, fmt)))
         self.version = v
         if v == 0:
             self.indexformat = indexformatv0
         else:
             self.indexformat = indexformatng
 
         if i:
             if not self.inlinedata() and st and st.st_size > 10000:
                 # big index, let's parse it on demand
                 parser = lazyparser(i, self, self.indexformat)
                 self.index = lazyindex(parser)
                 self.nodemap = lazymap(parser)
             else:
                 self.parseindex(i)
             if self.inlinedata():
                 # we've already got the entire data file read in, save it
                 # in the chunk data
                 self.chunkcache = (0, i)
             if self.version != 0:
                 e = list(self.index[0])
                 type = self.ngtype(e[0])
                 e[0] = self.offset_type(0, type)
                 self.index[0] = e
         else:
             self.nodemap = { nullid: -1}
             self.index = []
 
 
     def parseindex(self, data):
         s = struct.calcsize(self.indexformat)
         l = len(data)
         self.index = []
         self.nodemap =  {nullid: -1}
         inline = self.inlinedata()
         off = 0
         n = 0
         while off < l:
             e = struct.unpack(self.indexformat, data[off:off + s])
             self.index.append(e)
             self.nodemap[e[-1]] = n
             n += 1
             off += s
             if inline:
                 off += e[1]
 
     def ngoffset(self, q):
         if q & 0xFFFF:
             raise RevlogError(_('%s: incompatible revision flag %x') %
                                (self.indexfile, type))
         return long(q >> 16)
 
     def ngtype(self, q):
         return int(q & 0xFFFF)
 
     def offset_type(self, offset, type):
         return long(long(offset) << 16 | type)
 
     def loadindexmap(self):
         """loads both the map and the index from the lazy parser"""
         if isinstance(self.index, lazyindex):
             p = self.index.p
             p.load()
 
     def inlinedata(self): return self.version & REVLOGNGINLINEDATA
     def tip(self): return self.node(len(self.index) - 1)
     def count(self): return len(self.index)
     def node(self, rev):
         return (rev < 0) and nullid or self.index[rev][-1]
     def rev(self, node):
         try:
             return self.nodemap[node]
         except KeyError:
             raise RevlogError(_('%s: no node %s') % (self.indexfile, hex(node)))
     def linkrev(self, node): return self.index[self.rev(node)][-4]
     def parents(self, node):
         if node == nullid: return (nullid, nullid)
         r = self.rev(node)
         d = self.index[r][-3:-1]
         if self.version == 0:
             return d
         return [ self.node(x) for x in d ]
     def start(self, rev):
         if rev < 0:
             return -1
         if self.version != 0:
             return self.ngoffset(self.index[rev][0])
         return self.index[rev][0]
     def end(self, rev): return self.start(rev) + self.length(rev)
 
     def length(self, rev):
         if rev < 0:
             return 0
         else:
             return self.index[rev][1]
     def base(self, rev): return (rev < 0) and rev or self.index[rev][-5]
 
     def reachable(self, rev, stop=None):
         reachable = {}
         visit = [rev]
         reachable[rev] = 1
         if stop:
             stopn = self.rev(stop)
         else:
             stopn = 0
         while visit:
             n = visit.pop(0)
             if n == stop:
                 continue
             if n == nullid:
                 continue
             for p in self.parents(n):
                 if self.rev(p) < stopn:
                     continue
                 if p not in reachable:
                     reachable[p] = 1
                     visit.append(p)
         return reachable
 
     def nodesbetween(self, roots=None, heads=None):
         """Return a tuple containing three elements. Elements 1 and 2 contain
         a final list bases and heads after all the unreachable ones have been
         pruned.  Element 0 contains a topologically sorted list of all
 
         nodes that satisfy these constraints:
         1. All nodes must be descended from a node in roots (the nodes on
            roots are considered descended from themselves).
         2. All nodes must also be ancestors of a node in heads (the nodes in
            heads are considered to be their own ancestors).
 
         If roots is unspecified, nullid is assumed as the only root.
         If heads is unspecified, it is taken to be the output of the
         heads method (i.e. a list of all nodes in the repository that
         have no children)."""
         nonodes = ([], [], [])
         if roots is not None:
             roots = list(roots)
             if not roots:
                 return nonodes
             lowestrev = min([self.rev(n) for n in roots])
         else:
             roots = [nullid] # Everybody's a descendent of nullid
             lowestrev = -1
         if (lowestrev == -1) and (heads is None):
             # We want _all_ the nodes!
             return ([self.node(r) for r in xrange(0, self.count())],
                     [nullid], list(self.heads()))
         if heads is None:
             # All nodes are ancestors, so the latest ancestor is the last
             # node.
             highestrev = self.count() - 1
             # Set ancestors to None to signal that every node is an ancestor.
             ancestors = None
             # Set heads to an empty dictionary for later discovery of heads
             heads = {}
         else:
             heads = list(heads)
             if not heads:
                 return nonodes
             ancestors = {}
             # Start at the top and keep marking parents until we're done.
             nodestotag = heads[:]
             # Turn heads into a dictionary so we can remove 'fake' heads.
             # Also, later we will be using it to filter out the heads we can't
             # find from roots.
             heads = dict.fromkeys(heads, 0)
             # Remember where the top was so we can use it as a limit later.
             highestrev = max([self.rev(n) for n in nodestotag])
             while nodestotag:
                 # grab a node to tag
                 n = nodestotag.pop()
                 # Never tag nullid
                 if n == nullid:
                     continue
                 # A node's revision number represents its place in a
                 # topologically sorted list of nodes.
                 r = self.rev(n)
                 if r >= lowestrev:
                     if n not in ancestors:
                         # If we are possibly a descendent of one of the roots
                         # and we haven't already been marked as an ancestor
                         ancestors[n] = 1 # Mark as ancestor
                         # Add non-nullid parents to list of nodes to tag.
                         nodestotag.extend([p for p in self.parents(n) if
                                            p != nullid])
                     elif n in heads: # We've seen it before, is it a fake head?
                         # So it is, real heads should not be the ancestors of
                         # any other heads.
                         heads.pop(n)
             if not ancestors:
                 return nonodes
             # Now that we have our set of ancestors, we want to remove any
             # roots that are not ancestors.
 
             # If one of the roots was nullid, everything is included anyway.
             if lowestrev > -1:
                 # But, since we weren't, let's recompute the lowest rev to not
                 # include roots that aren't ancestors.
 
                 # Filter out roots that aren't ancestors of heads
                 roots = [n for n in roots if n in ancestors]
                 # Recompute the lowest revision
                 if roots:
                     lowestrev = min([self.rev(n) for n in roots])
                 else:
                     # No more roots?  Return empty list
                     return nonodes
             else:
                 # We are descending from nullid, and don't need to care about
                 # any other roots.
                 lowestrev = -1
                 roots = [nullid]
         # Transform our roots list into a 'set' (i.e. a dictionary where the
         # values don't matter.
         descendents = dict.fromkeys(roots, 1)
         # Also, keep the original roots so we can filter out roots that aren't
         # 'real' roots (i.e. are descended from other roots).
         roots = descendents.copy()
         # Our topologically sorted list of output nodes.
         orderedout = []
         # Don't start at nullid since we don't want nullid in our output list,
         # and if nullid shows up in descedents, empty parents will look like
         # they're descendents.
         for r in xrange(max(lowestrev, 0), highestrev + 1):
             n = self.node(r)
             isdescendent = False
             if lowestrev == -1:  # Everybody is a descendent of nullid
                 isdescendent = True
             elif n in descendents:
                 # n is already a descendent
                 isdescendent = True
                 # This check only needs to be done here because all the roots
                 # will start being marked is descendents before the loop.
                 if n in roots:
                     # If n was a root, check if it's a 'real' root.
                     p = tuple(self.parents(n))
                     # If any of its parents are descendents, it's not a root.
                     if (p[0] in descendents) or (p[1] in descendents):
                         roots.pop(n)
             else:
                 p = tuple(self.parents(n))
                 # A node is a descendent if either of its parents are
                 # descendents.  (We seeded the dependents list with the roots
                 # up there, remember?)
                 if (p[0] in descendents) or (p[1] in descendents):
                     descendents[n] = 1
                     isdescendent = True
             if isdescendent and ((ancestors is None) or (n in ancestors)):
                 # Only include nodes that are both descendents and ancestors.
                 orderedout.append(n)
                 if (ancestors is not None) and (n in heads):
                     # We're trying to figure out which heads are reachable
                     # from roots.
                     # Mark this head as having been reached
                     heads[n] = 1
                 elif ancestors is None:
                     # Otherwise, we're trying to discover the heads.
                     # Assume this is a head because if it isn't, the next step
                     # will eventually remove it.
                     heads[n] = 1
                     # But, obviously its parents aren't.
                     for p in self.parents(n):
                         heads.pop(p, None)
         heads = [n for n in heads.iterkeys() if heads[n] != 0]
         roots = roots.keys()
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def heads(self, start=None):
         """return the list of all nodes that have no children
 
         if start is specified, only heads that are descendants of
         start will be returned
 
         """
         if start is None:
             start = nullid
         reachable = {start: 1}
         heads = {start: 1}
         startrev = self.rev(start)
 
         for r in xrange(startrev + 1, self.count()):
             n = self.node(r)
             for pn in self.parents(n):
                 if pn in reachable:
                     reachable[n] = 1
                     heads[n] = 1
                 if pn in heads:
                     del heads[pn]
         return heads.keys()
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in range(p + 1, self.count()):
             n = self.node(r)
             for pn in self.parents(n):
                 if pn == node:
                     c.append(n)
                     continue
                 elif pn == nullid:
                     continue
         return c
 
     def lookup(self, id):
         """locate a node based on revision number or subset of hex nodeid"""
         try:
             rev = int(id)
             if str(rev) != id: raise ValueError
             if rev < 0: rev = self.count() + rev
             if rev < 0 or rev >= self.count(): raise ValueError
             return self.node(rev)
         except (ValueError, OverflowError):
             c = []
             for n in self.nodemap:
                 if hex(n).startswith(id):
                     c.append(n)
             if len(c) > 1: raise RevlogError(_("Ambiguous identifier"))
             if len(c) < 1: raise RevlogError(_("No match found"))
             return c[0]
 
         return None
 
     def diff(self, a, b):
         """return a delta between two revisions"""
         return mdiff.textdiff(a, b)
 
     def patches(self, t, pl):
         """apply a list of patches to a string"""
         return mdiff.patches(t, pl)
 
     def chunk(self, rev, df=None, cachelen=4096):
         start, length = self.start(rev), self.length(rev)
         inline = self.inlinedata()
         if inline:
             start += (rev + 1) * struct.calcsize(self.indexformat)
         end = start + length
         def loadcache(df):
             cache_length = max(cachelen, length) # 4k
             if not df:
                 if inline:
                     df = self.opener(self.indexfile)
                 else:
                     df = self.opener(self.datafile)
             df.seek(start)
             self.chunkcache = (start, df.read(cache_length))
 
         if not self.chunkcache:
             loadcache(df)
 
         cache_start = self.chunkcache[0]
         cache_end = cache_start + len(self.chunkcache[1])
         if start >= cache_start and end <= cache_end:
             # it is cached
             offset = start - cache_start
         else:
             loadcache(df)
             offset = 0
 
         #def checkchunk():
         #    df = self.opener(self.datafile)
         #    df.seek(start)
         #    return df.read(length)
         #assert s == checkchunk()
         return decompress(self.chunkcache[1][offset:offset + length])
 
     def delta(self, node):
         """return or calculate a delta between a node and its predecessor"""
         r = self.rev(node)
         return self.revdiff(r - 1, r)
 
     def revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions"""
         b1 = self.base(rev1)
         b2 = self.base(rev2)
         if b1 == b2 and rev1 + 1 == rev2:
             return self.chunk(rev2)
         else:
             return self.diff(self.revision(self.node(rev1)),
                              self.revision(self.node(rev2)))
 
     def revision(self, node):
         """return an uncompressed revision of a given"""
         if node == nullid: return ""
         if self.cache and self.cache[0] == node: return self.cache[2]
 
         # look up what we need to read
         text = None
         rev = self.rev(node)
         base = self.base(rev)
 
         if self.inlinedata():
             # we probably have the whole chunk cached
             df = None
         else:
             df = self.opener(self.datafile)
 
         # do we have useful data cached?
         if self.cache and self.cache[1] >= base and self.cache[1] < rev:
             base = self.cache[1]
             text = self.cache[2]
         else:
             text = self.chunk(base, df=df)
 
         bins = []
         for r in xrange(base + 1, rev + 1):
             bins.append(self.chunk(r, df=df))
 
         text = self.patches(text, bins)
 
         p1, p2 = self.parents(node)
         if node != hash(text, p1, p2):
             raise RevlogError(_("integrity check failed on %s:%d")
                           % (self.datafile, rev))
 
         self.cache = (node, rev, text)
         return text
 
     def checkinlinesize(self, tr, fp=None):
         if not self.inlinedata():
             return
         if not fp:
             fp = self.opener(self.indexfile, 'r')
         size = fp.tell()
         if size < 131072:
             return
         tr.add(self.datafile, 0)
         df = self.opener(self.datafile, 'w')
         calc = struct.calcsize(self.indexformat)
         for r in xrange(self.count()):
             start = self.start(r) + (r + 1) * calc
             length = self.length(r)
             fp.seek(start)
             d = fp.read(length)
             df.write(d)
         fp.close()
         df.close()
-        fp = self.opener(self.indexfile, 'w', atomic=True)
+        fp = self.opener(self.indexfile, 'w', atomictemp=True)
         self.version &= ~(REVLOGNGINLINEDATA)
         if self.count():
             x = self.index[0]
             e = struct.pack(self.indexformat, *x)[4:]
             l = struct.pack(versionformat, self.version)
             fp.write(l)
             fp.write(e)
 
         for i in xrange(1, self.count()):
             x = self.index[i]
             e = struct.pack(self.indexformat, *x)
             fp.write(e)
 
-        fp.close()
+        # if we don't call rename, the temp file will never replace the
+        # real index
+        fp.rename()
         self.chunkcache = None
 
     def addrevision(self, text, transaction, link, p1=None, p2=None, d=None):
         """add a revision to the log
 
         text - the revision data to add
         transaction - the transaction object used for rollback
         link - the linkrev data to add
         p1, p2 - the parent nodeids of the revision
         d - an optional precomputed delta
         """
         if text is None: text = ""
         if p1 is None: p1 = self.tip()
         if p2 is None: p2 = nullid
 
         node = hash(text, p1, p2)
 
         if node in self.nodemap:
             return node
 
         n = self.count()
         t = n - 1
 
         if n:
             base = self.base(t)
             start = self.start(base)
             end = self.end(t)
             if not d:
                 prev = self.revision(self.tip())
                 d = self.diff(prev, str(text))
             data = compress(d)
             l = len(data[1]) + len(data[0])
             dist = end - start + l
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if not n or dist > len(text) * 2:
             data = compress(text)
             l = len(data[1]) + len(data[0])
             base = n
         else:
             base = self.base(t)
 
         offset = 0
         if t >= 0:
             offset = self.end(t)
 
         if self.version == 0:
             e = (offset, l, base, link, p1, p2, node)
         else:
             e = (self.offset_type(offset, 0), l, len(text),
                  base, link, self.rev(p1), self.rev(p2), node)
 
         self.index.append(e)
         self.nodemap[node] = n
         entry = struct.pack(self.indexformat, *e)
 
         if not self.inlinedata():
             transaction.add(self.datafile, offset)
             transaction.add(self.indexfile, n * len(entry))
             f = self.opener(self.datafile, "a")
             if data[0]:
                 f.write(data[0])
             f.write(data[1])
             f = self.opener(self.indexfile, "a")
         else:
             f = self.opener(self.indexfile, "a+")
             transaction.add(self.indexfile, f.tell())
 
         if len(self.index) == 1 and self.version != 0:
             l = struct.pack(versionformat, self.version)
             f.write(l)
             entry = entry[4:]
 
         f.write(entry)
 
         if self.inlinedata():
             f.write(data[0])
             f.write(data[1])
             self.checkinlinesize(transaction, f)
 
         self.cache = (node, n, text)
         return node
 
     def ancestor(self, a, b):
         """calculate the least common ancestor of nodes a and b"""
         # calculate the distance of every node from root
         dist = {nullid: 0}
         for i in xrange(self.count()):
             n = self.node(i)
             p1, p2 = self.parents(n)
             dist[n] = max(dist[p1], dist[p2]) + 1
 
         # traverse ancestors in order of decreasing distance from root
         def ancestors(node):
             # we store negative distances because heap returns smallest member
             h = [(-dist[node], node)]
             seen = {}
             while h:
                 d, n = heapq.heappop(h)
                 if n not in seen:
                     seen[n] = 1
                     yield (-d, n)
                     for p in self.parents(n):
                         heapq.heappush(h, (-dist[p], p))
 
         def generations(node):
             sg, s = None, {}
             for g,n in ancestors(node):
                 if g != sg:
                     if sg:
                         yield sg, s
                     sg, s = g, {n:1}
                 else:
                     s[n] = 1
             yield sg, s
 
         x = generations(a)
         y = generations(b)
         gx = x.next()
         gy = y.next()
 
         # increment each ancestor list until it is closer to root than
         # the other, or they match
         while 1:
             #print "ancestor gen %s %s" % (gx[0], gy[0])
             if gx[0] == gy[0]:
                 # find the intersection
                 i = [ n for n in gx[1] if n in gy[1] ]
                 if i:
                     return i[0]
                 else:
                     #print "next"
                     gy = y.next()
                     gx = x.next()
             elif gx[0] < gy[0]:
                 #print "next y"
                 gy = y.next()
             else:
                 #print "next x"
                 gx = x.next()
 
     def group(self, nodelist, lookup, infocollect=None):
         """calculate a delta group
 
         Given a list of changeset revs, return a set of deltas and
         metadata corresponding to nodes. the first delta is
         parent(nodes[0]) -> nodes[0] the receiver is guaranteed to
         have this parent as it has all history before these
         changesets. parent is parent[0]
         """
         revs = [self.rev(n) for n in nodelist]
 
         # if we don't have any revisions touched by these changesets, bail
         if not revs:
             yield changegroup.closechunk()
             return
 
         # add the parent of the first rev
         p = self.parents(self.node(revs[0]))[0]
         revs.insert(0, self.rev(p))
 
         # build deltas
         for d in xrange(0, len(revs) - 1):
             a, b = revs[d], revs[d + 1]
             nb = self.node(b)
 
             if infocollect is not None:
                 infocollect(nb)
 
             d = self.revdiff(a, b)
             p = self.parents(nb)
             meta = nb + p[0] + p[1] + lookup(nb)
             yield changegroup.genchunk("%s%s" % (meta, d))
 
         yield changegroup.closechunk()
 
     def addgroup(self, revs, linkmapper, transaction, unique=0):
         """
         add a delta group
 
         given a set of deltas, add them to the revision log. the
         first delta is against its parent, which should be in our
         log, the rest are against the previous delta.
         """
 
         #track the base of the current delta log
         r = self.count()
         t = r - 1
         node = None
 
         base = prev = -1
         start = end = measure = 0
         if r:
             end = self.end(t)
 
         ifh = self.opener(self.indexfile, "a+")
         transaction.add(self.indexfile, ifh.tell())
         if self.inlinedata():
             dfh = None
         else:
             transaction.add(self.datafile, end)
             dfh = self.opener(self.datafile, "a")
 
         # loop through our set of deltas
         chain = None
         for chunk in revs:
             node, p1, p2, cs = struct.unpack("20s20s20s20s", chunk[:80])
             link = linkmapper(cs)
             if node in self.nodemap:
                 # this can happen if two branches make the same change
                 # if unique:
                 #    raise RevlogError(_("already have %s") % hex(node[:4]))
                 chain = node
                 continue
             delta = chunk[80:]
 
             for p in (p1, p2):
                 if not p in self.nodemap:
                     raise RevlogError(_("unknown parent %s") % short(p1))
 
             if not chain:
                 # retrieve the parent revision of the delta chain
                 chain = p1
                 if not chain in self.nodemap:
                     raise RevlogError(_("unknown base %s") % short(chain[:4]))
 
             # full versions are inserted when the needed deltas become
             # comparable to the uncompressed text or when the previous
             # version is not the one we have a delta against. We use
             # the size of the previous full rev as a proxy for the
             # current size.
 
             if chain == prev:
                 tempd = compress(delta)
                 cdelta = tempd[0] + tempd[1]
 
             if chain != prev or (end - start + len(cdelta)) > measure * 2:
                 # flush our writes here so we can read it in revision
                 if dfh:
                     dfh.flush()
                 ifh.flush()
                 text = self.revision(chain)
                 text = self.patches(text, [delta])
                 chk = self.addrevision(text, transaction, link, p1, p2)
                 if chk != node:
                     raise RevlogError(_("consistency error adding group"))
                 measure = len(text)
             else:
                 if self.version == 0:
                     e = (end, len(cdelta), base, link, p1, p2, node)
                 else:
                     e = (self.offset_type(end, 0), len(cdelta), -1, base,
                          link, self.rev(p1), self.rev(p2), node)
                 self.index.append(e)
                 self.nodemap[node] = r
                 if self.inlinedata():
                     ifh.write(struct.pack(self.indexformat, *e))
                     ifh.write(cdelta)
                     self.checkinlinesize(transaction, ifh)
                     if not self.inlinedata():
                         dfh = self.opener(self.datafile, "a")
                         ifh = self.opener(self.indexfile, "a")
                 else:
                     if not dfh:
                         # addrevision switched from inline to conventional
                         # reopen the index
                         dfh = self.opener(self.datafile, "a")
                         ifh = self.opener(self.indexfile, "a")
                     dfh.write(cdelta)
                     ifh.write(struct.pack(self.indexformat, *e))
 
             t, r, chain, prev = r, r + 1, node, node
             base = self.base(t)
             start = self.start(base)
             end = self.end(t)
 
         if node is None:
             raise RevlogError(_("group to be added is empty"))
         return node
 
     def strip(self, rev, minlink):
         if self.count() == 0 or rev >= self.count():
             return
 
         if isinstance(self.index, lazyindex):
             self.loadindexmap()
 
         # When stripping away a revision, we need to make sure it
         # does not actually belong to an older changeset.
         # The minlink parameter defines the oldest revision
         # we're allowed to strip away.
         while minlink > self.index[rev][-4]:
             rev += 1
             if rev >= self.count():
                 return
 
         # first truncate the files on disk
         end = self.start(rev)
         if not self.inlinedata():
             df = self.opener(self.datafile, "a")
             df.truncate(end)
             end = rev * struct.calcsize(self.indexformat)
         else:
             end += rev * struct.calcsize(self.indexformat)
 
         indexf = self.opener(self.indexfile, "a")
         indexf.truncate(end)
 
         # then reset internal state in memory to forget those revisions
         self.cache = None
         self.chunkcache = None
         for x in xrange(rev, self.count()):
             del self.nodemap[self.node(x)]
 
         del self.index[rev:]
 
     def checksize(self):
         expected = 0
         if self.count():
             expected = self.end(self.count() - 1)
 
         try:
             f = self.opener(self.datafile)
             f.seek(0, 2)
             actual = f.tell()
             dd = actual - expected
         except IOError, inst:
             if inst.errno != errno.ENOENT:
                 raise
             dd = 0
 
         try:
             f = self.opener(self.indexfile)
             f.seek(0, 2)
             actual = f.tell()
             s = struct.calcsize(self.indexformat)
             i = actual / s
             di = actual - (i * s)
             if self.inlinedata():
                 databytes = 0
                 for r in xrange(self.count()):
                     databytes += self.length(r)
                 dd = 0
                 di = actual - self.count() * s - databytes
         except IOError, inst:
             if inst.errno != errno.ENOENT:
                 raise
             di = 0
 
         return (dd, di)
 
 

mercurial/util.py

@@ -1,817 +1,832 b''
 """
 util.py - Mercurial utility functions and platform specfic implementations
 
  Copyright 2005 K. Thananchayan <[e-mail]>
 
 This software may be used and distributed according to the terms
 of the GNU General Public License, incorporated herein by reference.
 
 This contains helper routines that are independent of the SCM core and hide
 platform-specific details from the core.
 """
 
 import os, errno
 from i18n import gettext as _
 from demandload import *
 demandload(globals(), "cStringIO errno popen2 re shutil sys tempfile")
 demandload(globals(), "threading time")
 
 def pipefilter(s, cmd):
     '''filter string S through command CMD, returning its output'''
     (pout, pin) = popen2.popen2(cmd, -1, 'b')
     def writer():
         pin.write(s)
         pin.close()
 
     # we should use select instead on UNIX, but this will work on most
     # systems, including Windows
     w = threading.Thread(target=writer)
     w.start()
     f = pout.read()
     pout.close()
     w.join()
     return f
 
 def tempfilter(s, cmd):
     '''filter string S through a pair of temporary files with CMD.
     CMD is used as a template to create the real command to be run,
     with the strings INFILE and OUTFILE replaced by the real names of
     the temporary files generated.'''
     inname, outname = None, None
     try:
         infd, inname = tempfile.mkstemp(prefix='hgfin')
         fp = os.fdopen(infd, 'wb')
         fp.write(s)
         fp.close()
         outfd, outname = tempfile.mkstemp(prefix='hgfout')
         os.close(outfd)
         cmd = cmd.replace('INFILE', inname)
         cmd = cmd.replace('OUTFILE', outname)
         code = os.system(cmd)
         if code: raise Abort(_("command '%s' failed: %s") %
                              (cmd, explain_exit(code)))
         return open(outname, 'rb').read()
     finally:
         try:
             if inname: os.unlink(inname)
         except: pass
         try:
             if outname: os.unlink(outname)
         except: pass
 
 filtertable = {
     'tempfile:': tempfilter,
     'pipe:': pipefilter,
     }
 
 def filter(s, cmd):
     "filter a string through a command that transforms its input to its output"
     for name, fn in filtertable.iteritems():
         if cmd.startswith(name):
             return fn(s, cmd[len(name):].lstrip())
     return pipefilter(s, cmd)
 
 def patch(strip, patchname, ui):
     """apply the patch <patchname> to the working directory.
     a list of patched files is returned"""
     fp = os.popen('patch -p%d < "%s"' % (strip, patchname))
     files = {}
     for line in fp:
         line = line.rstrip()
         ui.status("%s\n" % line)
         if line.startswith('patching file '):
             pf = parse_patch_output(line)
             files.setdefault(pf, 1)
     code = fp.close()
     if code:
         raise Abort(_("patch command failed: %s") % explain_exit(code)[0])
     return files.keys()
 
 def binary(s):
     """return true if a string is binary data using diff's heuristic"""
     if s and '\0' in s[:4096]:
         return True
     return False
 
 def unique(g):
     """return the uniq elements of iterable g"""
     seen = {}
     for f in g:
         if f not in seen:
             seen[f] = 1
             yield f
 
 class Abort(Exception):
     """Raised if a command needs to print an error and exit."""
 
 def always(fn): return True
 def never(fn): return False
 
 def patkind(name, dflt_pat='glob'):
     """Split a string into an optional pattern kind prefix and the
     actual pattern."""
     for prefix in 're', 'glob', 'path', 'relglob', 'relpath', 'relre':
         if name.startswith(prefix + ':'): return name.split(':', 1)
     return dflt_pat, name
 
 def globre(pat, head='^', tail='$'):
     "convert a glob pattern into a regexp"
     i, n = 0, len(pat)
     res = ''
     group = False
     def peek(): return i < n and pat[i]
     while i < n:
         c = pat[i]
         i = i+1
         if c == '*':
             if peek() == '*':
                 i += 1
                 res += '.*'
             else:
                 res += '[^/]*'
         elif c == '?':
             res += '.'
         elif c == '[':
             j = i
             if j < n and pat[j] in '!]':
                 j += 1
             while j < n and pat[j] != ']':
                 j += 1
             if j >= n:
                 res += '\\['
             else:
                 stuff = pat[i:j].replace('\\','\\\\')
                 i = j + 1
                 if stuff[0] == '!':
                     stuff = '^' + stuff[1:]
                 elif stuff[0] == '^':
                     stuff = '\\' + stuff
                 res = '%s[%s]' % (res, stuff)
         elif c == '{':
             group = True
             res += '(?:'
         elif c == '}' and group:
             res += ')'
             group = False
         elif c == ',' and group:
             res += '|'
         elif c == '\\':
             p = peek()
             if p:
                 i += 1
                 res += re.escape(p)
             else:
                 res += re.escape(c)
         else:
             res += re.escape(c)
     return head + res + tail
 
 _globchars = {'[': 1, '{': 1, '*': 1, '?': 1}
 
 def pathto(n1, n2):
     '''return the relative path from one place to another.
     this returns a path in the form used by the local filesystem, not hg.'''
     if not n1: return localpath(n2)
     a, b = n1.split('/'), n2.split('/')
     a.reverse()
     b.reverse()
     while a and b and a[-1] == b[-1]:
         a.pop()
         b.pop()
     b.reverse()
     return os.sep.join((['..'] * len(a)) + b)
 
 def canonpath(root, cwd, myname):
     """return the canonical path of myname, given cwd and root"""
     if root == os.sep:
         rootsep = os.sep
     else:
         rootsep = root + os.sep
     name = myname
     if not name.startswith(os.sep):
         name = os.path.join(root, cwd, name)
     name = os.path.normpath(name)
     if name.startswith(rootsep):
         name = name[len(rootsep):]
         audit_path(name)
         return pconvert(name)
     elif name == root:
         return ''
     else:
         raise Abort('%s not under root' % myname)
 
 def matcher(canonroot, cwd='', names=['.'], inc=[], exc=[], head='', src=None):
     return _matcher(canonroot, cwd, names, inc, exc, head, 'glob', src)
 
 def cmdmatcher(canonroot, cwd='', names=['.'], inc=[], exc=[], head='', src=None):
     if os.name == 'nt':
         dflt_pat = 'glob'
     else:
         dflt_pat = 'relpath'
     return _matcher(canonroot, cwd, names, inc, exc, head, dflt_pat, src)
 
 def _matcher(canonroot, cwd, names, inc, exc, head, dflt_pat, src):
     """build a function to match a set of file patterns
 
     arguments:
     canonroot - the canonical root of the tree you're matching against
     cwd - the current working directory, if relevant
     names - patterns to find
     inc - patterns to include
     exc - patterns to exclude
     head - a regex to prepend to patterns to control whether a match is rooted
 
     a pattern is one of:
     'glob:<rooted glob>'
     're:<rooted regexp>'
     'path:<rooted path>'
     'relglob:<relative glob>'
     'relpath:<relative path>'
     'relre:<relative regexp>'
     '<rooted path or regexp>'
 
     returns:
     a 3-tuple containing
     - list of explicit non-pattern names passed in
     - a bool match(filename) function
     - a bool indicating if any patterns were passed in
 
     todo:
     make head regex a rooted bool
     """
 
     def contains_glob(name):
         for c in name:
             if c in _globchars: return True
         return False
 
     def regex(kind, name, tail):
         '''convert a pattern into a regular expression'''
         if kind == 're':
             return name
         elif kind == 'path':
             return '^' + re.escape(name) + '(?:/|$)'
         elif kind == 'relglob':
             return head + globre(name, '(?:|.*/)', tail)
         elif kind == 'relpath':
             return head + re.escape(name) + tail
         elif kind == 'relre':
             if name.startswith('^'):
                 return name
             return '.*' + name
         return head + globre(name, '', tail)
 
     def matchfn(pats, tail):
         """build a matching function from a set of patterns"""
         if not pats:
             return
         matches = []
         for k, p in pats:
             try:
                 pat = '(?:%s)' % regex(k, p, tail)
                 matches.append(re.compile(pat).match)
             except re.error:
                 if src: raise Abort("%s: invalid pattern (%s): %s" % (src, k, p))
                 else: raise Abort("invalid pattern (%s): %s" % (k, p))
 
         def buildfn(text):
             for m in matches:
                 r = m(text)
                 if r:
                     return r
 
         return buildfn
 
     def globprefix(pat):
         '''return the non-glob prefix of a path, e.g. foo/* -> foo'''
         root = []
         for p in pat.split(os.sep):
             if contains_glob(p): break
             root.append(p)
         return '/'.join(root)
 
     pats = []
     files = []
     roots = []
     for kind, name in [patkind(p, dflt_pat) for p in names]:
         if kind in ('glob', 'relpath'):
             name = canonpath(canonroot, cwd, name)
             if name == '':
                 kind, name = 'glob', '**'
         if kind in ('glob', 'path', 're'):
             pats.append((kind, name))
         if kind == 'glob':
             root = globprefix(name)
             if root: roots.append(root)
         elif kind == 'relpath':
             files.append((kind, name))
             roots.append(name)
 
     patmatch = matchfn(pats, '$') or always
     filematch = matchfn(files, '(?:/|$)') or always
     incmatch = always
     if inc:
         incmatch = matchfn(map(patkind, inc), '(?:/|$)')
     excmatch = lambda fn: False
     if exc:
         excmatch = matchfn(map(patkind, exc), '(?:/|$)')
 
     return (roots,
             lambda fn: (incmatch(fn) and not excmatch(fn) and
                         (fn.endswith('/') or
                          (not pats and not files) or
                          (pats and patmatch(fn)) or
                          (files and filematch(fn)))),
             (inc or exc or (pats and pats != [('glob', '**')])) and True)
 
 def system(cmd, environ={}, cwd=None, onerr=None, errprefix=None):
     '''enhanced shell command execution.
     run with environment maybe modified, maybe in different dir.
 
     if command fails and onerr is None, return status.  if ui object,
     print error message and return status, else raise onerr object as
     exception.'''
     oldenv = {}
     for k in environ:
         oldenv[k] = os.environ.get(k)
     if cwd is not None:
         oldcwd = os.getcwd()
     try:
         for k, v in environ.iteritems():
             os.environ[k] = str(v)
         if cwd is not None and oldcwd != cwd:
             os.chdir(cwd)
         rc = os.system(cmd)
         if rc and onerr:
             errmsg = '%s %s' % (os.path.basename(cmd.split(None, 1)[0]),
                                 explain_exit(rc)[0])
             if errprefix:
                 errmsg = '%s: %s' % (errprefix, errmsg)
             try:
                 onerr.warn(errmsg + '\n')
             except AttributeError:
                 raise onerr(errmsg)
         return rc
     finally:
         for k, v in oldenv.iteritems():
             if v is None:
                 del os.environ[k]
             else:
                 os.environ[k] = v
         if cwd is not None and oldcwd != cwd:
             os.chdir(oldcwd)
 
 def rename(src, dst):
     """forcibly rename a file"""
     try:
         os.rename(src, dst)
     except:
         os.unlink(dst)
         os.rename(src, dst)
 
 def unlink(f):
     """unlink and remove the directory if it is empty"""
     os.unlink(f)
     # try removing directories that might now be empty
     try: os.removedirs(os.path.dirname(f))
     except: pass
 
 def copyfiles(src, dst, hardlink=None):
     """Copy a directory tree using hardlinks if possible"""
 
     if hardlink is None:
         hardlink = (os.stat(src).st_dev ==
                     os.stat(os.path.dirname(dst)).st_dev)
 
     if os.path.isdir(src):
         os.mkdir(dst)
         for name in os.listdir(src):
             srcname = os.path.join(src, name)
             dstname = os.path.join(dst, name)
             copyfiles(srcname, dstname, hardlink)
     else:
         if hardlink:
             try:
                 os_link(src, dst)
             except:
                 hardlink = False
                 shutil.copy(src, dst)
         else:
             shutil.copy(src, dst)
 
 def audit_path(path):
     """Abort if path contains dangerous components"""
     parts = os.path.normcase(path).split(os.sep)
     if (os.path.splitdrive(path)[0] or parts[0] in ('.hg', '')
         or os.pardir in parts):
         raise Abort(_("path contains illegal component: %s\n") % path)
 
 def opener(base, audit=True):
     """
     return a function that opens files relative to base
 
     this function is used to hide the details of COW semantics and
     remote file access from higher level code.
     """
     p = base
     audit_p = audit
 
     def mktempcopy(name):
         d, fn = os.path.split(name)
         fd, temp = tempfile.mkstemp(prefix=fn, dir=d)
         fp = os.fdopen(fd, "wb")
         try:
             fp.write(file(name, "rb").read())
         except:
             try: os.unlink(temp)
             except: pass
             raise
         fp.close()
         st = os.lstat(name)
         os.chmod(temp, st.st_mode)
         return temp
 
-    class atomicfile(file):
-        """the file will only be copied on close"""
-        def __init__(self, name, mode, atomic=False):
+    class atomictempfile(file):
+        """the file will only be copied when rename is called"""
+        def __init__(self, name, mode):
             self.__name = name
             self.temp = mktempcopy(name)
             file.__init__(self, self.temp, mode)
-        def close(self):
+        def rename(self):
             if not self.closed:
                 file.close(self)
                 rename(self.temp, self.__name)
         def __del__(self):
-            self.close()
+            if not self.closed:
+                try:
+                    os.unlink(self.temp)
+                except: pass
+                file.close(self)
 
-    def o(path, mode="r", text=False, atomic=False):
+    class atomicfile(atomictempfile):
+        """the file will only be copied on close"""
+        def __init__(self, name, mode):
+            atomictempfile.__init__(self, name, mode)
+        def close(self):
+            self.rename()
+        def __del__(self):
+            self.rename()
+
+    def o(path, mode="r", text=False, atomic=False, atomictemp=False):
         if audit_p:
             audit_path(path)
         f = os.path.join(p, path)
 
         if not text:
             mode += "b" # for that other OS
 
         if mode[0] != "r":
             try:
                 nlink = nlinks(f)
             except OSError:
                 d = os.path.dirname(f)
                 if not os.path.isdir(d):
                     os.makedirs(d)
             else:
                 if atomic:
                     return atomicfile(f, mode)
+                elif atomictemp:
+                    return atomictempfile(f, mode)
                 if nlink > 1:
                     rename(mktempcopy(f), f)
         return file(f, mode)
 
     return o
 
 def _makelock_file(info, pathname):
     ld = os.open(pathname, os.O_CREAT | os.O_WRONLY | os.O_EXCL)
     os.write(ld, info)
     os.close(ld)
 
 def _readlock_file(pathname):
     return file(pathname).read()
 
 def nlinks(pathname):
     """Return number of hardlinks for the given file."""
     return os.stat(pathname).st_nlink
 
 if hasattr(os, 'link'):
     os_link = os.link
 else:
     def os_link(src, dst):
         raise OSError(0, _("Hardlinks not supported"))
 
 # Platform specific variants
 if os.name == 'nt':
     demandload(globals(), "msvcrt")
     nulldev = 'NUL:'
 
     class winstdout:
         '''stdout on windows misbehaves if sent through a pipe'''
 
         def __init__(self, fp):
             self.fp = fp
 
         def __getattr__(self, key):
             return getattr(self.fp, key)
 
         def close(self):
             try:
                 self.fp.close()
             except: pass
 
         def write(self, s):
             try:
                 return self.fp.write(s)
             except IOError, inst:
                 if inst.errno != 0: raise
                 self.close()
                 raise IOError(errno.EPIPE, 'Broken pipe')
 
     sys.stdout = winstdout(sys.stdout)
 
     def os_rcpath():
         '''return default os-specific hgrc search path'''
         try:
             import win32api, win32process
             proc = win32api.GetCurrentProcess()
             filename = win32process.GetModuleFileNameEx(proc, 0)
             systemrc = os.path.join(os.path.dirname(filename), 'mercurial.ini')
         except ImportError:
             systemrc = r'c:\mercurial\mercurial.ini'
 
         return [systemrc,
                 os.path.join(os.path.expanduser('~'), 'mercurial.ini')]
 
     def parse_patch_output(output_line):
         """parses the output produced by patch and returns the file name"""
         pf = output_line[14:]
         if pf[0] == '`':
             pf = pf[1:-1] # Remove the quotes
         return pf
 
     try: # Mark Hammond's win32all package allows better functionality on Windows
         import win32api, win32con, win32file, pywintypes
 
         # create hard links using win32file module
         def os_link(src, dst): # NB will only succeed on NTFS
             win32file.CreateHardLink(dst, src)
 
         def nlinks(pathname):
             """Return number of hardlinks for the given file."""
             try:
                 fh = win32file.CreateFile(pathname,
                     win32file.GENERIC_READ, win32file.FILE_SHARE_READ,
                     None, win32file.OPEN_EXISTING, 0, None)
                 res = win32file.GetFileInformationByHandle(fh)
                 fh.Close()
                 return res[7]
             except:
                 return os.stat(pathname).st_nlink
 
         def testpid(pid):
             '''return True if pid is still running or unable to
             determine, False otherwise'''
             try:
                 import win32process, winerror
                 handle = win32api.OpenProcess(
                     win32con.PROCESS_QUERY_INFORMATION, False, pid)
                 if handle:
                     status = win32process.GetExitCodeProcess(handle)
                     return status == win32con.STILL_ACTIVE
             except pywintypes.error, details:
                 return details[0] != winerror.ERROR_INVALID_PARAMETER
             return True
 
     except ImportError:
         def testpid(pid):
             '''return False if pid dead, True if running or not known'''
             return True
 
     def is_exec(f, last):
         return last
 
     def set_exec(f, mode):
         pass
 
     def set_binary(fd):
         msvcrt.setmode(fd.fileno(), os.O_BINARY)
 
     def pconvert(path):
         return path.replace("\\", "/")
 
     def localpath(path):
         return path.replace('/', '\\')
 
     def normpath(path):
         return pconvert(os.path.normpath(path))
 
     makelock = _makelock_file
     readlock = _readlock_file
 
     def explain_exit(code):
         return _("exited with status %d") % code, code
 
 else:
     nulldev = '/dev/null'
 
     def rcfiles(path):
         rcs = [os.path.join(path, 'hgrc')]
         rcdir = os.path.join(path, 'hgrc.d')
         try:
             rcs.extend([os.path.join(rcdir, f) for f in os.listdir(rcdir)
                         if f.endswith(".rc")])
         except OSError, inst: pass
         return rcs
 
     def os_rcpath():
         '''return default os-specific hgrc search path'''
         path = []
         if len(sys.argv) > 0:
             path.extend(rcfiles(os.path.dirname(sys.argv[0]) +
                                   '/../etc/mercurial'))
         path.extend(rcfiles('/etc/mercurial'))
         path.append(os.path.expanduser('~/.hgrc'))
         path = [os.path.normpath(f) for f in path]
         return path
 
     def parse_patch_output(output_line):
         """parses the output produced by patch and returns the file name"""
         pf = output_line[14:]
         if pf.startswith("'") and pf.endswith("'") and pf.find(" ") >= 0:
             pf = pf[1:-1] # Remove the quotes
         return pf
 
     def is_exec(f, last):
         """check whether a file is executable"""
         return (os.stat(f).st_mode & 0100 != 0)
 
     def set_exec(f, mode):
         s = os.stat(f).st_mode
         if (s & 0100 != 0) == mode:
             return
         if mode:
             # Turn on +x for every +r bit when making a file executable
             # and obey umask.
             umask = os.umask(0)
             os.umask(umask)
             os.chmod(f, s | (s & 0444) >> 2 & ~umask)
         else:
             os.chmod(f, s & 0666)
 
     def set_binary(fd):
         pass
 
     def pconvert(path):
         return path
 
     def localpath(path):
         return path
 
     normpath = os.path.normpath
 
     def makelock(info, pathname):
         try:
             os.symlink(info, pathname)
         except OSError, why:
             if why.errno == errno.EEXIST:
                 raise
             else:
                 _makelock_file(info, pathname)
 
     def readlock(pathname):
         try:
             return os.readlink(pathname)
         except OSError, why:
             if why.errno == errno.EINVAL:
                 return _readlock_file(pathname)
             else:
                 raise
 
     def testpid(pid):
         '''return False if pid dead, True if running or not sure'''
         try:
             os.kill(pid, 0)
             return True
         except OSError, inst:
             return inst.errno != errno.ESRCH
 
     def explain_exit(code):
         """return a 2-tuple (desc, code) describing a process's status"""
         if os.WIFEXITED(code):
             val = os.WEXITSTATUS(code)
             return _("exited with status %d") % val, val
         elif os.WIFSIGNALED(code):
             val = os.WTERMSIG(code)
             return _("killed by signal %d") % val, val
         elif os.WIFSTOPPED(code):
             val = os.WSTOPSIG(code)
             return _("stopped by signal %d") % val, val
         raise ValueError(_("invalid exit code"))
 
 class chunkbuffer(object):
     """Allow arbitrary sized chunks of data to be efficiently read from an
     iterator over chunks of arbitrary size."""
 
     def __init__(self, in_iter, targetsize = 2**16):
         """in_iter is the iterator that's iterating over the input chunks.
         targetsize is how big a buffer to try to maintain."""
         self.in_iter = iter(in_iter)
         self.buf = ''
         self.targetsize = int(targetsize)
         if self.targetsize <= 0:
             raise ValueError(_("targetsize must be greater than 0, was %d") %
                              targetsize)
         self.iterempty = False
 
     def fillbuf(self):
         """Ignore target size; read every chunk from iterator until empty."""
         if not self.iterempty:
             collector = cStringIO.StringIO()
             collector.write(self.buf)
             for ch in self.in_iter:
                 collector.write(ch)
             self.buf = collector.getvalue()
             self.iterempty = True
 
     def read(self, l):
         """Read L bytes of data from the iterator of chunks of data.
         Returns less than L bytes if the iterator runs dry."""
         if l > len(self.buf) and not self.iterempty:
             # Clamp to a multiple of self.targetsize
             targetsize = self.targetsize * ((l // self.targetsize) + 1)
             collector = cStringIO.StringIO()
             collector.write(self.buf)
             collected = len(self.buf)
             for chunk in self.in_iter:
                 collector.write(chunk)
                 collected += len(chunk)
                 if collected >= targetsize:
                     break
             if collected < targetsize:
                 self.iterempty = True
             self.buf = collector.getvalue()
         s, self.buf = self.buf[:l], buffer(self.buf, l)
         return s
 
 def filechunkiter(f, size = 65536):
     """Create a generator that produces all the data in the file size
     (default 65536) bytes at a time.  Chunks may be less than size
     bytes if the chunk is the last chunk in the file, or the file is a
     socket or some other type of file that sometimes reads less data
     than is requested."""
     s = f.read(size)
     while len(s) > 0:
         yield s
         s = f.read(size)
 
 def makedate():
     lt = time.localtime()
     if lt[8] == 1 and time.daylight:
         tz = time.altzone
     else:
         tz = time.timezone
     return time.mktime(lt), tz
 
 def datestr(date=None, format='%a %b %d %H:%M:%S %Y', timezone=True):
     """represent a (unixtime, offset) tuple as a localized time.
     unixtime is seconds since the epoch, and offset is the time zone's
     number of seconds away from UTC. if timezone is false, do not
     append time zone to string."""
     t, tz = date or makedate()
     s = time.strftime(format, time.gmtime(float(t) - tz))
     if timezone:
         s += " %+03d%02d" % (-tz / 3600, ((-tz % 3600) / 60))
     return s
 
 def shortuser(user):
     """Return a short representation of a user name or email address."""
     f = user.find('@')
     if f >= 0:
         user = user[:f]
     f = user.find('<')
     if f >= 0:
         user = user[f+1:]
     return user
 
 def walkrepos(path):
     '''yield every hg repository under path, recursively.'''
     def errhandler(err):
         if err.filename == path:
             raise err
 
     for root, dirs, files in os.walk(path, onerror=errhandler):
         for d in dirs:
             if d == '.hg':
                 yield root
                 dirs[:] = []
                 break
 
 _rcpath = None
 
 def rcpath():
     '''return hgrc search path. if env var HGRCPATH is set, use it.
     for each item in path, if directory, use files ending in .rc,
     else use item.
     make HGRCPATH empty to only look in .hg/hgrc of current repo.
     if no HGRCPATH, use default os-specific path.'''
     global _rcpath
     if _rcpath is None:
         if 'HGRCPATH' in os.environ:
             _rcpath = []
             for p in os.environ['HGRCPATH'].split(os.pathsep):
                 if not p: continue
                 if os.path.isdir(p):
                     for f in os.listdir(p):
                         if f.endswith('.rc'):
                             _rcpath.append(os.path.join(p, f))
                 else:
                     _rcpath.append(p)
         else:
             _rcpath = os_rcpath()
     return _rcpath