merge with stable

mercurial/hgweb/hgweb_mod.py

@@ -1,391 +1,393 b''
 # hgweb/hgweb_mod.py - Web interface for a repository.
 #
 # Copyright 21 May 2005 - (c) 2005 Jake Edge <[e-mail]>
 # Copyright 2005-2007 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.
 
 import os, re
 from mercurial import ui, hg, hook, error, encoding, templater, util, repoview
 from mercurial.templatefilters import websub
 from mercurial.i18n import _
 from common import get_stat, ErrorResponse, permhooks, caching
 from common import HTTP_OK, HTTP_NOT_MODIFIED, HTTP_BAD_REQUEST
 from common import HTTP_NOT_FOUND, HTTP_SERVER_ERROR
 from request import wsgirequest
 import webcommands, protocol, webutil
 
 perms = {
     'changegroup': 'pull',
     'changegroupsubset': 'pull',
     'getbundle': 'pull',
     'stream_out': 'pull',
     'listkeys': 'pull',
     'unbundle': 'push',
     'pushkey': 'push',
 }
 
 def makebreadcrumb(url, prefix=''):
     '''Return a 'URL breadcrumb' list
 
     A 'URL breadcrumb' is a list of URL-name pairs,
     corresponding to each of the path items on a URL.
     This can be used to create path navigation entries.
     '''
     if url.endswith('/'):
         url = url[:-1]
     if prefix:
         url = '/' + prefix + url
     relpath = url
     if relpath.startswith('/'):
         relpath = relpath[1:]
 
     breadcrumb = []
     urlel = url
     pathitems = [''] + relpath.split('/')
     for pathel in reversed(pathitems):
         if not pathel or not urlel:
             break
         breadcrumb.append({'url': urlel, 'name': pathel})
         urlel = os.path.dirname(urlel)
     return reversed(breadcrumb)
 
 
 class hgweb(object):
     def __init__(self, repo, name=None, baseui=None):
         if isinstance(repo, str):
             if baseui:
                 u = baseui.copy()
             else:
                 u = ui.ui()
-            self.repo = hg.repository(u, repo)
+            r = hg.repository(u, repo)
         else:
-            self.repo = repo
+            r = repo
 
-        self.repo = self._getview(self.repo)
-        self.repo.ui.setconfig('ui', 'report_untrusted', 'off')
-        self.repo.baseui.setconfig('ui', 'report_untrusted', 'off')
-        self.repo.ui.setconfig('ui', 'nontty', 'true')
-        self.repo.baseui.setconfig('ui', 'nontty', 'true')
+        r = self._getview(r)
+        r.ui.setconfig('ui', 'report_untrusted', 'off')
+        r.baseui.setconfig('ui', 'report_untrusted', 'off')
+        r.ui.setconfig('ui', 'nontty', 'true')
+        r.baseui.setconfig('ui', 'nontty', 'true')
+        self.repo = r
         hook.redirect(True)
         self.mtime = -1
         self.size = -1
         self.reponame = name
         self.archives = 'zip', 'gz', 'bz2'
         self.stripecount = 1
         # a repo owner may set web.templates in .hg/hgrc to get any file
         # readable by the user running the CGI script
         self.templatepath = self.config('web', 'templates')
         self.websubtable = self.loadwebsub()
 
     # The CGI scripts are often run by a user different from the repo owner.
     # Trust the settings from the .hg/hgrc files by default.
     def config(self, section, name, default=None, untrusted=True):
         return self.repo.ui.config(section, name, default,
                                    untrusted=untrusted)
 
     def configbool(self, section, name, default=False, untrusted=True):
         return self.repo.ui.configbool(section, name, default,
                                        untrusted=untrusted)
 
     def configlist(self, section, name, default=None, untrusted=True):
         return self.repo.ui.configlist(section, name, default,
                                        untrusted=untrusted)
 
     def _getview(self, repo):
-        viewconfig = self.config('web', 'view', 'served')
+        viewconfig = repo.ui.config('web', 'view', 'served',
+                                    untrusted=True)
         if viewconfig == 'all':
             return repo.unfiltered()
         elif viewconfig in repoview.filtertable:
             return repo.filtered(viewconfig)
         else:
             return repo.filtered('served')
 
     def refresh(self, request=None):
         st = get_stat(self.repo.spath)
         # compare changelog size in addition to mtime to catch
         # rollbacks made less than a second ago
         if st.st_mtime != self.mtime or st.st_size != self.size:
             self.mtime = st.st_mtime
             self.size = st.st_size
             r = hg.repository(self.repo.baseui, self.repo.root)
             self.repo = self._getview(r)
             self.maxchanges = int(self.config("web", "maxchanges", 10))
             self.stripecount = int(self.config("web", "stripes", 1))
             self.maxshortchanges = int(self.config("web", "maxshortchanges",
                                                    60))
             self.maxfiles = int(self.config("web", "maxfiles", 10))
             self.allowpull = self.configbool("web", "allowpull", True)
             encoding.encoding = self.config("web", "encoding",
                                             encoding.encoding)
         if request:
             self.repo.ui.environ = request.env
 
     def run(self):
         if not os.environ.get('GATEWAY_INTERFACE', '').startswith("CGI/1."):
             raise RuntimeError("This function is only intended to be "
                                "called while running as a CGI script.")
         import mercurial.hgweb.wsgicgi as wsgicgi
         wsgicgi.launch(self)
 
     def __call__(self, env, respond):
         req = wsgirequest(env, respond)
         return self.run_wsgi(req)
 
     def run_wsgi(self, req):
 
         self.refresh(req)
 
         # work with CGI variables to create coherent structure
         # use SCRIPT_NAME, PATH_INFO and QUERY_STRING as well as our REPO_NAME
 
         req.url = req.env['SCRIPT_NAME']
         if not req.url.endswith('/'):
             req.url += '/'
         if 'REPO_NAME' in req.env:
             req.url += req.env['REPO_NAME'] + '/'
 
         if 'PATH_INFO' in req.env:
             parts = req.env['PATH_INFO'].strip('/').split('/')
             repo_parts = req.env.get('REPO_NAME', '').split('/')
             if parts[:len(repo_parts)] == repo_parts:
                 parts = parts[len(repo_parts):]
             query = '/'.join(parts)
         else:
             query = req.env['QUERY_STRING'].split('&', 1)[0]
             query = query.split(';', 1)[0]
 
         # process this if it's a protocol request
         # protocol bits don't need to create any URLs
         # and the clients always use the old URL structure
 
         cmd = req.form.get('cmd', [''])[0]
         if protocol.iscmd(cmd):
             try:
                 if query:
                     raise ErrorResponse(HTTP_NOT_FOUND)
                 if cmd in perms:
                     self.check_perm(req, perms[cmd])
                 return protocol.call(self.repo, req, cmd)
             except ErrorResponse, inst:
                 # A client that sends unbundle without 100-continue will
                 # break if we respond early.
                 if (cmd == 'unbundle' and
                     (req.env.get('HTTP_EXPECT',
                                  '').lower() != '100-continue') or
                     req.env.get('X-HgHttp2', '')):
                     req.drain()
                 else:
                     req.headers.append(('Connection', 'Close'))
                 req.respond(inst, protocol.HGTYPE,
                             body='0\n%s\n' % inst.message)
                 return ''
 
         # translate user-visible url structure to internal structure
 
         args = query.split('/', 2)
         if 'cmd' not in req.form and args and args[0]:
 
             cmd = args.pop(0)
             style = cmd.rfind('-')
             if style != -1:
                 req.form['style'] = [cmd[:style]]
                 cmd = cmd[style + 1:]
 
             # avoid accepting e.g. style parameter as command
             if util.safehasattr(webcommands, cmd):
                 req.form['cmd'] = [cmd]
             else:
                 cmd = ''
 
             if cmd == 'static':
                 req.form['file'] = ['/'.join(args)]
             else:
                 if args and args[0]:
                     node = args.pop(0)
                     req.form['node'] = [node]
                 if args:
                     req.form['file'] = args
 
             ua = req.env.get('HTTP_USER_AGENT', '')
             if cmd == 'rev' and 'mercurial' in ua:
                 req.form['style'] = ['raw']
 
             if cmd == 'archive':
                 fn = req.form['node'][0]
                 for type_, spec in self.archive_specs.iteritems():
                     ext = spec[2]
                     if fn.endswith(ext):
                         req.form['node'] = [fn[:-len(ext)]]
                         req.form['type'] = [type_]
 
         # process the web interface request
 
         try:
             tmpl = self.templater(req)
             ctype = tmpl('mimetype', encoding=encoding.encoding)
             ctype = templater.stringify(ctype)
 
             # check read permissions non-static content
             if cmd != 'static':
                 self.check_perm(req, None)
 
             if cmd == '':
                 req.form['cmd'] = [tmpl.cache['default']]
                 cmd = req.form['cmd'][0]
 
             if self.configbool('web', 'cache', True):
                 caching(self, req) # sets ETag header or raises NOT_MODIFIED
             if cmd not in webcommands.__all__:
                 msg = 'no such method: %s' % cmd
                 raise ErrorResponse(HTTP_BAD_REQUEST, msg)
             elif cmd == 'file' and 'raw' in req.form.get('style', []):
                 self.ctype = ctype
                 content = webcommands.rawfile(self, req, tmpl)
             else:
                 content = getattr(webcommands, cmd)(self, req, tmpl)
                 req.respond(HTTP_OK, ctype)
 
             return content
 
         except (error.LookupError, error.RepoLookupError), err:
             req.respond(HTTP_NOT_FOUND, ctype)
             msg = str(err)
             if (util.safehasattr(err, 'name') and
                 not isinstance(err,  error.ManifestLookupError)):
                 msg = 'revision not found: %s' % err.name
             return tmpl('error', error=msg)
         except (error.RepoError, error.RevlogError), inst:
             req.respond(HTTP_SERVER_ERROR, ctype)
             return tmpl('error', error=str(inst))
         except ErrorResponse, inst:
             req.respond(inst, ctype)
             if inst.code == HTTP_NOT_MODIFIED:
                 # Not allowed to return a body on a 304
                 return ['']
             return tmpl('error', error=inst.message)
 
     def loadwebsub(self):
         websubtable = []
         websubdefs = self.repo.ui.configitems('websub')
         # we must maintain interhg backwards compatibility
         websubdefs += self.repo.ui.configitems('interhg')
         for key, pattern in websubdefs:
             # grab the delimiter from the character after the "s"
             unesc = pattern[1]
             delim = re.escape(unesc)
 
             # identify portions of the pattern, taking care to avoid escaped
             # delimiters. the replace format and flags are optional, but
             # delimiters are required.
             match = re.match(
                 r'^s%s(.+)(?:(?<=\\\\)|(?<!\\))%s(.*)%s([ilmsux])*$'
                 % (delim, delim, delim), pattern)
             if not match:
                 self.repo.ui.warn(_("websub: invalid pattern for %s: %s\n")
                                   % (key, pattern))
                 continue
 
             # we need to unescape the delimiter for regexp and format
             delim_re = re.compile(r'(?<!\\)\\%s' % delim)
             regexp = delim_re.sub(unesc, match.group(1))
             format = delim_re.sub(unesc, match.group(2))
 
             # the pattern allows for 6 regexp flags, so set them if necessary
             flagin = match.group(3)
             flags = 0
             if flagin:
                 for flag in flagin.upper():
                     flags |= re.__dict__[flag]
 
             try:
                 regexp = re.compile(regexp, flags)
                 websubtable.append((regexp, format))
             except re.error:
                 self.repo.ui.warn(_("websub: invalid regexp for %s: %s\n")
                                   % (key, regexp))
         return websubtable
 
     def templater(self, req):
 
         # determine scheme, port and server name
         # this is needed to create absolute urls
 
         proto = req.env.get('wsgi.url_scheme')
         if proto == 'https':
             proto = 'https'
             default_port = "443"
         else:
             proto = 'http'
             default_port = "80"
 
         port = req.env["SERVER_PORT"]
         port = port != default_port and (":" + port) or ""
         urlbase = '%s://%s%s' % (proto, req.env['SERVER_NAME'], port)
         logourl = self.config("web", "logourl", "http://mercurial.selenic.com/")
         logoimg = self.config("web", "logoimg", "hglogo.png")
         staticurl = self.config("web", "staticurl") or req.url + 'static/'
         if not staticurl.endswith('/'):
             staticurl += '/'
 
         # some functions for the templater
 
         def motd(**map):
             yield self.config("web", "motd", "")
 
         # figure out which style to use
 
         vars = {}
         styles = (
             req.form.get('style', [None])[0],
             self.config('web', 'style'),
             'paper',
         )
         style, mapfile = templater.stylemap(styles, self.templatepath)
         if style == styles[0]:
             vars['style'] = style
 
         start = req.url[-1] == '?' and '&' or '?'
         sessionvars = webutil.sessionvars(vars, start)
 
         if not self.reponame:
             self.reponame = (self.config("web", "name")
                              or req.env.get('REPO_NAME')
                              or req.url.strip('/') or self.repo.root)
 
         def websubfilter(text):
             return websub(text, self.websubtable)
 
         # create the templater
 
         tmpl = templater.templater(mapfile,
                                    filters={"websub": websubfilter},
                                    defaults={"url": req.url,
                                              "logourl": logourl,
                                              "logoimg": logoimg,
                                              "staticurl": staticurl,
                                              "urlbase": urlbase,
                                              "repo": self.reponame,
                                              "encoding": encoding.encoding,
                                              "motd": motd,
                                              "sessionvars": sessionvars,
                                              "pathdef": makebreadcrumb(req.url),
                                             })
         return tmpl
 
     def archivelist(self, nodeid):
         allowed = self.configlist("web", "allow_archive")
         for i, spec in self.archive_specs.iteritems():
             if i in allowed or self.configbool("web", "allow" + i):
                 yield {"type" : i, "extension" : spec[2], "node" : nodeid}
 
     archive_specs = {
         'bz2': ('application/x-bzip2', 'tbz2', '.tar.bz2', None),
         'gz': ('application/x-gzip', 'tgz', '.tar.gz', None),
         'zip': ('application/zip', 'zip', '.zip', None),
         }
 
     def check_perm(self, req, op):
         for hook in permhooks:
             hook(self, req, op)

mercurial/mpatch.c

@@ -1,429 +1,420 b''
 /*
  mpatch.c - efficient binary patching for Mercurial
 
  This implements a patch algorithm that's O(m + nlog n) where m is the
  size of the output and n is the number of patches.
 
  Given a list of binary patches, it unpacks each into a hunk list,
  then combines the hunk lists with a treewise recursion to form a
  single hunk list. This hunk list is then applied to the original
  text.
 
  The text (or binary) fragments are copied directly from their source
  Python objects into a preallocated output string to avoid the
  allocation of intermediate Python objects. Working memory is about 2x
  the total number of hunks.
 
  Copyright 2005, 2006 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.
 */
 
 #define PY_SSIZE_T_CLEAN
 #include <Python.h>
 #include <stdlib.h>
 #include <string.h>
 
 #include "util.h"
 
 static char mpatch_doc[] = "Efficient binary patching.";
 static PyObject *mpatch_Error;
 
 struct frag {
 	int start, end, len;
 	const char *data;
 };
 
 struct flist {
 	struct frag *base, *head, *tail;
 };
 
 static struct flist *lalloc(Py_ssize_t size)
 {
 	struct flist *a = NULL;
 
 	if (size < 1)
 		size = 1;
 
 	a = (struct flist *)malloc(sizeof(struct flist));
 	if (a) {
 		a->base = (struct frag *)malloc(sizeof(struct frag) * size);
 		if (a->base) {
 			a->head = a->tail = a->base;
 			return a;
 		}
 		free(a);
 		a = NULL;
 	}
 	if (!PyErr_Occurred())
 		PyErr_NoMemory();
 	return NULL;
 }
 
 static void lfree(struct flist *a)
 {
 	if (a) {
 		free(a->base);
 		free(a);
 	}
 }
 
 static Py_ssize_t lsize(struct flist *a)
 {
 	return a->tail - a->head;
 }
 
 /* move hunks in source that are less cut to dest, compensating
    for changes in offset. the last hunk may be split if necessary.
 */
 static int gather(struct flist *dest, struct flist *src, int cut, int offset)
 {
 	struct frag *d = dest->tail, *s = src->head;
 	int postend, c, l;
 
 	while (s != src->tail) {
 		if (s->start + offset >= cut)
 			break; /* we've gone far enough */
 
 		postend = offset + s->start + s->len;
 		if (postend <= cut) {
 			/* save this hunk */
 			offset += s->start + s->len - s->end;
 			*d++ = *s++;
 		}
 		else {
 			/* break up this hunk */
 			c = cut - offset;
 			if (s->end < c)
 				c = s->end;
 			l = cut - offset - s->start;
 			if (s->len < l)
 				l = s->len;
 
 			offset += s->start + l - c;
 
 			d->start = s->start;
 			d->end = c;
 			d->len = l;
 			d->data = s->data;
 			d++;
 			s->start = c;
 			s->len = s->len - l;
 			s->data = s->data + l;
 
 			break;
 		}
 	}
 
 	dest->tail = d;
 	src->head = s;
 	return offset;
 }
 
 /* like gather, but with no output list */
 static int discard(struct flist *src, int cut, int offset)
 {
 	struct frag *s = src->head;
 	int postend, c, l;
 
 	while (s != src->tail) {
 		if (s->start + offset >= cut)
 			break;
 
 		postend = offset + s->start + s->len;
 		if (postend <= cut) {
 			offset += s->start + s->len - s->end;
 			s++;
 		}
 		else {
 			c = cut - offset;
 			if (s->end < c)
 				c = s->end;
 			l = cut - offset - s->start;
 			if (s->len < l)
 				l = s->len;
 
 			offset += s->start + l - c;
 			s->start = c;
 			s->len = s->len - l;
 			s->data = s->data + l;
 
 			break;
 		}
 	}
 
 	src->head = s;
 	return offset;
 }
 
 /* combine hunk lists a and b, while adjusting b for offset changes in a/
    this deletes a and b and returns the resultant list. */
 static struct flist *combine(struct flist *a, struct flist *b)
 {
 	struct flist *c = NULL;
 	struct frag *bh, *ct;
 	int offset = 0, post;
 
 	if (a && b)
 		c = lalloc((lsize(a) + lsize(b)) * 2);
 
 	if (c) {
 
 		for (bh = b->head; bh != b->tail; bh++) {
 			/* save old hunks */
 			offset = gather(c, a, bh->start, offset);
 
 			/* discard replaced hunks */
 			post = discard(a, bh->end, offset);
 
 			/* insert new hunk */
 			ct = c->tail;
 			ct->start = bh->start - offset;
 			ct->end = bh->end - post;
 			ct->len = bh->len;
 			ct->data = bh->data;
 			c->tail++;
 			offset = post;
 		}
 
 		/* hold on to tail from a */
 		memcpy(c->tail, a->head, sizeof(struct frag) * lsize(a));
 		c->tail += lsize(a);
 	}
 
 	lfree(a);
 	lfree(b);
 	return c;
 }
 
 /* decode a binary patch into a hunk list */
 static struct flist *decode(const char *bin, Py_ssize_t len)
 {
 	struct flist *l;
 	struct frag *lt;
-	const char *data = bin + 12, *end = bin + len;
+	int pos = 0;
 
 	/* assume worst case size, we won't have many of these lists */
 	l = lalloc(len / 12);
 	if (!l)
 		return NULL;
 
 	lt = l->tail;
 
-	while (data <= end) {
-		lt->start = getbe32(bin);
-		lt->end = getbe32(bin + 4);
-		lt->len = getbe32(bin + 8);
+	while (pos >= 0 && pos < len) {
+		lt->start = getbe32(bin + pos);
+		lt->end = getbe32(bin + pos + 4);
+		lt->len = getbe32(bin + pos + 8);
 		if (lt->start > lt->end)
 			break; /* sanity check */
-		bin = data + lt->len;
-		if (bin < data)
-			break; /* big data + big (bogus) len can wrap around */
-		lt->data = data;
-		data = bin + 12;
+		lt->data = bin + pos + 12;
+		pos += 12 + lt->len;
 		lt++;
 	}
 
-	if (bin != end) {
+	if (pos != len) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(mpatch_Error, "patch cannot be decoded");
 		lfree(l);
 		return NULL;
 	}
 
 	l->tail = lt;
 	return l;
 }
 
 /* calculate the size of resultant text */
 static Py_ssize_t calcsize(Py_ssize_t len, struct flist *l)
 {
 	Py_ssize_t outlen = 0, last = 0;
 	struct frag *f = l->head;
 
 	while (f != l->tail) {
 		if (f->start < last || f->end > len) {
 			if (!PyErr_Occurred())
 				PyErr_SetString(mpatch_Error,
 				                "invalid patch");
 			return -1;
 		}
 		outlen += f->start - last;
 		last = f->end;
 		outlen += f->len;
 		f++;
 	}
 
 	outlen += len - last;
 	return outlen;
 }
 
 static int apply(char *buf, const char *orig, Py_ssize_t len, struct flist *l)
 {
 	struct frag *f = l->head;
 	int last = 0;
 	char *p = buf;
 
 	while (f != l->tail) {
 		if (f->start < last || f->end > len) {
 			if (!PyErr_Occurred())
 				PyErr_SetString(mpatch_Error,
 				                "invalid patch");
 			return 0;
 		}
 		memcpy(p, orig + last, f->start - last);
 		p += f->start - last;
 		memcpy(p, f->data, f->len);
 		last = f->end;
 		p += f->len;
 		f++;
 	}
 	memcpy(p, orig + last, len - last);
 	return 1;
 }
 
 /* recursively generate a patch of all bins between start and end */
 static struct flist *fold(PyObject *bins, Py_ssize_t start, Py_ssize_t end)
 {
 	Py_ssize_t len, blen;
 	const char *buffer;
 
 	if (start + 1 == end) {
 		/* trivial case, output a decoded list */
 		PyObject *tmp = PyList_GetItem(bins, start);
 		if (!tmp)
 			return NULL;
 		if (PyObject_AsCharBuffer(tmp, &buffer, &blen))
 			return NULL;
 		return decode(buffer, blen);
 	}
 
 	/* divide and conquer, memory management is elsewhere */
 	len = (end - start) / 2;
 	return combine(fold(bins, start, start + len),
 		       fold(bins, start + len, end));
 }
 
 static PyObject *
 patches(PyObject *self, PyObject *args)
 {
 	PyObject *text, *bins, *result;
 	struct flist *patch;
 	const char *in;
 	char *out;
 	Py_ssize_t len, outlen, inlen;
 
 	if (!PyArg_ParseTuple(args, "OO:mpatch", &text, &bins))
 		return NULL;
 
 	len = PyList_Size(bins);
 	if (!len) {
 		/* nothing to do */
 		Py_INCREF(text);
 		return text;
 	}
 
 	if (PyObject_AsCharBuffer(text, &in, &inlen))
 		return NULL;
 
 	patch = fold(bins, 0, len);
 	if (!patch)
 		return NULL;
 
 	outlen = calcsize(inlen, patch);
 	if (outlen < 0) {
 		result = NULL;
 		goto cleanup;
 	}
 	result = PyBytes_FromStringAndSize(NULL, outlen);
 	if (!result) {
 		result = NULL;
 		goto cleanup;
 	}
 	out = PyBytes_AsString(result);
 	if (!apply(out, in, inlen, patch)) {
 		Py_DECREF(result);
 		result = NULL;
 	}
 cleanup:
 	lfree(patch);
 	return result;
 }
 
 /* calculate size of a patched file directly */
 static PyObject *
 patchedsize(PyObject *self, PyObject *args)
 {
-	long orig, start, end, len, outlen = 0, last = 0;
+	long orig, start, end, len, outlen = 0, last = 0, pos = 0;
 	Py_ssize_t patchlen;
-	char *bin, *binend, *data;
+	char *bin;
 
 	if (!PyArg_ParseTuple(args, "ls#", &orig, &bin, &patchlen))
 		return NULL;
 
-	binend = bin + patchlen;
-	data = bin + 12;
-
-	while (data <= binend) {
-		start = getbe32(bin);
-		end = getbe32(bin + 4);
-		len = getbe32(bin + 8);
+	while (pos >= 0 && pos < patchlen) {
+		start = getbe32(bin + pos);
+		end = getbe32(bin + pos + 4);
+		len = getbe32(bin + pos + 8);
 		if (start > end)
 			break; /* sanity check */
-		bin = data + len;
-		if (bin < data)
-			break; /* big data + big (bogus) len can wrap around */
-		data = bin + 12;
+		pos += 12 + len;
 		outlen += start - last;
 		last = end;
 		outlen += len;
 	}
 
-	if (bin != binend) {
+	if (pos != patchlen) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(mpatch_Error, "patch cannot be decoded");
 		return NULL;
 	}
 
 	outlen += orig - last;
 	return Py_BuildValue("l", outlen);
 }
 
 static PyMethodDef methods[] = {
 	{"patches", patches, METH_VARARGS, "apply a series of patches\n"},
 	{"patchedsize", patchedsize, METH_VARARGS, "calculed patched size\n"},
 	{NULL, NULL}
 };
 
 #ifdef IS_PY3K
 static struct PyModuleDef mpatch_module = {
 	PyModuleDef_HEAD_INIT,
 	"mpatch",
 	mpatch_doc,
 	-1,
 	methods
 };
 
 PyMODINIT_FUNC PyInit_mpatch(void)
 {
 	PyObject *m;
 
 	m = PyModule_Create(&mpatch_module);
 	if (m == NULL)
 		return NULL;
 
 	mpatch_Error = PyErr_NewException("mpatch.mpatchError", NULL, NULL);
 	Py_INCREF(mpatch_Error);
 	PyModule_AddObject(m, "mpatchError", mpatch_Error);
 
 	return m;
 }
 #else
 PyMODINIT_FUNC
 initmpatch(void)
 {
 	Py_InitModule3("mpatch", methods, mpatch_doc);
 	mpatch_Error = PyErr_NewException("mpatch.mpatchError", NULL, NULL);
 }
 #endif

mercurial/parsers.c

@@ -1,1965 +1,1959 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.
 */
 
 #include <Python.h>
 #include <ctype.h>
 #include <stddef.h>
 #include <string.h>
 
 #include "util.h"
 
 static int8_t hextable[256] = {
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, -1, -1, -1, -1, -1, -1, /* 0-9 */
 	-1, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* A-F */
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* a-f */
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
 };
 
 static inline int hexdigit(const char *p, Py_ssize_t off)
 {
 	int8_t val = hextable[(unsigned char)p[off]];
 
 	if (val >= 0) {
 		return val;
 	}
 
 	PyErr_SetString(PyExc_ValueError, "input contains non-hex character");
 	return 0;
 }
 
 /*
  * Turn a hex-encoded string into binary.
  */
 static PyObject *unhexlify(const char *str, int len)
 {
 	PyObject *ret;
 	char *d;
 	int i;
 
 	ret = PyBytes_FromStringAndSize(NULL, len / 2);
 
 	if (!ret)
 		return NULL;
 
 	d = PyBytes_AsString(ret);
 
 	for (i = 0; i < len;) {
 		int hi = hexdigit(str, i++);
 		int lo = hexdigit(str, i++);
 		*d++ = (hi << 4) | lo;
 	}
 
 	return ret;
 }
 
 /*
  * This code assumes that a manifest is stitched together with newline
  * ('\n') characters.
  */
 static PyObject *parse_manifest(PyObject *self, PyObject *args)
 {
 	PyObject *mfdict, *fdict;
 	char *str, *start, *end;
 	int len;
 
 	if (!PyArg_ParseTuple(args, "O!O!s#:parse_manifest",
 			      &PyDict_Type, &mfdict,
 			      &PyDict_Type, &fdict,
 			      &str, &len))
 		goto quit;
 
 	start = str;
 	end = str + len;
 	while (start < end) {
 		PyObject *file = NULL, *node = NULL;
 		PyObject *flags = NULL;
 		char *zero = NULL, *newline = NULL;
 		ptrdiff_t nlen;
 
 		zero = memchr(start, '\0', end - start);
 		if (!zero) {
 			PyErr_SetString(PyExc_ValueError,
 					"manifest entry has no separator");
 			goto quit;
 		}
 
 		newline = memchr(zero + 1, '\n', end - (zero + 1));
 		if (!newline) {
 			PyErr_SetString(PyExc_ValueError,
 					"manifest contains trailing garbage");
 			goto quit;
 		}
 
 		file = PyBytes_FromStringAndSize(start, zero - start);
 
 		if (!file)
 			goto bail;
 
 		nlen = newline - zero - 1;
 
 		node = unhexlify(zero + 1, nlen > 40 ? 40 : (int)nlen);
 		if (!node)
 			goto bail;
 
 		if (nlen > 40) {
 			flags = PyBytes_FromStringAndSize(zero + 41,
 							   nlen - 40);
 			if (!flags)
 				goto bail;
 
 			if (PyDict_SetItem(fdict, file, flags) == -1)
 				goto bail;
 		}
 
 		if (PyDict_SetItem(mfdict, file, node) == -1)
 			goto bail;
 
 		start = newline + 1;
 
 		Py_XDECREF(flags);
 		Py_XDECREF(node);
 		Py_XDECREF(file);
 		continue;
 	bail:
 		Py_XDECREF(flags);
 		Py_XDECREF(node);
 		Py_XDECREF(file);
 		goto quit;
 	}
 
 	Py_INCREF(Py_None);
 	return Py_None;
 quit:
 	return NULL;
 }
 
 static PyObject *parse_dirstate(PyObject *self, PyObject *args)
 {
 	PyObject *dmap, *cmap, *parents = NULL, *ret = NULL;
 	PyObject *fname = NULL, *cname = NULL, *entry = NULL;
-	char state, *str, *cur, *end, *cpos;
+	char state, *cur, *str, *cpos;
 	int mode, size, mtime;
 	unsigned int flen;
-	int len;
+	int len, pos = 40;
 
 	if (!PyArg_ParseTuple(args, "O!O!s#:parse_dirstate",
 			      &PyDict_Type, &dmap,
 			      &PyDict_Type, &cmap,
 			      &str, &len))
 		goto quit;
 
 	/* read parents */
 	if (len < 40)
 		goto quit;
 
 	parents = Py_BuildValue("s#s#", str, 20, str + 20, 20);
 	if (!parents)
 		goto quit;
 
 	/* read filenames */
-	cur = str + 40;
-	end = str + len;
-
-	while (cur < end - 17) {
+	while (pos >= 40 && pos < len) {
+		cur = str + pos;
 		/* unpack header */
 		state = *cur;
 		mode = getbe32(cur + 1);
 		size = getbe32(cur + 5);
 		mtime = getbe32(cur + 9);
 		flen = getbe32(cur + 13);
+		pos += 17;
 		cur += 17;
-		if (cur + flen > end || cur + flen < cur) {
+		if (flen > len - pos || flen < 0) {
 			PyErr_SetString(PyExc_ValueError, "overflow in dirstate");
 			goto quit;
 		}
 
 		entry = Py_BuildValue("ciii", state, mode, size, mtime);
 		if (!entry)
 			goto quit;
 		PyObject_GC_UnTrack(entry); /* don't waste time with this */
 
 		cpos = memchr(cur, 0, flen);
 		if (cpos) {
 			fname = PyBytes_FromStringAndSize(cur, cpos - cur);
 			cname = PyBytes_FromStringAndSize(cpos + 1,
 							   flen - (cpos - cur) - 1);
 			if (!fname || !cname ||
 			    PyDict_SetItem(cmap, fname, cname) == -1 ||
 			    PyDict_SetItem(dmap, fname, entry) == -1)
 				goto quit;
 			Py_DECREF(cname);
 		} else {
 			fname = PyBytes_FromStringAndSize(cur, flen);
 			if (!fname ||
 			    PyDict_SetItem(dmap, fname, entry) == -1)
 				goto quit;
 		}
-		cur += flen;
 		Py_DECREF(fname);
 		Py_DECREF(entry);
 		fname = cname = entry = NULL;
+		pos += flen;
 	}
 
 	ret = parents;
 	Py_INCREF(ret);
 quit:
 	Py_XDECREF(fname);
 	Py_XDECREF(cname);
 	Py_XDECREF(entry);
 	Py_XDECREF(parents);
 	return ret;
 }
 
 static inline int getintat(PyObject *tuple, int off, uint32_t *v)
 {
 	PyObject *o = PyTuple_GET_ITEM(tuple, off);
 	long val;
 
 	if (PyInt_Check(o))
 		val = PyInt_AS_LONG(o);
 	else if (PyLong_Check(o)) {
 		val = PyLong_AsLong(o);
 		if (val == -1 && PyErr_Occurred())
 			return -1;
 	} else {
 		PyErr_SetString(PyExc_TypeError, "expected an int or long");
 		return -1;
 	}
 	if (LONG_MAX > INT_MAX && (val > INT_MAX || val < INT_MIN)) {
 		PyErr_SetString(PyExc_OverflowError,
 				"Python value to large to convert to uint32_t");
 		return -1;
 	}
 	*v = (uint32_t)val;
 	return 0;
 }
 
 static PyObject *dirstate_unset;
 
 /*
  * Efficiently pack a dirstate object into its on-disk format.
  */
 static PyObject *pack_dirstate(PyObject *self, PyObject *args)
 {
 	PyObject *packobj = NULL;
 	PyObject *map, *copymap, *pl;
 	Py_ssize_t nbytes, pos, l;
 	PyObject *k, *v, *pn;
 	char *p, *s;
 	double now;
 
 	if (!PyArg_ParseTuple(args, "O!O!Od:pack_dirstate",
 			      &PyDict_Type, &map, &PyDict_Type, &copymap,
 			      &pl, &now))
 		return NULL;
 
 	if (!PySequence_Check(pl) || PySequence_Size(pl) != 2) {
 		PyErr_SetString(PyExc_TypeError, "expected 2-element sequence");
 		return NULL;
 	}
 
 	/* Figure out how much we need to allocate. */
 	for (nbytes = 40, pos = 0; PyDict_Next(map, &pos, &k, &v);) {
 		PyObject *c;
 		if (!PyString_Check(k)) {
 			PyErr_SetString(PyExc_TypeError, "expected string key");
 			goto bail;
 		}
 		nbytes += PyString_GET_SIZE(k) + 17;
 		c = PyDict_GetItem(copymap, k);
 		if (c) {
 			if (!PyString_Check(c)) {
 				PyErr_SetString(PyExc_TypeError,
 						"expected string key");
 				goto bail;
 			}
 			nbytes += PyString_GET_SIZE(c) + 1;
 		}
 	}
 
 	packobj = PyString_FromStringAndSize(NULL, nbytes);
 	if (packobj == NULL)
 		goto bail;
 
 	p = PyString_AS_STRING(packobj);
 
 	pn = PySequence_ITEM(pl, 0);
 	if (PyString_AsStringAndSize(pn, &s, &l) == -1 || l != 20) {
 		PyErr_SetString(PyExc_TypeError, "expected a 20-byte hash");
 		goto bail;
 	}
 	memcpy(p, s, l);
 	p += 20;
 	pn = PySequence_ITEM(pl, 1);
 	if (PyString_AsStringAndSize(pn, &s, &l) == -1 || l != 20) {
 		PyErr_SetString(PyExc_TypeError, "expected a 20-byte hash");
 		goto bail;
 	}
 	memcpy(p, s, l);
 	p += 20;
 
 	for (pos = 0; PyDict_Next(map, &pos, &k, &v); ) {
 		uint32_t mode, size, mtime;
 		Py_ssize_t len, l;
 		PyObject *o;
 		char *s, *t;
 
 		if (!PyTuple_Check(v) || PyTuple_GET_SIZE(v) != 4) {
 			PyErr_SetString(PyExc_TypeError, "expected a 4-tuple");
 			goto bail;
 		}
 		o = PyTuple_GET_ITEM(v, 0);
 		if (PyString_AsStringAndSize(o, &s, &l) == -1 || l != 1) {
 			PyErr_SetString(PyExc_TypeError, "expected one byte");
 			goto bail;
 		}
 		*p++ = *s;
 		if (getintat(v, 1, &mode) == -1)
 			goto bail;
 		if (getintat(v, 2, &size) == -1)
 			goto bail;
 		if (getintat(v, 3, &mtime) == -1)
 			goto bail;
 		if (*s == 'n' && mtime == (uint32_t)now) {
 			/* See pure/parsers.py:pack_dirstate for why we do
 			 * this. */
 			if (PyDict_SetItem(map, k, dirstate_unset) == -1)
 				goto bail;
 			mtime = -1;
 		}
 		putbe32(mode, p);
 		putbe32(size, p + 4);
 		putbe32(mtime, p + 8);
 		t = p + 12;
 		p += 16;
 		len = PyString_GET_SIZE(k);
 		memcpy(p, PyString_AS_STRING(k), len);
 		p += len;
 		o = PyDict_GetItem(copymap, k);
 		if (o) {
 			*p++ = '\0';
 			l = PyString_GET_SIZE(o);
 			memcpy(p, PyString_AS_STRING(o), l);
 			p += l;
 			len += l + 1;
 		}
 		putbe32((uint32_t)len, t);
 	}
 
 	pos = p - PyString_AS_STRING(packobj);
 	if (pos != nbytes) {
 		PyErr_Format(PyExc_SystemError, "bad dirstate size: %ld != %ld",
                              (long)pos, (long)nbytes);
 		goto bail;
 	}
 
 	return packobj;
 bail:
 	Py_XDECREF(packobj);
 	return NULL;
 }
 
 /*
  * 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 + 1)
  * Zero is empty
  */
 typedef struct {
 	int children[16];
 } nodetree;
 
 /*
  * This class has two behaviours.
  *
  * When used in a list-like way (with integer keys), we decode an
  * entry in a RevlogNG index file on demand. Our last entry is a
  * sentinel, always a nullid.  We have limited support for
  * integer-keyed insert and delete, only at elements right before the
  * sentinel.
  *
  * With string keys, we lazily perform a reverse mapping from node to
  * rev, using a base-16 trie.
  */
 typedef struct {
 	PyObject_HEAD
 	/* Type-specific fields go here. */
 	PyObject *data;        /* raw bytes of index */
 	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 */
 	nodetree *nt;          /* base-16 trie */
 	int ntlength;          /* # nodes in use */
 	int ntcapacity;        /* # nodes allocated */
 	int ntdepth;           /* maximum depth of tree */
 	int ntsplits;          /* # splits performed */
 	int ntrev;             /* last rev scanned */
 	int ntlookups;         /* # lookups */
 	int ntmisses;          /* # lookups that miss the cache */
 	int inlined;
 } indexObject;
 
 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;
 static const char nullid[20];
 
 static long inline_scan(indexObject *self, const char **offsets);
 
 #if LONG_MAX == 0x7fffffffL
 static char *tuple_format = "Kiiiiiis#";
 #else
 static char *tuple_format = "kiiiiiis#";
 #endif
 
 /* A RevlogNG v1 index entry is 64 bytes long. */
 static const long v1_hdrsize = 64;
 
 /*
  * 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 = malloc(self->raw_length *
 					       sizeof(*self->offsets));
 			if (self->offsets == NULL)
 				return (const char *)PyErr_NoMemory();
 			inline_scan(self, self->offsets);
 		}
 		return self->offsets[pos];
 	}
 
 	return PyString_AS_STRING(self->data) + pos * v1_hdrsize;
 }
 
 /*
  * 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 < 0)
 		pos += length;
 
 	if (pos < 0 || pos >= length) {
 		PyErr_SetString(PyExc_IndexError, "revlog index out of range");
 		return NULL;
 	}
 
 	if (pos == length - 1) {
 		Py_INCREF(nullentry);
 		return nullentry;
 	}
 
 	if (pos >= self->length - 1) {
 		PyObject *obj;
 		obj = PyList_GET_ITEM(self->added, pos - self->length + 1);
 		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, 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 == length - 1 || pos == INT_MAX)
 		return nullid;
 
 	if (pos >= length)
 		return NULL;
 
 	if (pos >= self->length - 1) {
 		PyObject *tuple, *str;
 		tuple = PyList_GET_ITEM(self->added, pos - self->length + 1);
 		str = PyTuple_GetItem(tuple, 7);
 		return str ? PyString_AS_STRING(str) : NULL;
 	}
 
 	data = index_deref(self, pos);
 	return data ? data + 32 : NULL;
 }
 
 static int nt_insert(indexObject *self, const char *node, int rev);
 
 static int node_check(PyObject *obj, char **node, Py_ssize_t *nodelen)
 {
 	if (PyString_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_insert(indexObject *self, PyObject *args)
 {
 	PyObject *obj;
 	char *node;
 	long offset;
 	Py_ssize_t len, nodelen;
 
 	if (!PyArg_ParseTuple(args, "lO", &offset, &obj))
 		return NULL;
 
 	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, &nodelen) == -1)
 		return NULL;
 
 	len = index_length(self);
 
 	if (offset < 0)
 		offset += len;
 
 	if (offset != len - 1) {
 		PyErr_SetString(PyExc_IndexError,
 				"insert only supported at index -1");
 		return NULL;
 	}
 
 	if (offset > INT_MAX) {
 		PyErr_SetString(PyExc_ValueError,
 				"currently only 2**31 revs supported");
 		return NULL;
 	}
 
 	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->nt)
 		nt_insert(self, node, (int)offset);
 
 	Py_CLEAR(self->headrevs);
 	Py_RETURN_NONE;
 }
 
 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) {
 		free(self->offsets);
 		self->offsets = NULL;
 	}
 	if (self->nt) {
 		free(self->nt);
 		self->nt = NULL;
 	}
 	Py_CLEAR(self->headrevs);
 }
 
 static PyObject *index_clearcaches(indexObject *self)
 {
 	_index_clearcaches(self);
 	self->ntlength = self->ntcapacity = 0;
 	self->ntdepth = self->ntsplits = 0;
 	self->ntrev = -1;
 	self->ntlookups = self->ntmisses = 0;
 	Py_RETURN_NONE;
 }
 
 static PyObject *index_stats(indexObject *self)
 {
 	PyObject *obj = PyDict_New();
 
 	if (obj == NULL)
 		return NULL;
 
 #define istat(__n, __d) \
 	if (PyDict_SetItemString(obj, __d, PyInt_FromSsize_t(self->__n)) == -1) \
 		goto bail;
 
 	if (self->added) {
 		Py_ssize_t len = PyList_GET_SIZE(self->added);
 		if (PyDict_SetItemString(obj, "index entries added",
 					 PyInt_FromSsize_t(len)) == -1)
 			goto bail;
 	}
 
 	if (self->raw_length != self->length - 1)
 		istat(raw_length, "revs on disk");
 	istat(length, "revs in memory");
 	istat(ntcapacity, "node trie capacity");
 	istat(ntdepth, "node trie depth");
 	istat(ntlength, "node trie count");
 	istat(ntlookups, "node trie lookups");
 	istat(ntmisses, "node trie misses");
 	istat(ntrev, "node trie last rev scanned");
 	istat(ntsplits, "node trie splits");
 
 #undef istat
 
 	return obj;
 
 bail:
 	Py_XDECREF(obj);
 	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 PyObject *index_headrevs(indexObject *self)
 {
 	Py_ssize_t i, len, addlen;
 	char *nothead = NULL;
 	PyObject *heads;
 
 	if (self->headrevs)
 		return list_copy(self->headrevs);
 
 	len = index_length(self) - 1;
 	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)
 		goto bail;
 
 	for (i = 0; i < self->raw_length; i++) {
 		const char *data = index_deref(self, i);
 		int parent_1 = getbe32(data + 24);
 		int parent_2 = getbe32(data + 28);
 		if (parent_1 >= 0)
 			nothead[parent_1] = 1;
 		if (parent_2 >= 0)
 			nothead[parent_2] = 1;
 	}
 
 	addlen = self->added ? PyList_GET_SIZE(self->added) : 0;
 
 	for (i = 0; i < addlen; i++) {
 		PyObject *rev = PyList_GET_ITEM(self->added, i);
 		PyObject *p1 = PyTuple_GET_ITEM(rev, 5);
 		PyObject *p2 = PyTuple_GET_ITEM(rev, 6);
 		long parent_1, parent_2;
 
 		if (!PyInt_Check(p1) || !PyInt_Check(p2)) {
 			PyErr_SetString(PyExc_TypeError,
 					"revlog parents are invalid");
 			goto bail;
 		}
 		parent_1 = PyInt_AS_LONG(p1);
 		parent_2 = PyInt_AS_LONG(p2);
 		if (parent_1 >= 0)
 			nothead[parent_1] = 1;
 		if (parent_2 >= 0)
 			nothead[parent_2] = 1;
 	}
 
 	for (i = 0; i < len; i++) {
 		PyObject *head;
 
 		if (nothead[i])
 			continue;
 		head = PyInt_FromLong(i);
 		if (head == NULL || PyList_Append(heads, head) == -1) {
 			Py_XDECREF(head);
 			goto bail;
 		}
 	}
 
 done:
 	self->headrevs = heads;
 	free(nothead);
 	return list_copy(self->headrevs);
 bail:
 	Py_XDECREF(heads);
 	free(nothead);
 	return NULL;
 }
 
 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(indexObject *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 (self->nt == NULL)
 		return -2;
 
 	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);
 		nodetree *n = &self->nt[off];
 		int v = n->children[k];
 
 		if (v < 0) {
 			const char *n;
 			Py_ssize_t i;
 
 			v = -v - 1;
 			n = index_node(self, 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(indexObject *self)
 {
 	if (self->ntlength == self->ntcapacity) {
 		self->ntcapacity *= 2;
 		self->nt = realloc(self->nt,
 				   self->ntcapacity * sizeof(nodetree));
 		if (self->nt == NULL) {
 			PyErr_SetString(PyExc_MemoryError, "out of memory");
 			return -1;
 		}
 		memset(&self->nt[self->ntlength], 0,
 		       sizeof(nodetree) * (self->ntcapacity - self->ntlength));
 	}
 	return self->ntlength++;
 }
 
 static int nt_insert(indexObject *self, const char *node, int rev)
 {
 	int level = 0;
 	int off = 0;
 
 	while (level < 40) {
 		int k = nt_level(node, level);
 		nodetree *n;
 		int v;
 
 		n = &self->nt[off];
 		v = n->children[k];
 
 		if (v == 0) {
 			n->children[k] = -rev - 1;
 			return 0;
 		}
 		if (v < 0) {
 			const char *oldnode = index_node(self, -v - 1);
 			int noff;
 
 			if (!oldnode || !memcmp(oldnode, node, 20)) {
 				n->children[k] = -rev - 1;
 				return 0;
 			}
 			noff = nt_new(self);
 			if (noff == -1)
 				return -1;
 			/* self->nt may have been changed by realloc */
 			self->nt[off].children[k] = noff;
 			off = noff;
 			n = &self->nt[off];
 			n->children[nt_level(oldnode, ++level)] = v;
 			if (level > self->ntdepth)
 				self->ntdepth = level;
 			self->ntsplits += 1;
 		} else {
 			level += 1;
 			off = v;
 		}
 	}
 
 	return -1;
 }
 
 static int nt_init(indexObject *self)
 {
 	if (self->nt == NULL) {
 		if (self->raw_length > INT_MAX) {
 			PyErr_SetString(PyExc_ValueError, "overflow in nt_init");
 			return -1;
 		}
 		self->ntcapacity = self->raw_length < 4
 			? 4 : (int)self->raw_length / 2;
 
 		self->nt = calloc(self->ntcapacity, sizeof(nodetree));
 		if (self->nt == NULL) {
 			PyErr_NoMemory();
 			return -1;
 		}
 		self->ntlength = 1;
 		self->ntrev = (int)index_length(self) - 1;
 		self->ntlookups = 1;
 		self->ntmisses = 0;
 		if (nt_insert(self, nullid, INT_MAX) == -1)
 			return -1;
 	}
 	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;
 
 	self->ntlookups++;
 	rev = nt_find(self, node, nodelen, 0);
 	if (rev >= -1)
 		return rev;
 
 	if (nt_init(self) == -1)
 		return -3;
 
 	/*
 	 * 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(self, rev);
 			if (n == NULL)
 				return -2;
 			if (memcmp(node, n, nodelen > 20 ? 20 : nodelen) == 0) {
 				if (nt_insert(self, n, rev) == -1)
 					return -3;
 				break;
 			}
 		}
 	} else {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node(self, rev);
 			if (n == NULL) {
 				self->ntrev = rev + 1;
 				return -2;
 			}
 			if (nt_insert(self, 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 *raise_revlog_error(void)
 {
 	static PyObject *errclass;
 	PyObject *mod = NULL, *errobj;
 
 	if (errclass == NULL) {
 		PyObject *dict;
 
 		mod = PyImport_ImportModule("mercurial.error");
 		if (mod == NULL)
 			goto classfail;
 
 		dict = PyModule_GetDict(mod);
 		if (dict == NULL)
 			goto classfail;
 
 		errclass = PyDict_GetItemString(dict, "RevlogError");
 		if (errclass == NULL) {
 			PyErr_SetString(PyExc_SystemError,
 					"could not find RevlogError");
 			goto classfail;
 		}
 		Py_INCREF(errclass);
 	}
 
 	errobj = PyObject_CallFunction(errclass, NULL);
 	if (errobj == NULL)
 		return NULL;
 	PyErr_SetObject(errclass, errobj);
 	return errobj;
 
 classfail:
 	Py_XDECREF(mod);
 	return NULL;
 }
 
 static PyObject *index_getitem(indexObject *self, PyObject *value)
 {
 	char *node;
 	Py_ssize_t nodelen;
 	int rev;
 
 	if (PyInt_Check(value))
 		return index_get(self, PyInt_AS_LONG(value));
 
 	if (node_check(value, &node, &nodelen) == -1)
 		return NULL;
 	rev = index_find_node(self, node, nodelen);
 	if (rev >= -1)
 		return PyInt_FromLong(rev);
 	if (rev == -2)
 		raise_revlog_error();
 	return NULL;
 }
 
 static int nt_partialmatch(indexObject *self, const char *node,
 			   Py_ssize_t nodelen)
 {
 	int rev;
 
 	if (nt_init(self) == -1)
 		return -3;
 
 	if (self->ntrev > 0) {
 		/* ensure that the radix tree is fully populated */
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node(self, rev);
 			if (n == NULL)
 				return -2;
 			if (nt_insert(self, n, rev) == -1)
 				return -3;
 		}
 		self->ntrev = rev;
 	}
 
 	return nt_find(self, node, nodelen, 1);
 }
 
 static PyObject *index_partialmatch(indexObject *self, PyObject *args)
 {
 	const char *fullnode;
 	int nodelen;
 	char *node;
 	int rev, i;
 
 	if (!PyArg_ParseTuple(args, "s#", &node, &nodelen))
 		return NULL;
 
 	if (nodelen < 4) {
 		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;
 	}
 
 	rev = nt_partialmatch(self, node, nodelen);
 
 	switch (rev) {
 	case -4:
 		raise_revlog_error();
 	case -3:
 		return NULL;
 	case -2:
 		Py_RETURN_NONE;
 	case -1:
 		return PyString_FromStringAndSize(nullid, 20);
 	}
 
 	fullnode = index_node(self, rev);
 	if (fullnode == NULL) {
 		PyErr_Format(PyExc_IndexError,
 			     "could not access rev %d", rev);
 		return NULL;
 	}
 	return PyString_FromStringAndSize(fullnode, 20);
 }
 
 static PyObject *index_m_get(indexObject *self, PyObject *args)
 {
 	Py_ssize_t nodelen;
 	PyObject *val;
 	char *node;
 	int rev;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(val, &node, &nodelen) == -1)
 		return NULL;
 	rev = index_find_node(self, node, nodelen);
 	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;
 	Py_ssize_t nodelen;
 
 	if (PyInt_Check(value)) {
 		long rev = PyInt_AS_LONG(value);
 		return rev >= -1 && rev < index_length(self);
 	}
 
 	if (node_check(value, &node, &nodelen) == -1)
 		return -1;
 
 	switch (index_find_node(self, node, nodelen)) {
 	case -3:
 		return -1;
 	case -2:
 		return 0;
 	default:
 		return 1;
 	}
 }
 
 static inline void index_get_parents(indexObject *self, int rev, int *ps)
 {
 	if (rev >= self->length - 1) {
 		PyObject *tuple = PyList_GET_ITEM(self->added,
 						  rev - self->length + 1);
 		ps[0] = (int)PyInt_AS_LONG(PyTuple_GET_ITEM(tuple, 5));
 		ps[1] = (int)PyInt_AS_LONG(PyTuple_GET_ITEM(tuple, 6));
 	} else {
 		const char *data = index_deref(self, rev);
 		ps[0] = getbe32(data + 24);
 		ps[1] = getbe32(data + 28);
 	}
 }
 
 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, left;
 	int maxrev = -1;
 	long 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 = left = revcount;
 
 	for (v = maxrev; v >= 0 && interesting; v--) {
 		long 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) {
 						if (--left <= 1)
 							goto done;
 						break;
 					}
 				}
 			}
 		}
 		index_get_parents(self, v, parents);
 
 		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;
 	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), 2 << 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;
 	}
 
 	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];
 		index_get_parents(self, v, parents);
 
 		for (i = 0; i < 2; i++) {
 			int p = parents[i];
 			long nsp, sp;
 			int dp;
 
 			if (p == -1)
 				continue;
 
 			dp = depth[p];
 			nsp = sp = seen[p];
 			if (dp <= dv) {
 				depth[p] = dv + 1;
 				if (sp != sv) {
 					interesting[sv] += 1;
 					nsp = seen[p] = sv;
 					if (sp) {
 						interesting[sp] -= 1;
 						if (interesting[sp] == 0)
 							ninteresting -= 1;
 					}
 				}
 			}
 			else if (dv == dp - 1) {
 				nsp = sp | sv;
 				if (nsp == sp)
 					continue;
 				seen[p] = nsp;
 				interesting[sp] -= 1;
 				if (interesting[sp] == 0 && 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)
 		return PyList_New(0);
 
 	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);
 
 	free(depth);
 	free(seen);
 	free(interesting);
 	Py_DECREF(dict);
 
 	return keys;
 bail:
 	free(depth);
 	free(seen);
 	free(interesting);
 	Py_XDECREF(dict);
 
 	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 = NULL, *gca = NULL;
 	Py_ssize_t argcount, i, len;
 	bitmask repeat = 0;
 	int revcount = 0;
 	int *revs;
 
 	argcount = PySequence_Length(args);
 	revs = malloc(argcount * sizeof(*revs));
 	if (argcount > 0 && revs == NULL)
 		return PyErr_NoMemory();
 	len = index_length(self) - 1;
 
 	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");
 			goto bail;
 		}
 		val = PyInt_AsLong(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;
 	}
 
 	gca = find_gca_candidates(self, revs, revcount);
 	if (gca == NULL)
 		goto bail;
 
 	if (PyList_GET_SIZE(gca) <= 1) {
 		ret = gca;
 		Py_INCREF(gca);
 	}
 	else if (PyList_GET_SIZE(gca) == 1) {
 		ret = PyList_GET_ITEM(gca, 0);
 		Py_INCREF(ret);
 	}
 	else ret = find_deepest(self, gca);
 
 done:
 	free(revs);
 	Py_XDECREF(gca);
 
 	return ret;
 
 bail:
 	free(revs);
 	Py_XDECREF(gca);
 	Py_XDECREF(ret);
 	return NULL;
 }
 
 /*
  * Invalidate any trie entries introduced by added revs.
  */
 static void nt_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_insert(self, PyString_AS_STRING(node), -1);
 	}
 
 	if (start == 0)
 		Py_CLEAR(self->added);
 }
 
 /*
  * Delete a numeric range of revs, which must be at the end of the
  * range, but exclude the sentinel nullid entry.
  */
 static int index_slice_del(indexObject *self, PyObject *item)
 {
 	Py_ssize_t start, stop, step, slicelength;
 	Py_ssize_t length = index_length(self);
 	int ret = 0;
 
 	if (PySlice_GetIndicesEx((PySliceObject*)item, length,
 				 &start, &stop, &step, &slicelength) < 0)
 		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 - 1) {
 		if (self->nt) {
 			Py_ssize_t i;
 
 			for (i = start + 1; i < self->length - 1; i++) {
 				const char *node = index_node(self, i);
 
 				if (node)
 					nt_insert(self, node, -1);
 			}
 			if (self->added)
 				nt_invalidate_added(self, 0);
 			if (self->ntrev > start)
 				self->ntrev = (int)start;
 		}
 		self->length = start + 1;
 		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->nt) {
 		nt_invalidate_added(self, start - self->length + 1);
 		if (self->ntrev > start)
 			self->ntrev = (int)start;
 	}
 	if (self->added)
 		ret = PyList_SetSlice(self->added, start - self->length + 1,
 				      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;
 	Py_ssize_t nodelen;
 	long rev;
 
 	if (PySlice_Check(item) && value == NULL)
 		return index_slice_del(self, item);
 
 	if (node_check(item, &node, &nodelen) == -1)
 		return -1;
 
 	if (value == NULL)
 		return self->nt ? nt_insert(self, node, -1) : 0;
 	rev = PyInt_AsLong(value);
 	if (rev > INT_MAX || rev < 0) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "rev out of range");
 		return -1;
 	}
 	return nt_insert(self, node, (int)rev);
 }
 
 /*
  * Find all RevlogNG entries in an index that has inline data. Update
  * the optional "offsets" table with those entries.
  */
 static long inline_scan(indexObject *self, const char **offsets)
 {
 	const char *data = PyString_AS_STRING(self->data);
-	const char *end = data + PyString_GET_SIZE(self->data);
+	Py_ssize_t pos = 0;
+	Py_ssize_t end = PyString_GET_SIZE(self->data);
 	long incr = v1_hdrsize;
 	Py_ssize_t len = 0;
 
-	while (data + v1_hdrsize <= end) {
+	while (pos + v1_hdrsize <= end && pos >= 0) {
 		uint32_t comp_len;
-		const char *old_data;
 		/* 3rd element of header is length of compressed inline data */
-		comp_len = getbe32(data + 8);
+		comp_len = getbe32(data + pos + 8);
 		incr = v1_hdrsize + comp_len;
-		if (incr < v1_hdrsize)
-			break;
 		if (offsets)
-			offsets[len] = data;
+			offsets[len] = data + pos;
 		len++;
-		old_data = data;
-		data += incr;
-		if (data <= old_data)
-			break;
+		pos += incr;
 	}
 
-	if (data != end && data + v1_hdrsize != end) {
+	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;
 	self->headrevs = NULL;
 	self->nt = NULL;
 	self->offsets = NULL;
 
 	if (!PyArg_ParseTuple(args, "OO", &data_obj, &inlined_obj))
 		return -1;
 	if (!PyString_Check(data_obj)) {
 		PyErr_SetString(PyExc_TypeError, "data is not a string");
 		return -1;
 	}
 	size = PyString_GET_SIZE(data_obj);
 
 	self->inlined = inlined_obj && PyObject_IsTrue(inlined_obj);
 	self->data = data_obj;
 
 	self->ntlength = self->ntcapacity = 0;
 	self->ntdepth = self->ntsplits = 0;
 	self->ntlookups = self->ntmisses = 0;
 	self->ntrev = -1;
 	Py_INCREF(self->data);
 
 	if (self->inlined) {
 		long len = inline_scan(self, NULL);
 		if (len == -1)
 			goto bail;
 		self->raw_length = len;
 		self->length = len + 1;
 	} 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 + 1;
 	}
 
 	return 0;
 bail:
 	return -1;
 }
 
 static PyObject *index_nodemap(indexObject *self)
 {
 	Py_INCREF(self);
 	return (PyObject *)self;
 }
 
 static void index_dealloc(indexObject *self)
 {
 	_index_clearcaches(self);
 	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"},
 	{"clearcaches", (PyCFunction)index_clearcaches, METH_NOARGS,
 	 "clear the index caches"},
 	{"get", (PyCFunction)index_m_get, METH_VARARGS,
 	 "get an index entry"},
 	{"headrevs", (PyCFunction)index_headrevs, METH_NOARGS,
 	 "get head revisions"},
 	{"insert", (PyCFunction)index_insert, METH_VARARGS,
 	 "insert an index entry"},
 	{"partialmatch", (PyCFunction)index_partialmatch, METH_VARARGS,
 	 "match a potentially ambiguous node 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 */
 };
 
 static PyTypeObject indexType = {
 	PyObject_HEAD_INIT(NULL)
 	0,                         /* ob_size */
 	"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 (index_file_content, 0), else None
  *
  * added complications are for backwards compatibility
  */
 static PyObject *parse_index2(PyObject *self, PyObject *args)
 {
 	PyObject *tuple = NULL, *cache = NULL;
 	indexObject *idx;
 	int ret;
 
 	idx = PyObject_New(indexObject, &indexType);
 	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 char parsers_doc[] = "Efficient content parsing.";
 
 PyObject *encodedir(PyObject *self, PyObject *args);
 PyObject *pathencode(PyObject *self, PyObject *args);
 PyObject *lowerencode(PyObject *self, PyObject *args);
 
 static PyMethodDef methods[] = {
 	{"pack_dirstate", pack_dirstate, METH_VARARGS, "pack a dirstate\n"},
 	{"parse_manifest", parse_manifest, METH_VARARGS, "parse a manifest\n"},
 	{"parse_dirstate", parse_dirstate, METH_VARARGS, "parse a dirstate\n"},
 	{"parse_index2", parse_index2, METH_VARARGS, "parse a revlog index\n"},
 	{"encodedir", encodedir, METH_VARARGS, "encodedir a path\n"},
 	{"pathencode", pathencode, METH_VARARGS, "fncache-encode a path\n"},
 	{"lowerencode", lowerencode, METH_VARARGS, "lower-encode a path\n"},
 	{NULL, NULL}
 };
 
 void dirs_module_init(PyObject *mod);
 
 static void module_init(PyObject *mod)
 {
 	dirs_module_init(mod);
 
 	indexType.tp_new = PyType_GenericNew;
 	if (PyType_Ready(&indexType) < 0)
 		return;
 	Py_INCREF(&indexType);
 
 	PyModule_AddObject(mod, "index", (PyObject *)&indexType);
 
 	nullentry = Py_BuildValue("iiiiiiis#", 0, 0, 0,
 				  -1, -1, -1, -1, nullid, 20);
 	if (nullentry)
 		PyObject_GC_UnTrack(nullentry);
 
 	dirstate_unset = Py_BuildValue("ciii", 'n', 0, -1, -1);
 }
 
 #ifdef IS_PY3K
 static struct PyModuleDef parsers_module = {
 	PyModuleDef_HEAD_INIT,
 	"parsers",
 	parsers_doc,
 	-1,
 	methods
 };
 
 PyMODINIT_FUNC PyInit_parsers(void)
 {
 	PyObject *mod = PyModule_Create(&parsers_module);
 	module_init(mod);
 	return mod;
 }
 #else
 PyMODINIT_FUNC initparsers(void)
 {
 	PyObject *mod = Py_InitModule3("parsers", methods, parsers_doc);
 	module_init(mod);
 }
 #endif