sapling/mercurial/dirstate.py

826 lines
28 KiB
Python
Raw Normal View History

# dirstate.py - working directory tracking for mercurial
#
# Copyright 2005-2007 Matt Mackall <mpm@selenic.com>
#
# This software may be used and distributed according to the terms of the
2010-01-20 07:20:08 +03:00
# GNU General Public License version 2 or any later version.
import errno
from node import nullid
2006-12-15 05:25:19 +03:00
from i18n import _
import scmutil, util, ignore, osutil, parsers, encoding
import os, stat, errno, gc
2009-04-30 05:47:18 +04:00
propertycache = util.propertycache
filecache = scmutil.filecache
_rangemask = 0x7fffffff
class repocache(filecache):
"""filecache for files in .hg/"""
def join(self, obj, fname):
return obj._opener.join(fname)
2007-06-18 22:24:34 +04:00
class rootcache(filecache):
"""filecache for files in the repository root"""
def join(self, obj, fname):
return obj._join(fname)
def _finddirs(path):
pos = path.rfind('/')
while pos != -1:
yield path[:pos]
pos = path.rfind('/', 0, pos)
def _incdirs(dirs, path):
for base in _finddirs(path):
if base in dirs:
dirs[base] += 1
return
dirs[base] = 1
def _decdirs(dirs, path):
for base in _finddirs(path):
if dirs[base] > 1:
dirs[base] -= 1
return
del dirs[base]
class dirstate(object):
def __init__(self, opener, ui, root, validate):
'''Create a new dirstate object.
opener is an open()-like callable that can be used to open the
dirstate file; root is the root of the directory tracked by
the dirstate.
'''
self._opener = opener
self._validate = validate
self._root = root
self._rootdir = os.path.join(root, '')
self._dirty = False
2007-07-22 01:44:38 +04:00
self._dirtypl = False
self._lastnormaltime = 0
self._ui = ui
2012-03-01 19:39:58 +04:00
self._filecache = {}
2009-04-30 05:47:18 +04:00
@propertycache
def _map(self):
'''Return the dirstate contents as a map from filename to
(state, mode, size, time).'''
2009-04-30 05:47:18 +04:00
self._read()
return self._map
@propertycache
def _copymap(self):
self._read()
return self._copymap
@propertycache
def _foldmap(self):
f = {}
for name in self._map:
f[util.normcase(name)] = name
for name in self._dirs:
f[util.normcase(name)] = name
f['.'] = '.' # prevents useless util.fspath() invocation
2009-04-30 05:47:18 +04:00
return f
@repocache('branch')
2009-04-30 05:47:18 +04:00
def _branch(self):
try:
return self._opener.read("branch").strip() or "default"
except IOError, inst:
if inst.errno != errno.ENOENT:
raise
2009-04-30 05:47:18 +04:00
return "default"
@propertycache
def _pl(self):
try:
fp = self._opener("dirstate")
st = fp.read(40)
fp.close()
l = len(st)
if l == 40:
2009-04-30 05:47:18 +04:00
return st[:20], st[20:40]
elif l > 0 and l < 40:
raise util.Abort(_('working directory state appears damaged!'))
2009-04-30 05:47:18 +04:00
except IOError, err:
2010-01-25 09:05:27 +03:00
if err.errno != errno.ENOENT:
raise
2009-04-30 05:47:18 +04:00
return [nullid, nullid]
@propertycache
def _dirs(self):
dirs = {}
2010-01-25 09:05:27 +03:00
for f, s in self._map.iteritems():
2009-04-30 05:47:18 +04:00
if s[0] != 'r':
_incdirs(dirs, f)
return dirs
def dirs(self):
return self._dirs
@rootcache('.hgignore')
2009-04-30 05:47:18 +04:00
def _ignore(self):
files = [self._join('.hgignore')]
for name, path in self._ui.configitems("ui"):
if name == 'ignore' or name.startswith('ignore.'):
files.append(util.expandpath(path))
2009-04-30 05:47:18 +04:00
return ignore.ignore(self._root, files, self._ui.warn)
@propertycache
def _slash(self):
return self._ui.configbool('ui', 'slash') and os.sep != '/'
@propertycache
def _checklink(self):
return util.checklink(self._root)
@propertycache
def _checkexec(self):
return util.checkexec(self._root)
@propertycache
def _checkcase(self):
return not util.checkcase(self._join('.hg'))
2007-07-22 01:02:09 +04:00
def _join(self, f):
# much faster than os.path.join()
# it's safe because f is always a relative path
return self._rootdir + f
def flagfunc(self, buildfallback):
if self._checklink and self._checkexec:
def f(x):
p = self._join(x)
if os.path.islink(p):
return 'l'
if util.isexec(p):
return 'x'
return ''
return f
fallback = buildfallback()
if self._checklink:
def f(x):
if os.path.islink(self._join(x)):
return 'l'
if 'x' in fallback(x):
return 'x'
return ''
return f
if self._checkexec:
def f(x):
if 'l' in fallback(x):
return 'l'
2011-05-08 22:45:47 +04:00
if util.isexec(self._join(x)):
return 'x'
return ''
return f
else:
return fallback
def getcwd(self):
cwd = os.getcwd()
2010-01-25 09:05:27 +03:00
if cwd == self._root:
return ''
# self._root ends with a path separator if self._root is '/' or 'C:\'
rootsep = self._root
if not util.endswithsep(rootsep):
rootsep += os.sep
if cwd.startswith(rootsep):
return cwd[len(rootsep):]
else:
# we're outside the repo. return an absolute path.
return cwd
def pathto(self, f, cwd=None):
if cwd is None:
cwd = self.getcwd()
path = util.pathto(self._root, cwd, f)
if self._slash:
return util.normpath(path)
return path
def __getitem__(self, key):
'''Return the current state of key (a filename) in the dirstate.
States are:
n normal
m needs merging
r marked for removal
a marked for addition
? not tracked
'''
return self._map.get(key, ("?",))[0]
def __contains__(self, key):
return key in self._map
def __iter__(self):
for x in sorted(self._map):
yield x
def parents(self):
return [self._validate(p) for p in self._pl]
def p1(self):
return self._validate(self._pl[0])
def p2(self):
return self._validate(self._pl[1])
def branch(self):
return encoding.tolocal(self._branch)
def setparents(self, p1, p2=nullid):
"""Set dirstate parents to p1 and p2.
When moving from two parents to one, 'm' merged entries a
adjusted to normal and previous copy records discarded and
returned by the call.
See localrepo.setparents()
"""
2007-07-22 01:44:38 +04:00
self._dirty = self._dirtypl = True
rebase: skip resolved but emptied revisions When rebasing, if a conflict occurs and is resolved in a way the rebased revision becomes empty, it is not skipped, unlike revisions being emptied without conflicts. The reason is: - File 'x' is merged and resolved, merge.update() marks it as 'm' in the dirstate. - rebase.concludenode() calls localrepo.commit(), which calls localrepo.status() which calls dirstate.status(). 'x' shows up as 'm' and is unconditionnally added to the modified files list, instead of being checked again. - localrepo.commit() detects 'x' as changed an create a new revision where only the manifest parents and linkrev differ. Marking 'x' as modified without checking it makes sense for regular merges. But in rebase case, the merge looks normal but the second parent is usually discarded. When this happens, 'm' files in dirstate are a bit irrelevant and should be considered 'n' possibly dirty instead. That is what the current patch does. Another approach, maybe more efficient, would be to pass another flag to merge.update() saying the 'branchmerge' is a bit of a lie and recordupdate() should call dirstate.normallookup() instead of merge(). It is also tempting to add this logic to dirstate.setparents(), moving from two to one parent is what invalidates the 'm' markers. But this is a far bigger change to make. v2: succumb to the temptation and move the logic in dirstate.setparents(). mpm suggested trying _filecommit() first but it is called by commitctx() which knows nothing about the dirstate and comes too late into the game. A second approach was to rewrite the 'm' state into 'n' on the fly in dirstate.status() which failed for graft in the following case: $ hg init repo $ cd repo $ echo a > a $ hg ci -qAm0 $ echo a >> a $ hg ci -m1 $ hg up 0 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg mv a b $ echo c > b $ hg ci -m2 created new head $ hg graft 1 --tool internal:local grafting revision 1 $ hg --config extensions.graphlog= glog --template '{rev} {desc|firstline}\n' @ 3 1 | o 2 2 | | o 1 1 |/ o 0 0 $ hg log -r 3 --debug --patch --git --copies changeset: 3:19cd7d1417952af13161b94c32e901769104560c tag: tip phase: draft parent: 2:b5c505595c9e9a12d5dd457919c143e05fc16fb8 parent: -1:0000000000000000000000000000000000000000 manifest: 3:3d27ce8d02241aa59b60804805edf103c5c0cda4 user: test date: Thu Jan 01 00:00:00 1970 +0000 extra: branch=default extra: source=a03df74c41413a75c0a42997fc36c2de97b26658 description: 1 Here, revision 3 is created because there is a copy record for 'b' in the dirstate and thus 'b' is considered modified. But this information is discarded at commit time since 'b' content is unchanged. I do not know if discarding this information is correct or not, but at this time we cannot represent it anyway. This patch therefore implements the last solution of moving the logic into dirstate.setparents(). It does not sound crazy as 'm' files makes no sense with only one parent. It also makes dirstate.merge() calls .lookupnormal() if there is one parent, to preserve the invariant. I am a bit concerned about introducing this kind of stateful behaviour to existing code which historically treated setparents() as a basic setter without side-effects. And doing that during the code freeze.
2012-04-22 22:06:36 +04:00
oldp2 = self._pl[1]
self._pl = p1, p2
copies = {}
rebase: skip resolved but emptied revisions When rebasing, if a conflict occurs and is resolved in a way the rebased revision becomes empty, it is not skipped, unlike revisions being emptied without conflicts. The reason is: - File 'x' is merged and resolved, merge.update() marks it as 'm' in the dirstate. - rebase.concludenode() calls localrepo.commit(), which calls localrepo.status() which calls dirstate.status(). 'x' shows up as 'm' and is unconditionnally added to the modified files list, instead of being checked again. - localrepo.commit() detects 'x' as changed an create a new revision where only the manifest parents and linkrev differ. Marking 'x' as modified without checking it makes sense for regular merges. But in rebase case, the merge looks normal but the second parent is usually discarded. When this happens, 'm' files in dirstate are a bit irrelevant and should be considered 'n' possibly dirty instead. That is what the current patch does. Another approach, maybe more efficient, would be to pass another flag to merge.update() saying the 'branchmerge' is a bit of a lie and recordupdate() should call dirstate.normallookup() instead of merge(). It is also tempting to add this logic to dirstate.setparents(), moving from two to one parent is what invalidates the 'm' markers. But this is a far bigger change to make. v2: succumb to the temptation and move the logic in dirstate.setparents(). mpm suggested trying _filecommit() first but it is called by commitctx() which knows nothing about the dirstate and comes too late into the game. A second approach was to rewrite the 'm' state into 'n' on the fly in dirstate.status() which failed for graft in the following case: $ hg init repo $ cd repo $ echo a > a $ hg ci -qAm0 $ echo a >> a $ hg ci -m1 $ hg up 0 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg mv a b $ echo c > b $ hg ci -m2 created new head $ hg graft 1 --tool internal:local grafting revision 1 $ hg --config extensions.graphlog= glog --template '{rev} {desc|firstline}\n' @ 3 1 | o 2 2 | | o 1 1 |/ o 0 0 $ hg log -r 3 --debug --patch --git --copies changeset: 3:19cd7d1417952af13161b94c32e901769104560c tag: tip phase: draft parent: 2:b5c505595c9e9a12d5dd457919c143e05fc16fb8 parent: -1:0000000000000000000000000000000000000000 manifest: 3:3d27ce8d02241aa59b60804805edf103c5c0cda4 user: test date: Thu Jan 01 00:00:00 1970 +0000 extra: branch=default extra: source=a03df74c41413a75c0a42997fc36c2de97b26658 description: 1 Here, revision 3 is created because there is a copy record for 'b' in the dirstate and thus 'b' is considered modified. But this information is discarded at commit time since 'b' content is unchanged. I do not know if discarding this information is correct or not, but at this time we cannot represent it anyway. This patch therefore implements the last solution of moving the logic into dirstate.setparents(). It does not sound crazy as 'm' files makes no sense with only one parent. It also makes dirstate.merge() calls .lookupnormal() if there is one parent, to preserve the invariant. I am a bit concerned about introducing this kind of stateful behaviour to existing code which historically treated setparents() as a basic setter without side-effects. And doing that during the code freeze.
2012-04-22 22:06:36 +04:00
if oldp2 != nullid and p2 == nullid:
# Discard 'm' markers when moving away from a merge state
for f, s in self._map.iteritems():
if s[0] == 'm':
if f in self._copymap:
copies[f] = self._copymap[f]
rebase: skip resolved but emptied revisions When rebasing, if a conflict occurs and is resolved in a way the rebased revision becomes empty, it is not skipped, unlike revisions being emptied without conflicts. The reason is: - File 'x' is merged and resolved, merge.update() marks it as 'm' in the dirstate. - rebase.concludenode() calls localrepo.commit(), which calls localrepo.status() which calls dirstate.status(). 'x' shows up as 'm' and is unconditionnally added to the modified files list, instead of being checked again. - localrepo.commit() detects 'x' as changed an create a new revision where only the manifest parents and linkrev differ. Marking 'x' as modified without checking it makes sense for regular merges. But in rebase case, the merge looks normal but the second parent is usually discarded. When this happens, 'm' files in dirstate are a bit irrelevant and should be considered 'n' possibly dirty instead. That is what the current patch does. Another approach, maybe more efficient, would be to pass another flag to merge.update() saying the 'branchmerge' is a bit of a lie and recordupdate() should call dirstate.normallookup() instead of merge(). It is also tempting to add this logic to dirstate.setparents(), moving from two to one parent is what invalidates the 'm' markers. But this is a far bigger change to make. v2: succumb to the temptation and move the logic in dirstate.setparents(). mpm suggested trying _filecommit() first but it is called by commitctx() which knows nothing about the dirstate and comes too late into the game. A second approach was to rewrite the 'm' state into 'n' on the fly in dirstate.status() which failed for graft in the following case: $ hg init repo $ cd repo $ echo a > a $ hg ci -qAm0 $ echo a >> a $ hg ci -m1 $ hg up 0 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg mv a b $ echo c > b $ hg ci -m2 created new head $ hg graft 1 --tool internal:local grafting revision 1 $ hg --config extensions.graphlog= glog --template '{rev} {desc|firstline}\n' @ 3 1 | o 2 2 | | o 1 1 |/ o 0 0 $ hg log -r 3 --debug --patch --git --copies changeset: 3:19cd7d1417952af13161b94c32e901769104560c tag: tip phase: draft parent: 2:b5c505595c9e9a12d5dd457919c143e05fc16fb8 parent: -1:0000000000000000000000000000000000000000 manifest: 3:3d27ce8d02241aa59b60804805edf103c5c0cda4 user: test date: Thu Jan 01 00:00:00 1970 +0000 extra: branch=default extra: source=a03df74c41413a75c0a42997fc36c2de97b26658 description: 1 Here, revision 3 is created because there is a copy record for 'b' in the dirstate and thus 'b' is considered modified. But this information is discarded at commit time since 'b' content is unchanged. I do not know if discarding this information is correct or not, but at this time we cannot represent it anyway. This patch therefore implements the last solution of moving the logic into dirstate.setparents(). It does not sound crazy as 'm' files makes no sense with only one parent. It also makes dirstate.merge() calls .lookupnormal() if there is one parent, to preserve the invariant. I am a bit concerned about introducing this kind of stateful behaviour to existing code which historically treated setparents() as a basic setter without side-effects. And doing that during the code freeze.
2012-04-22 22:06:36 +04:00
self.normallookup(f)
return copies
def setbranch(self, branch):
self._branch = encoding.fromlocal(branch)
2012-04-19 19:11:42 +04:00
f = self._opener('branch', 'w', atomictemp=True)
try:
f.write(self._branch + '\n')
f.close()
# make sure filecache has the correct stat info for _branch after
# replacing the underlying file
ce = self._filecache['_branch']
if ce:
ce.refresh()
except: # re-raises
f.discard()
raise
2007-06-18 22:24:34 +04:00
def _read(self):
self._map = {}
self._copymap = {}
2007-06-18 22:24:33 +04:00
try:
st = self._opener.read("dirstate")
2007-06-18 22:24:33 +04:00
except IOError, err:
2010-01-25 09:05:27 +03:00
if err.errno != errno.ENOENT:
raise
2007-06-18 22:24:33 +04:00
return
if not st:
return
# Python's garbage collector triggers a GC each time a certain number
# of container objects (the number being defined by
# gc.get_threshold()) are allocated. parse_dirstate creates a tuple
# for each file in the dirstate. The C version then immediately marks
# them as not to be tracked by the collector. However, this has no
# effect on when GCs are triggered, only on what objects the GC looks
# into. This means that O(number of files) GCs are unavoidable.
# Depending on when in the process's lifetime the dirstate is parsed,
# this can get very expensive. As a workaround, disable GC while
# parsing the dirstate.
gcenabled = gc.isenabled()
gc.disable()
try:
p = parsers.parse_dirstate(self._map, self._copymap, st)
finally:
if gcenabled:
gc.enable()
if not self._dirtypl:
2008-10-13 00:21:08 +04:00
self._pl = p
def invalidate(self):
for a in ("_map", "_copymap", "_foldmap", "_branch", "_pl", "_dirs",
"_ignore"):
if a in self.__dict__:
delattr(self, a)
self._lastnormaltime = 0
self._dirty = False
def copy(self, source, dest):
"""Mark dest as a copy of source. Unmark dest if source is None."""
if source == dest:
return
self._dirty = True
if source is not None:
self._copymap[dest] = source
elif dest in self._copymap:
del self._copymap[dest]
def copied(self, file):
return self._copymap.get(file, None)
def copies(self):
return self._copymap
def _droppath(self, f):
if self[f] not in "?r" and "_dirs" in self.__dict__:
_decdirs(self._dirs, f)
2007-06-18 22:24:34 +04:00
def _addpath(self, f, state, mode, size, mtime):
oldstate = self[f]
if state == 'a' or oldstate == 'r':
scmutil.checkfilename(f)
if f in self._dirs:
raise util.Abort(_('directory %r already in dirstate') % f)
# shadows
for d in _finddirs(f):
if d in self._dirs:
break
if d in self._map and self[d] != 'r':
raise util.Abort(
_('file %r in dirstate clashes with %r') % (d, f))
if oldstate in "?r" and "_dirs" in self.__dict__:
_incdirs(self._dirs, f)
self._dirty = True
self._map[f] = (state, mode, size, mtime)
def normal(self, f):
'''Mark a file normal and clean.'''
2007-07-22 01:02:09 +04:00
s = os.lstat(self._join(f))
mtime = int(s.st_mtime)
self._addpath(f, 'n', s.st_mode,
s.st_size & _rangemask, mtime & _rangemask)
2008-01-20 16:39:25 +03:00
if f in self._copymap:
del self._copymap[f]
if mtime > self._lastnormaltime:
# Remember the most recent modification timeslot for status(),
# to make sure we won't miss future size-preserving file content
# modifications that happen within the same timeslot.
self._lastnormaltime = mtime
dirstate: avoid a race with multiple commits in the same process (issue2264, issue2516) The race happens when two commits in a row change the same file without changing its size, *if* those two commits happen in the same second in the same process while holding the same repo lock. For example: commit 1: M a M b commit 2: # same process, same second, same repo lock M b # modify b without changing its size M c This first manifested in transplant, which is the most common way to do multiple commits in the same process. But it can manifest in any script or extension that does multiple commits under the same repo lock. (Thus, the test script tests both transplant and a custom script.) The problem was that dirstate.status() failed to notice the change to b when localrepo is about to do the second commit, meaning that change gets left in the working directory. In the context of transplant, that means either a crash ("RuntimeError: nothing committed after transplant") or a silently inaccurate transplant, depending on whether any other files were modified by the second transplanted changeset. The fix is to make status() work a little harder when we have previously marked files as clean (state 'normal') in the same process. Specifically, dirstate.normal() adds files to self._lastnormal, and other state-changing methods remove them. Then dirstate.status() puts any files in self._lastnormal into state 'lookup', which will make localrepository.status() read file contents to see if it has really changed. So we pay a small performance penalty for the second (and subsequent) commits in the same process, without affecting the common case. Anything that does lots of status updates and checks in the same process could suffer a performance hit. Incidentally, there is a simpler fix: call dirstate.normallookup() on every file updated by commit() at the end of the commit. The trouble with that solution is that it imposes a performance penalty on the common case: it means the next status-dependent hg command after every "hg commit" will be a little bit slower. The patch here is more complex, but only affects performance for the uncommon case.
2011-03-21 00:41:09 +03:00
def normallookup(self, f):
'''Mark a file normal, but possibly dirty.'''
if self._pl[1] != nullid and f in self._map:
# if there is a merge going on and the file was either
# in state 'm' (-1) or coming from other parent (-2) before
# being removed, restore that state.
entry = self._map[f]
if entry[0] == 'r' and entry[2] in (-1, -2):
source = self._copymap.get(f)
if entry[2] == -1:
self.merge(f)
elif entry[2] == -2:
self.otherparent(f)
if source:
self.copy(source, f)
return
if entry[0] == 'm' or entry[0] == 'n' and entry[2] == -2:
return
self._addpath(f, 'n', 0, -1, -1)
if f in self._copymap:
del self._copymap[f]
def otherparent(self, f):
'''Mark as coming from the other parent, always dirty.'''
if self._pl[1] == nullid:
raise util.Abort(_("setting %r to other parent "
"only allowed in merges") % f)
self._addpath(f, 'n', 0, -2, -1)
if f in self._copymap:
del self._copymap[f]
def add(self, f):
'''Mark a file added.'''
self._addpath(f, 'a', 0, -1, -1)
if f in self._copymap:
del self._copymap[f]
def remove(self, f):
'''Mark a file removed.'''
self._dirty = True
self._droppath(f)
size = 0
if self._pl[1] != nullid and f in self._map:
# backup the previous state
entry = self._map[f]
if entry[0] == 'm': # merge
size = -1
elif entry[0] == 'n' and entry[2] == -2: # other parent
size = -2
2008-10-13 00:21:08 +04:00
self._map[f] = ('r', 0, size, 0)
if size == 0 and f in self._copymap:
del self._copymap[f]
def merge(self, f):
'''Mark a file merged.'''
rebase: skip resolved but emptied revisions When rebasing, if a conflict occurs and is resolved in a way the rebased revision becomes empty, it is not skipped, unlike revisions being emptied without conflicts. The reason is: - File 'x' is merged and resolved, merge.update() marks it as 'm' in the dirstate. - rebase.concludenode() calls localrepo.commit(), which calls localrepo.status() which calls dirstate.status(). 'x' shows up as 'm' and is unconditionnally added to the modified files list, instead of being checked again. - localrepo.commit() detects 'x' as changed an create a new revision where only the manifest parents and linkrev differ. Marking 'x' as modified without checking it makes sense for regular merges. But in rebase case, the merge looks normal but the second parent is usually discarded. When this happens, 'm' files in dirstate are a bit irrelevant and should be considered 'n' possibly dirty instead. That is what the current patch does. Another approach, maybe more efficient, would be to pass another flag to merge.update() saying the 'branchmerge' is a bit of a lie and recordupdate() should call dirstate.normallookup() instead of merge(). It is also tempting to add this logic to dirstate.setparents(), moving from two to one parent is what invalidates the 'm' markers. But this is a far bigger change to make. v2: succumb to the temptation and move the logic in dirstate.setparents(). mpm suggested trying _filecommit() first but it is called by commitctx() which knows nothing about the dirstate and comes too late into the game. A second approach was to rewrite the 'm' state into 'n' on the fly in dirstate.status() which failed for graft in the following case: $ hg init repo $ cd repo $ echo a > a $ hg ci -qAm0 $ echo a >> a $ hg ci -m1 $ hg up 0 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg mv a b $ echo c > b $ hg ci -m2 created new head $ hg graft 1 --tool internal:local grafting revision 1 $ hg --config extensions.graphlog= glog --template '{rev} {desc|firstline}\n' @ 3 1 | o 2 2 | | o 1 1 |/ o 0 0 $ hg log -r 3 --debug --patch --git --copies changeset: 3:19cd7d1417952af13161b94c32e901769104560c tag: tip phase: draft parent: 2:b5c505595c9e9a12d5dd457919c143e05fc16fb8 parent: -1:0000000000000000000000000000000000000000 manifest: 3:3d27ce8d02241aa59b60804805edf103c5c0cda4 user: test date: Thu Jan 01 00:00:00 1970 +0000 extra: branch=default extra: source=a03df74c41413a75c0a42997fc36c2de97b26658 description: 1 Here, revision 3 is created because there is a copy record for 'b' in the dirstate and thus 'b' is considered modified. But this information is discarded at commit time since 'b' content is unchanged. I do not know if discarding this information is correct or not, but at this time we cannot represent it anyway. This patch therefore implements the last solution of moving the logic into dirstate.setparents(). It does not sound crazy as 'm' files makes no sense with only one parent. It also makes dirstate.merge() calls .lookupnormal() if there is one parent, to preserve the invariant. I am a bit concerned about introducing this kind of stateful behaviour to existing code which historically treated setparents() as a basic setter without side-effects. And doing that during the code freeze.
2012-04-22 22:06:36 +04:00
if self._pl[1] == nullid:
return self.normallookup(f)
2007-07-22 01:02:09 +04:00
s = os.lstat(self._join(f))
self._addpath(f, 'm', s.st_mode,
s.st_size & _rangemask, int(s.st_mtime) & _rangemask)
if f in self._copymap:
del self._copymap[f]
def drop(self, f):
'''Drop a file from the dirstate'''
if f in self._map:
self._dirty = True
self._droppath(f)
del self._map[f]
def _normalize(self, path, isknown, ignoremissing=False, exists=None):
normed = util.normcase(path)
folded = self._foldmap.get(normed, None)
if folded is None:
if isknown:
folded = path
else:
if exists is None:
exists = os.path.lexists(os.path.join(self._root, path))
if not exists:
# Maybe a path component exists
if not ignoremissing and '/' in path:
d, f = path.rsplit('/', 1)
d = self._normalize(d, isknown, ignoremissing, None)
folded = d + "/" + f
else:
# No path components, preserve original case
folded = path
else:
# recursively normalize leading directory components
# against dirstate
if '/' in normed:
d, f = normed.rsplit('/', 1)
d = self._normalize(d, isknown, ignoremissing, True)
r = self._root + "/" + d
folded = d + "/" + util.fspath(f, r)
else:
folded = util.fspath(normed, self._root)
self._foldmap[normed] = folded
return folded
def normalize(self, path, isknown=False, ignoremissing=False):
'''
normalize the case of a pathname when on a casefolding filesystem
isknown specifies whether the filename came from walking the
disk, to avoid extra filesystem access.
If ignoremissing is True, missing path are returned
unchanged. Otherwise, we try harder to normalize possibly
existing path components.
The normalized case is determined based on the following precedence:
- version of name already stored in the dirstate
- version of name stored on disk
- version provided via command arguments
'''
if self._checkcase:
return self._normalize(path, isknown, ignoremissing)
return path
def clear(self):
self._map = {}
if "_dirs" in self.__dict__:
2010-02-08 17:36:34 +03:00
delattr(self, "_dirs")
self._copymap = {}
self._pl = [nullid, nullid]
self._lastnormaltime = 0
2007-08-06 08:00:10 +04:00
self._dirty = True
def rebuild(self, parent, allfiles, changedfiles=None):
changedfiles = changedfiles or allfiles
oldmap = self._map
self.clear()
for f in allfiles:
if f not in changedfiles:
self._map[f] = oldmap[f]
else:
if 'x' in allfiles.flags(f):
self._map[f] = ('n', 0777, -1, 0)
else:
self._map[f] = ('n', 0666, -1, 0)
self._pl = (parent, nullid)
self._dirty = True
def write(self):
2007-06-18 22:24:34 +04:00
if not self._dirty:
return
st = self._opener("dirstate", "w", atomictemp=True)
def finish(s):
st.write(s)
st.close()
self._lastnormaltime = 0
self._dirty = self._dirtypl = False
# use the modification time of the newly created temporary file as the
# filesystem's notion of 'now'
now = util.fstat(st).st_mtime
finish(parsers.pack_dirstate(self._map, self._copymap, self._pl, now))
def _dirignore(self, f):
if f == '.':
return False
if self._ignore(f):
return True
for p in _finddirs(f):
if self._ignore(p):
return True
return False
def walk(self, match, subrepos, unknown, ignored):
'''
Walk recursively through the directory tree, finding all files
matched by match.
Return a dict mapping filename to stat-like object (either
mercurial.osutil.stat instance or return value of os.stat()).
'''
def fwarn(f, msg):
self._ui.warn('%s: %s\n' % (self.pathto(f), msg))
return False
def badtype(mode):
2009-05-08 09:54:00 +04:00
kind = _('unknown')
2010-01-25 09:05:27 +03:00
if stat.S_ISCHR(mode):
kind = _('character device')
elif stat.S_ISBLK(mode):
kind = _('block device')
elif stat.S_ISFIFO(mode):
kind = _('fifo')
elif stat.S_ISSOCK(mode):
kind = _('socket')
elif stat.S_ISDIR(mode):
kind = _('directory')
return _('unsupported file type (type is %s)') % kind
2007-06-18 22:24:34 +04:00
ignore = self._ignore
dirignore = self._dirignore
if ignored:
ignore = util.never
dirignore = util.never
elif not unknown:
# if unknown and ignored are False, skip step 2
ignore = util.always
dirignore = util.always
matchfn = match.matchfn
2009-06-01 02:54:18 +04:00
badfn = match.bad
dmap = self._map
normpath = util.normpath
listdir = osutil.listdir
lstat = os.lstat
getkind = stat.S_IFMT
dirkind = stat.S_IFDIR
regkind = stat.S_IFREG
lnkkind = stat.S_IFLNK
join = self._join
work = []
wadd = work.append
exact = skipstep3 = False
if matchfn == match.exact: # match.exact
exact = True
dirignore = util.always # skip step 2
elif match.files() and not match.anypats(): # match.match, no patterns
skipstep3 = True
if not exact and self._checkcase:
normalize = self._normalize
skipstep3 = False
else:
normalize = None
files = sorted(match.files())
subrepos.sort()
i, j = 0, 0
while i < len(files) and j < len(subrepos):
subpath = subrepos[j] + "/"
if files[i] < subpath:
i += 1
continue
while i < len(files) and files[i].startswith(subpath):
del files[i]
j += 1
if not files or '.' in files:
files = ['']
results = dict.fromkeys(subrepos)
results['.hg'] = None
# step 1: find all explicit files
for ff in files:
if normalize:
nf = normalize(normpath(ff), False, True)
else:
nf = normpath(ff)
if nf in results:
continue
try:
st = lstat(join(nf))
kind = getkind(st.st_mode)
if kind == dirkind:
skipstep3 = False
if nf in dmap:
2009-05-28 01:39:41 +04:00
#file deleted on disk but still in dirstate
results[nf] = None
match.dir(nf)
if not dirignore(nf):
wadd(nf)
2010-09-24 21:46:54 +04:00
elif kind == regkind or kind == lnkkind:
results[nf] = st
else:
badfn(ff, badtype(kind))
if nf in dmap:
results[nf] = None
except OSError, inst:
2009-06-01 02:54:18 +04:00
if nf in dmap: # does it exactly match a file?
results[nf] = None
else: # does it match a directory?
prefix = nf + "/"
for fn in dmap:
if fn.startswith(prefix):
match.dir(nf)
skipstep3 = False
2009-06-01 02:54:18 +04:00
break
else:
badfn(ff, inst.strerror)
# step 2: visit subdirectories
while work:
nd = work.pop()
skip = None
if nd == '.':
nd = ''
else:
skip = '.hg'
try:
entries = listdir(join(nd), stat=True, skip=skip)
except OSError, inst:
if inst.errno in (errno.EACCES, errno.ENOENT):
fwarn(nd, inst.strerror)
continue
raise
for f, kind, st in entries:
if normalize:
nf = normalize(nd and (nd + "/" + f) or f, True, True)
else:
nf = nd and (nd + "/" + f) or f
if nf not in results:
if kind == dirkind:
if not ignore(nf):
match.dir(nf)
wadd(nf)
if nf in dmap and matchfn(nf):
results[nf] = None
elif kind == regkind or kind == lnkkind:
if nf in dmap:
if matchfn(nf):
results[nf] = st
elif matchfn(nf) and not ignore(nf):
results[nf] = st
elif nf in dmap and matchfn(nf):
results[nf] = None
# step 3: report unseen items in the dmap hash
if not skipstep3 and not exact:
visit = sorted([f for f in dmap if f not in results and matchfn(f)])
if unknown:
# unknown == True means we walked the full directory tree above.
# So if a file is not seen it was either a) not matching matchfn
# b) ignored, c) missing, or d) under a symlink directory.
audit_path = scmutil.pathauditor(self._root)
for nf in iter(visit):
# Report ignored items in the dmap as long as they are not
# under a symlink directory.
if ignore(nf) and audit_path.check(nf):
try:
results[nf] = lstat(join(nf))
except OSError:
# file doesn't exist
results[nf] = None
else:
# It's either missing or under a symlink directory
results[nf] = None
else:
# We may not have walked the full directory tree above,
# so stat everything we missed.
nf = iter(visit).next
for st in util.statfiles([join(i) for i in visit]):
results[nf()] = st
for s in subrepos:
del results[s]
del results['.hg']
return results
def status(self, match, subrepos, ignored, clean, unknown):
'''Determine the status of the working copy relative to the
dirstate and return a tuple of lists (unsure, modified, added,
removed, deleted, unknown, ignored, clean), where:
unsure:
files that might have been modified since the dirstate was
written, but need to be read to be sure (size is the same
but mtime differs)
modified:
files that have definitely been modified since the dirstate
was written (different size or mode)
added:
files that have been explicitly added with hg add
removed:
files that have been explicitly removed with hg remove
deleted:
files that have been deleted through other means ("missing")
unknown:
files not in the dirstate that are not ignored
ignored:
files not in the dirstate that are ignored
(by _dirignore())
clean:
files that have definitely not been modified since the
dirstate was written
'''
2008-06-26 23:35:50 +04:00
listignored, listclean, listunknown = ignored, clean, unknown
lookup, modified, added, unknown, ignored = [], [], [], [], []
removed, deleted, clean = [], [], []
dmap = self._map
ladd = lookup.append # aka "unsure"
madd = modified.append
aadd = added.append
uadd = unknown.append
iadd = ignored.append
radd = removed.append
dadd = deleted.append
cadd = clean.append
mexact = match.exact
dirignore = self._dirignore
checkexec = self._checkexec
checklink = self._checklink
copymap = self._copymap
lastnormaltime = self._lastnormaltime
dirstate: ignore symlinks when fs cannot handle them (issue1888) When the filesystem cannot handle the executable bit, we currently ignore it completely when looking for modified files. Similarly, it is impossible to set or clear the bit when the filesystem ignores it. This patch makes Mercurial treat symbolic links the same way. Symlinks are a little different since they manifest themselves as small files containing a filename (the symlink target). On Windows, these files show up as regular files, and on Linux and Mac they show up as real symlinks. Issue1888 presents a case where the symlink files are better ignored from the Windows side. A Linux client creates symlinks in a working copy which is shared over a network between Linux and Windows clients. The Samba server is helpful and defererences the symlink when the Windows client looks at it. This means that Mercurial on the Windows side sees file content instead of a file name in the symlink, and hence flags the link as modified. Ignoring the change would be much more helpful, similarly to how Mercurial does not report any changes when executable bits are ignored in a checkout on Windows. An initial checkout of a symbolic link on a file system that cannot handle symbolic links will still result in a regular file containing the target file name as its content. Sharing such a checkout with a Linux client will not turn the file into a symlink automatically, but 'hg revert' can fix that. After the revert, the Windows client will see the correct file content (provided by the Samba server when it follows the link on the Linux side) and otherwise ignore the change. Running 'hg perfstatus' 10 times gives these results: Before: After: min: 0.544703 min: 0.546549 med: 0.547592 med: 0.548881 avg: 0.549146 avg: 0.548549 max: 0.564112 max: 0.551504 The median time is increased about 0.24%.
2010-08-09 17:31:56 +04:00
lnkkind = stat.S_IFLNK
for fn, st in self.walk(match, subrepos, listunknown,
listignored).iteritems():
2008-05-12 20:37:08 +04:00
if fn not in dmap:
if (listignored or mexact(fn)) and dirignore(fn):
2008-06-26 23:35:50 +04:00
if listignored:
iadd(fn)
2008-06-26 23:35:50 +04:00
elif listunknown:
uadd(fn)
continue
2008-05-12 20:37:08 +04:00
2008-10-13 00:21:08 +04:00
state, mode, size, time = dmap[fn]
2008-05-12 20:37:08 +04:00
if not st and state in "nma":
dadd(fn)
elif state == 'n':
dirstate: ignore symlinks when fs cannot handle them (issue1888) When the filesystem cannot handle the executable bit, we currently ignore it completely when looking for modified files. Similarly, it is impossible to set or clear the bit when the filesystem ignores it. This patch makes Mercurial treat symbolic links the same way. Symlinks are a little different since they manifest themselves as small files containing a filename (the symlink target). On Windows, these files show up as regular files, and on Linux and Mac they show up as real symlinks. Issue1888 presents a case where the symlink files are better ignored from the Windows side. A Linux client creates symlinks in a working copy which is shared over a network between Linux and Windows clients. The Samba server is helpful and defererences the symlink when the Windows client looks at it. This means that Mercurial on the Windows side sees file content instead of a file name in the symlink, and hence flags the link as modified. Ignoring the change would be much more helpful, similarly to how Mercurial does not report any changes when executable bits are ignored in a checkout on Windows. An initial checkout of a symbolic link on a file system that cannot handle symbolic links will still result in a regular file containing the target file name as its content. Sharing such a checkout with a Linux client will not turn the file into a symlink automatically, but 'hg revert' can fix that. After the revert, the Windows client will see the correct file content (provided by the Samba server when it follows the link on the Linux side) and otherwise ignore the change. Running 'hg perfstatus' 10 times gives these results: Before: After: min: 0.544703 min: 0.546549 med: 0.547592 med: 0.548881 avg: 0.549146 avg: 0.548549 max: 0.564112 max: 0.551504 The median time is increased about 0.24%.
2010-08-09 17:31:56 +04:00
# The "mode & lnkkind != lnkkind or self._checklink"
# lines are an expansion of "islink => checklink"
# where islink means "is this a link?" and checklink
# means "can we check links?".
mtime = int(st.st_mtime)
if (size >= 0 and
((size != st.st_size and size != st.st_size & _rangemask)
or ((mode ^ st.st_mode) & 0100 and checkexec))
and (mode & lnkkind != lnkkind or checklink)
or size == -2 # other parent
or fn in copymap):
madd(fn)
elif ((time != mtime and time != mtime & _rangemask)
and (mode & lnkkind != lnkkind or checklink)):
ladd(fn)
elif mtime == lastnormaltime:
# fn may have been changed in the same timeslot without
# changing its size. This can happen if we quickly do
# multiple commits in a single transaction.
# Force lookup, so we don't miss such a racy file change.
dirstate: avoid a race with multiple commits in the same process (issue2264, issue2516) The race happens when two commits in a row change the same file without changing its size, *if* those two commits happen in the same second in the same process while holding the same repo lock. For example: commit 1: M a M b commit 2: # same process, same second, same repo lock M b # modify b without changing its size M c This first manifested in transplant, which is the most common way to do multiple commits in the same process. But it can manifest in any script or extension that does multiple commits under the same repo lock. (Thus, the test script tests both transplant and a custom script.) The problem was that dirstate.status() failed to notice the change to b when localrepo is about to do the second commit, meaning that change gets left in the working directory. In the context of transplant, that means either a crash ("RuntimeError: nothing committed after transplant") or a silently inaccurate transplant, depending on whether any other files were modified by the second transplanted changeset. The fix is to make status() work a little harder when we have previously marked files as clean (state 'normal') in the same process. Specifically, dirstate.normal() adds files to self._lastnormal, and other state-changing methods remove them. Then dirstate.status() puts any files in self._lastnormal into state 'lookup', which will make localrepository.status() read file contents to see if it has really changed. So we pay a small performance penalty for the second (and subsequent) commits in the same process, without affecting the common case. Anything that does lots of status updates and checks in the same process could suffer a performance hit. Incidentally, there is a simpler fix: call dirstate.normallookup() on every file updated by commit() at the end of the commit. The trouble with that solution is that it imposes a performance penalty on the common case: it means the next status-dependent hg command after every "hg commit" will be a little bit slower. The patch here is more complex, but only affects performance for the uncommon case.
2011-03-21 00:41:09 +03:00
ladd(fn)
2008-06-26 23:35:50 +04:00
elif listclean:
cadd(fn)
2008-05-12 20:37:08 +04:00
elif state == 'm':
madd(fn)
2008-05-12 20:37:08 +04:00
elif state == 'a':
aadd(fn)
2008-05-12 20:37:08 +04:00
elif state == 'r':
radd(fn)
return (lookup, modified, added, removed, deleted, unknown, ignored,
clean)