revlog: use an LRU cache for delta chain bases
Profiling using statprof revealed a hotspot during changegroup
application calculating delta chain bases on generaldelta repos.
Essentially, revlog._addrevision() was performing a lot of redundant
work tracing the delta chain as part of determining when the chain
distance was acceptable. This was most pronounced when adding
revisions to manifests, which can have delta chains thousands of
revisions long.
There was a delta chain base cache on revlogs before, but it only
captured a single revision. This was acceptable before generaldelta,
when _addrevision would build deltas from the previous revision and
thus we'd pretty much guarantee a cache hit when resolving the delta
chain base on a subsequent _addrevision call. However, it isn't
suitable for generaldelta because parent revisions aren't necessarily
the last processed revision.
This patch converts the delta chain base cache to an LRU dict cache.
The cache can hold multiple entries, so generaldelta repos have a
higher chance of getting a cache hit.
The impact of this change when processing changegroup additions is
significant. On a generaldelta conversion of the "mozilla-unified"
repo (which contains heads of the main Firefox repositories in
chronological order - this means there are lots of transitions between
heads in revlog order), this change has the following impact when
performing an `hg unbundle` of an uncompressed bundle of the repo:
before: 5:42 CPU time
after: 4:34 CPU time
Most of this time is saved when applying the changelog and manifest
revlogs:
before: 2:30 CPU time
after: 1:17 CPU time
That nearly a 50% reduction in CPU time applying changesets and
manifests!
Applying a gzipped bundle of the same repo (effectively simulating a
`hg clone` over HTTP) showed a similar speedup:
before: 5:53 CPU time
after: 4:46 CPU time
Wall time improvements were basically the same as CPU time.
I didn't measure explicitly, but it feels like most of the time
is saved when processing manifests. This makes sense, as large
manifests tend to have very long delta chains and thus benefit the
most from this cache.
So, this change effectively makes changegroup application (which is
used by `hg unbundle`, `hg clone`, `hg pull`, `hg unshelve`, and
various other commands) significantly faster when delta chains are
long (which can happen on repos with large numbers of files and thus
large manifests).
In theory, this change can result in more memory utilization. However,
we're caching a dict of ints. At most we have 200 ints + Python object
overhead per revlog. And, the cache is really only populated when
performing read-heavy operations, such as adding changegroups or
scanning an individual revlog. For memory bloat to be an issue, we'd
need to scan/read several revisions from several revlogs all while
having active references to several revlogs. I don't think there are
many operations that do this, so I don't think memory bloat from the
cache will be an issue.
"""automatically manage newlines in repository files
This extension allows you to manage the type of line endings (CRLF or
LF) that are used in the repository and in the local working
directory. That way you can get CRLF line endings on Windows and LF on
Unix/Mac, thereby letting everybody use their OS native line endings.
The extension reads its configuration from a versioned ``.hgeol``
configuration file found in the root of the working directory. The
``.hgeol`` file use the same syntax as all other Mercurial
configuration files. It uses two sections, ``[patterns]`` and
``[repository]``.
The ``[patterns]`` section specifies how line endings should be
converted between the working directory and the repository. The format is
specified by a file pattern. The first match is used, so put more
specific patterns first. The available line endings are ``LF``,
``CRLF``, and ``BIN``.
Files with the declared format of ``CRLF`` or ``LF`` are always
checked out and stored in the repository in that format and files
declared to be binary (``BIN``) are left unchanged. Additionally,
``native`` is an alias for checking out in the platform's default line
ending: ``LF`` on Unix (including Mac OS X) and ``CRLF`` on
Windows. Note that ``BIN`` (do nothing to line endings) is Mercurial's
default behavior; it is only needed if you need to override a later,
more general pattern.
The optional ``[repository]`` section specifies the line endings to
use for files stored in the repository. It has a single setting,
``native``, which determines the storage line endings for files
declared as ``native`` in the ``[patterns]`` section. It can be set to
``LF`` or ``CRLF``. The default is ``LF``. For example, this means
that on Windows, files configured as ``native`` (``CRLF`` by default)
will be converted to ``LF`` when stored in the repository. Files
declared as ``LF``, ``CRLF``, or ``BIN`` in the ``[patterns]`` section
are always stored as-is in the repository.
Example versioned ``.hgeol`` file::
[patterns]
**.py = native
**.vcproj = CRLF
**.txt = native
Makefile = LF
**.jpg = BIN
[repository]
native = LF
.. note::
The rules will first apply when files are touched in the working
directory, e.g. by updating to null and back to tip to touch all files.
The extension uses an optional ``[eol]`` section read from both the
normal Mercurial configuration files and the ``.hgeol`` file, with the
latter overriding the former. You can use that section to control the
overall behavior. There are three settings:
- ``eol.native`` (default ``os.linesep``) can be set to ``LF`` or
``CRLF`` to override the default interpretation of ``native`` for
checkout. This can be used with :hg:`archive` on Unix, say, to
generate an archive where files have line endings for Windows.
- ``eol.only-consistent`` (default True) can be set to False to make
the extension convert files with inconsistent EOLs. Inconsistent
means that there is both ``CRLF`` and ``LF`` present in the file.
Such files are normally not touched under the assumption that they
have mixed EOLs on purpose.
- ``eol.fix-trailing-newline`` (default False) can be set to True to
ensure that converted files end with a EOL character (either ``\\n``
or ``\\r\\n`` as per the configured patterns).
The extension provides ``cleverencode:`` and ``cleverdecode:`` filters
like the deprecated win32text extension does. This means that you can
disable win32text and enable eol and your filters will still work. You
only need to these filters until you have prepared a ``.hgeol`` file.
The ``win32text.forbid*`` hooks provided by the win32text extension
have been unified into a single hook named ``eol.checkheadshook``. The
hook will lookup the expected line endings from the ``.hgeol`` file,
which means you must migrate to a ``.hgeol`` file first before using
the hook. ``eol.checkheadshook`` only checks heads, intermediate
invalid revisions will be pushed. To forbid them completely, use the
``eol.checkallhook`` hook. These hooks are best used as
``pretxnchangegroup`` hooks.
See :hg:`help patterns` for more information about the glob patterns
used.
"""
from __future__ import absolute_import
import os
import re
from mercurial.i18n import _
from mercurial import (
config,
error,
extensions,
match,
util,
)
# Note for extension authors: ONLY specify testedwith = 'internal' for
# extensions which SHIP WITH MERCURIAL. Non-mainline extensions should
# be specifying the version(s) of Mercurial they are tested with, or
# leave the attribute unspecified.
testedwith = 'internal'
# Matches a lone LF, i.e., one that is not part of CRLF.
singlelf = re.compile('(^|[^\r])\n')
# Matches a single EOL which can either be a CRLF where repeated CR
# are removed or a LF. We do not care about old Macintosh files, so a
# stray CR is an error.
eolre = re.compile('\r*\n')
def inconsistenteol(data):
return '\r\n' in data and singlelf.search(data)
def tolf(s, params, ui, **kwargs):
"""Filter to convert to LF EOLs."""
if util.binary(s):
return s
if ui.configbool('eol', 'only-consistent', True) and inconsistenteol(s):
return s
if (ui.configbool('eol', 'fix-trailing-newline', False)
and s and s[-1] != '\n'):
s = s + '\n'
return eolre.sub('\n', s)
def tocrlf(s, params, ui, **kwargs):
"""Filter to convert to CRLF EOLs."""
if util.binary(s):
return s
if ui.configbool('eol', 'only-consistent', True) and inconsistenteol(s):
return s
if (ui.configbool('eol', 'fix-trailing-newline', False)
and s and s[-1] != '\n'):
s = s + '\n'
return eolre.sub('\r\n', s)
def isbinary(s, params):
"""Filter to do nothing with the file."""
return s
filters = {
'to-lf': tolf,
'to-crlf': tocrlf,
'is-binary': isbinary,
# The following provide backwards compatibility with win32text
'cleverencode:': tolf,
'cleverdecode:': tocrlf
}
class eolfile(object):
def __init__(self, ui, root, data):
self._decode = {'LF': 'to-lf', 'CRLF': 'to-crlf', 'BIN': 'is-binary'}
self._encode = {'LF': 'to-lf', 'CRLF': 'to-crlf', 'BIN': 'is-binary'}
self.cfg = config.config()
# Our files should not be touched. The pattern must be
# inserted first override a '** = native' pattern.
self.cfg.set('patterns', '.hg*', 'BIN', 'eol')
# We can then parse the user's patterns.
self.cfg.parse('.hgeol', data)
isrepolf = self.cfg.get('repository', 'native') != 'CRLF'
self._encode['NATIVE'] = isrepolf and 'to-lf' or 'to-crlf'
iswdlf = ui.config('eol', 'native', os.linesep) in ('LF', '\n')
self._decode['NATIVE'] = iswdlf and 'to-lf' or 'to-crlf'
include = []
exclude = []
for pattern, style in self.cfg.items('patterns'):
key = style.upper()
if key == 'BIN':
exclude.append(pattern)
else:
include.append(pattern)
# This will match the files for which we need to care
# about inconsistent newlines.
self.match = match.match(root, '', [], include, exclude)
def copytoui(self, ui):
for pattern, style in self.cfg.items('patterns'):
key = style.upper()
try:
ui.setconfig('decode', pattern, self._decode[key], 'eol')
ui.setconfig('encode', pattern, self._encode[key], 'eol')
except KeyError:
ui.warn(_("ignoring unknown EOL style '%s' from %s\n")
% (style, self.cfg.source('patterns', pattern)))
# eol.only-consistent can be specified in ~/.hgrc or .hgeol
for k, v in self.cfg.items('eol'):
ui.setconfig('eol', k, v, 'eol')
def checkrev(self, repo, ctx, files):
failed = []
for f in (files or ctx.files()):
if f not in ctx:
continue
for pattern, style in self.cfg.items('patterns'):
if not match.match(repo.root, '', [pattern])(f):
continue
target = self._encode[style.upper()]
data = ctx[f].data()
if (target == "to-lf" and "\r\n" in data
or target == "to-crlf" and singlelf.search(data)):
failed.append((f, target, str(ctx)))
break
return failed
def parseeol(ui, repo, nodes):
try:
for node in nodes:
try:
if node is None:
# Cannot use workingctx.data() since it would load
# and cache the filters before we configure them.
data = repo.wfile('.hgeol').read()
else:
data = repo[node]['.hgeol'].data()
return eolfile(ui, repo.root, data)
except (IOError, LookupError):
pass
except error.ParseError as inst:
ui.warn(_("warning: ignoring .hgeol file due to parse error "
"at %s: %s\n") % (inst.args[1], inst.args[0]))
return None
def _checkhook(ui, repo, node, headsonly):
# Get revisions to check and touched files at the same time
files = set()
revs = set()
for rev in xrange(repo[node].rev(), len(repo)):
revs.add(rev)
if headsonly:
ctx = repo[rev]
files.update(ctx.files())
for pctx in ctx.parents():
revs.discard(pctx.rev())
failed = []
for rev in revs:
ctx = repo[rev]
eol = parseeol(ui, repo, [ctx.node()])
if eol:
failed.extend(eol.checkrev(repo, ctx, files))
if failed:
eols = {'to-lf': 'CRLF', 'to-crlf': 'LF'}
msgs = []
for f, target, node in sorted(failed):
msgs.append(_(" %s in %s should not have %s line endings") %
(f, node, eols[target]))
raise error.Abort(_("end-of-line check failed:\n") + "\n".join(msgs))
def checkallhook(ui, repo, node, hooktype, **kwargs):
"""verify that files have expected EOLs"""
_checkhook(ui, repo, node, False)
def checkheadshook(ui, repo, node, hooktype, **kwargs):
"""verify that files have expected EOLs"""
_checkhook(ui, repo, node, True)
# "checkheadshook" used to be called "hook"
hook = checkheadshook
def preupdate(ui, repo, hooktype, parent1, parent2):
repo.loadeol([parent1])
return False
def uisetup(ui):
ui.setconfig('hooks', 'preupdate.eol', preupdate, 'eol')
def extsetup(ui):
try:
extensions.find('win32text')
ui.warn(_("the eol extension is incompatible with the "
"win32text extension\n"))
except KeyError:
pass
def reposetup(ui, repo):
uisetup(repo.ui)
if not repo.local():
return
for name, fn in filters.iteritems():
repo.adddatafilter(name, fn)
ui.setconfig('patch', 'eol', 'auto', 'eol')
class eolrepo(repo.__class__):
def loadeol(self, nodes):
eol = parseeol(self.ui, self, nodes)
if eol is None:
return None
eol.copytoui(self.ui)
return eol.match
def _hgcleardirstate(self):
self._eolfile = self.loadeol([None, 'tip'])
if not self._eolfile:
self._eolfile = util.never
return
try:
cachemtime = os.path.getmtime(self.join("eol.cache"))
except OSError:
cachemtime = 0
try:
eolmtime = os.path.getmtime(self.wjoin(".hgeol"))
except OSError:
eolmtime = 0
if eolmtime > cachemtime:
self.ui.debug("eol: detected change in .hgeol\n")
wlock = None
try:
wlock = self.wlock()
for f in self.dirstate:
if self.dirstate[f] == 'n':
# all normal files need to be looked at
# again since the new .hgeol file might no
# longer match a file it matched before
self.dirstate.normallookup(f)
# Create or touch the cache to update mtime
self.vfs("eol.cache", "w").close()
wlock.release()
except error.LockUnavailable:
# If we cannot lock the repository and clear the
# dirstate, then a commit might not see all files
# as modified. But if we cannot lock the
# repository, then we can also not make a commit,
# so ignore the error.
pass
def commitctx(self, ctx, haserror=False):
for f in sorted(ctx.added() + ctx.modified()):
if not self._eolfile(f):
continue
fctx = ctx[f]
if fctx is None:
continue
data = fctx.data()
if util.binary(data):
# We should not abort here, since the user should
# be able to say "** = native" to automatically
# have all non-binary files taken care of.
continue
if inconsistenteol(data):
raise error.Abort(_("inconsistent newline style "
"in %s\n") % f)
return super(eolrepo, self).commitctx(ctx, haserror)
repo.__class__ = eolrepo
repo._hgcleardirstate()