annotate mercurial/changelog.py @ 30155:b7a966ce89ed

changelog: disable delta chains

This patch disables delta chains on changelogs. After this patch, new
entries on changelogs - including existing changelogs - will be stored
as the fulltext of that data (likely compressed). No delta computation
will be performed.

An overview of delta chains and data justifying this change follows.

Revlogs try to store entries as a delta against a previous entry (either
a parent revision in the case of generaldelta or the previous physical
revision when not using generaldelta). Most of the time this is the
correct thing to do: it frequently results in less CPU usage and smaller
storage.

Delta chains are most effective when the base revision being deltad
against is similar to the current data. This tends to occur naturally
for manifests and file data, since only small parts of each tend to
change with each revision. Changelogs, however, are a different story.

Changelog entries represent changesets/commits. And unless commits in a
repository are homogonous (same author, changing same files, similar
commit messages, etc), a delta from one entry to the next tends to be
relatively large compared to the size of the entry. This means that
delta chains tend to be short. How short? Here is the full vs delta
revision breakdown on some real world repos:

Repo             % Full    % Delta   Max Length
hg                45.8       54.2        6
mozilla-central   42.4       57.6        8
mozilla-unified   42.5       57.5       17
pypy              46.1       53.9        6
python-zstandard  46.1       53.9        3

(I threw in python-zstandard as an example of a repo that is homogonous.
It contains a small Python project with changes all from the same
author.)

Contrast this with the manifest revlog for these repos, where 99+% of
revisions are deltas and delta chains run into the thousands.

So delta chains aren't as useful on changelogs. But even a short delta
chain may provide benefits. Let's measure that.

Delta chains may require less CPU to read revisions if the CPU time
spent reading smaller deltas is less than the CPU time used to
decompress larger individual entries. We can measure this via
`hg perfrevlog -c -d 1` to iterate a revlog to resolve each revision's
fulltext. Here are the results of that command on a repo using delta
chains in its changelog and on a repo without delta chains:

hg (forward)
! wall 0.407008 comb 0.410000 user 0.410000 sys 0.000000 (best of 25)
! wall 0.390061 comb 0.390000 user 0.390000 sys 0.000000 (best of 26)

hg (reverse)
! wall 0.515221 comb 0.520000 user 0.520000 sys 0.000000 (best of 19)
! wall 0.400018 comb 0.400000 user 0.390000 sys 0.010000 (best of 25)

mozilla-central (forward)
! wall 4.508296 comb 4.490000 user 4.490000 sys 0.000000 (best of 3)
! wall 4.370222 comb 4.370000 user 4.350000 sys 0.020000 (best of 3)

mozilla-central (reverse)
! wall 5.758995 comb 5.760000 user 5.720000 sys 0.040000 (best of 3)
! wall 4.346503 comb 4.340000 user 4.320000 sys 0.020000 (best of 3)

mozilla-unified (forward)
! wall 4.957088 comb 4.950000 user 4.940000 sys 0.010000 (best of 3)
! wall 4.660528 comb 4.650000 user 4.630000 sys 0.020000 (best of 3)

mozilla-unified (reverse)
! wall 6.119827 comb 6.110000 user 6.090000 sys 0.020000 (best of 3)
! wall 4.675136 comb 4.670000 user 4.670000 sys 0.000000 (best of 3)

pypy (forward)
! wall 1.231122 comb 1.240000 user 1.230000 sys 0.010000 (best of 8)
! wall 1.164896 comb 1.160000 user 1.160000 sys 0.000000 (best of 9)

pypy (reverse)
! wall 1.467049 comb 1.460000 user 1.460000 sys 0.000000 (best of 7)
! wall 1.160200 comb 1.170000 user 1.160000 sys 0.010000 (best of 9)

The data clearly shows that it takes less wall and CPU time to resolve
revisions when there are no delta chains in the changelogs, regardless
of the direction of traversal. Furthermore, not using a delta chain
means that fulltext resolution in reverse is as fast as iterating
forward. So not using delta chains on the changelog is a clear CPU win
for reading operations.

An example of a user-visible operation showing this speed-up is revset
evaluation. Here are results for
`hg perfrevset 'author(gps) or author(mpm)'`:

hg
! wall 1.655506 comb 1.660000 user 1.650000 sys 0.010000 (best of 6)
! wall 1.612723 comb 1.610000 user 1.600000 sys 0.010000 (best of 7)

mozilla-central
! wall 17.629826 comb 17.640000 user 17.600000 sys 0.040000 (best of 3)
! wall 17.311033 comb 17.300000 user 17.260000 sys 0.040000 (best of 3)

What about 00changelog.i size?

Repo                Delta Chains     No Delta Chains
hg                    7,033,250         6,976,771
mozilla-central      82,978,748        81,574,623
mozilla-unified      88,112,349        86,702,162
pypy                 20,740,699        20,659,741

The data shows that removing delta chains from the changelog makes the
changelog smaller.

Delta chains are also used during changegroup generation. This
operation essentially converts a series of revisions to one large
delta chain. And changegroup generation is smart: if the delta in
the revlog matches what the changegroup is emitting, it will reuse
the delta instead of recalculating it. We can measure the impact
removing changelog delta chains has on changegroup generation via
`hg perfchangegroupchangelog`:

hg
! wall 1.589245 comb 1.590000 user 1.590000 sys 0.000000 (best of 7)
! wall 1.788060 comb 1.790000 user 1.790000 sys 0.000000 (best of 6)

mozilla-central
! wall 17.382585 comb 17.380000 user 17.340000 sys 0.040000 (best of 3)
! wall 20.161357 comb 20.160000 user 20.120000 sys 0.040000 (best of 3)

mozilla-unified
! wall 18.722839 comb 18.720000 user 18.680000 sys 0.040000 (best of 3)
! wall 21.168075 comb 21.170000 user 21.130000 sys 0.040000 (best of 3)

pypy
! wall 4.828317 comb 4.830000 user 4.820000 sys 0.010000 (best of 3)
! wall 5.415455 comb 5.420000 user 5.410000 sys 0.010000 (best of 3)

The data shows eliminating delta chains makes the changelog part of
changegroup generation slower. This is expected since we now have to
compute deltas for revisions where we could recycle the delta before.

It is worth putting this regression into context of overall changegroup
times. Here is the rough total CPU time spent in changegroup generation
for various repos while using delta chains on the changelog:

Repo              CPU Time (s)    CPU Time w/ compression
hg                  4.50              7.05
mozilla-central   111.1             222.0
pypy               28.68             75.5

Before compression, removing delta chains from the changegroup adds
~4.4% overhead to hg changegroup generation, 1.3% to mozilla-central,
and 2.0% to pypy. When you factor in zlib compression, these percentages
are roughly divided by 2.

While the increased CPU usage for changegroup generation is unfortunate,
I think it is acceptable because the percentage is small, server
operators (those likely impacted most by this) have other mechanisms
to mitigate CPU consumption (namely reducing zlib compression level and
pre-generated clone bundles), and because there is room to optimize this
in the future. For example, we could use the nullid as the base revision,
effectively encoding the full revision for each entry in the changegroup.
When doing this, `hg perfchangegroupchangelog` nearly halves:

mozilla-unified
! wall 21.168075 comb 21.170000 user 21.130000 sys 0.040000 (best of 3)
! wall 11.196461 comb 11.200000 user 11.190000 sys 0.010000 (best of 3)

This looks very promising as a future optimization opportunity.

It's worth that the changes in test-acl.t to the changegroup part size.
This is because revision 6 in the changegroup had a delta chain of
length 2 before and after this patch the base revision is nullrev.
When the base revision is nullrev, cg2packer.deltaparent() hardcodes
the *previous* revision from the changegroup as the delta parent.
This caused the delta in the changegroup to switch base revisions,
the delta to change, and the size to change accordingly. While the
size increased in this case, I think sizes will remain the same
on average, as the delta base for changelog revisions doesn't matter
too much (as this patch shows). So, I don't consider this a regression.

# changelog.py - changelog class for mercurial
#
# 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.

from __future__ import absolute_import

import collections

from .i18n import _
from .node import (
    bin,
    hex,
    nullid,
)

from . import (
    encoding,
    error,
    revlog,
    util,
)

_defaultextra = {'branch': 'default'}

def _string_escape(text):
    """
    >>> d = {'nl': chr(10), 'bs': chr(92), 'cr': chr(13), 'nul': chr(0)}
    >>> s = "ab%(nl)scd%(bs)s%(bs)sn%(nul)sab%(cr)scd%(bs)s%(nl)s" % d
    >>> s
    'ab\\ncd\\\\\\\\n\\x00ab\\rcd\\\\\\n'
    >>> res = _string_escape(s)
    >>> s == res.decode('string_escape')
    True
    """
    # subset of the string_escape codec
    text = text.replace('\\', '\\\\').replace('\n', '\\n').replace('\r', '\\r')
    return text.replace('\0', '\\0')

def decodeextra(text):
    """
    >>> sorted(decodeextra(encodeextra({'foo': 'bar', 'baz': chr(0) + '2'})
    ...                    ).iteritems())
    [('baz', '\\x002'), ('branch', 'default'), ('foo', 'bar')]
    >>> sorted(decodeextra(encodeextra({'foo': 'bar',
    ...                                 'baz': chr(92) + chr(0) + '2'})
    ...                    ).iteritems())
    [('baz', '\\\\\\x002'), ('branch', 'default'), ('foo', 'bar')]
    """
    extra = _defaultextra.copy()
    for l in text.split('\0'):
        if l:
            if '\\0' in l:
                # fix up \0 without getting into trouble with \\0
                l = l.replace('\\\\', '\\\\\n')
                l = l.replace('\\0', '\0')
                l = l.replace('\n', '')
            k, v = l.decode('string_escape').split(':', 1)
            extra[k] = v
    return extra

def encodeextra(d):
    # keys must be sorted to produce a deterministic changelog entry
    items = [_string_escape('%s:%s' % (k, d[k])) for k in sorted(d)]
    return "\0".join(items)

def stripdesc(desc):
    """strip trailing whitespace and leading and trailing empty lines"""
    return '\n'.join([l.rstrip() for l in desc.splitlines()]).strip('\n')

class appender(object):
    '''the changelog index must be updated last on disk, so we use this class
    to delay writes to it'''
    def __init__(self, vfs, name, mode, buf):
        self.data = buf
        fp = vfs(name, mode)
        self.fp = fp
        self.offset = fp.tell()
        self.size = vfs.fstat(fp).st_size

    def end(self):
        return self.size + len("".join(self.data))
    def tell(self):
        return self.offset
    def flush(self):
        pass
    def close(self):
        self.fp.close()

    def seek(self, offset, whence=0):
        '''virtual file offset spans real file and data'''
        if whence == 0:
            self.offset = offset
        elif whence == 1:
            self.offset += offset
        elif whence == 2:
            self.offset = self.end() + offset
        if self.offset < self.size:
            self.fp.seek(self.offset)

    def read(self, count=-1):
        '''only trick here is reads that span real file and data'''
        ret = ""
        if self.offset < self.size:
            s = self.fp.read(count)
            ret = s
            self.offset += len(s)
            if count > 0:
                count -= len(s)
        if count != 0:
            doff = self.offset - self.size
            self.data.insert(0, "".join(self.data))
            del self.data[1:]
            s = self.data[0][doff:doff + count]
            self.offset += len(s)
            ret += s
        return ret

    def write(self, s):
        self.data.append(str(s))
        self.offset += len(s)

def _divertopener(opener, target):
    """build an opener that writes in 'target.a' instead of 'target'"""
    def _divert(name, mode='r', checkambig=False):
        if name != target:
            return opener(name, mode)
        return opener(name + ".a", mode)
    return _divert

def _delayopener(opener, target, buf):
    """build an opener that stores chunks in 'buf' instead of 'target'"""
    def _delay(name, mode='r', checkambig=False):
        if name != target:
            return opener(name, mode)
        return appender(opener, name, mode, buf)
    return _delay

_changelogrevision = collections.namedtuple(u'changelogrevision',
                                            (u'manifest', u'user', u'date',
                                             u'files', u'description',
                                             u'extra'))

class changelogrevision(object):
    """Holds results of a parsed changelog revision.

    Changelog revisions consist of multiple pieces of data, including
    the manifest node, user, and date. This object exposes a view into
    the parsed object.
    """

    __slots__ = (
        u'_offsets',
        u'_text',
    )

    def __new__(cls, text):
        if not text:
            return _changelogrevision(
                manifest=nullid,
                user='',
                date=(0, 0),
                files=[],
                description='',
                extra=_defaultextra,
            )

        self = super(changelogrevision, cls).__new__(cls)
        # We could return here and implement the following as an __init__.
        # But doing it here is equivalent and saves an extra function call.

        # format used:
        # nodeid\n        : manifest node in ascii
        # user\n          : user, no \n or \r allowed
        # time tz extra\n : date (time is int or float, timezone is int)
        #                 : extra is metadata, encoded and separated by '\0'
        #                 : older versions ignore it
        # files\n\n       : files modified by the cset, no \n or \r allowed
        # (.*)            : comment (free text, ideally utf-8)
        #
        # changelog v0 doesn't use extra

        nl1 = text.index('\n')
        nl2 = text.index('\n', nl1 + 1)
        nl3 = text.index('\n', nl2 + 1)

        # The list of files may be empty. Which means nl3 is the first of the
        # double newline that precedes the description.
        if text[nl3 + 1] == '\n':
            doublenl = nl3
        else:
            doublenl = text.index('\n\n', nl3 + 1)

        self._offsets = (nl1, nl2, nl3, doublenl)
        self._text = text

        return self

    @property
    def manifest(self):
        return bin(self._text[0:self._offsets[0]])

    @property
    def user(self):
        off = self._offsets
        return encoding.tolocal(self._text[off[0] + 1:off[1]])

    @property
    def _rawdate(self):
        off = self._offsets
        dateextra = self._text[off[1] + 1:off[2]]
        return dateextra.split(' ', 2)[0:2]

    @property
    def _rawextra(self):
        off = self._offsets
        dateextra = self._text[off[1] + 1:off[2]]
        fields = dateextra.split(' ', 2)
        if len(fields) != 3:
            return None

        return fields[2]

    @property
    def date(self):
        raw = self._rawdate
        time = float(raw[0])
        # Various tools did silly things with the timezone.
        try:
            timezone = int(raw[1])
        except ValueError:
            timezone = 0

        return time, timezone

    @property
    def extra(self):
        raw = self._rawextra
        if raw is None:
            return _defaultextra

        return decodeextra(raw)

    @property
    def files(self):
        off = self._offsets
        if off[2] == off[3]:
            return []

        return self._text[off[2] + 1:off[3]].split('\n')

    @property
    def description(self):
        return encoding.tolocal(self._text[self._offsets[3] + 2:])

class changelog(revlog.revlog):
    def __init__(self, opener):
        revlog.revlog.__init__(self, opener, "00changelog.i",
                               checkambig=True)
        if self._initempty:
            # changelogs don't benefit from generaldelta
            self.version &= ~revlog.REVLOGGENERALDELTA
            self._generaldelta = False

        # Delta chains for changelogs tend to be very small because entries
        # tend to be small and don't delta well with each. So disable delta
        # chains.
        self._storedeltachains = False

        self._realopener = opener
        self._delayed = False
        self._delaybuf = None
        self._divert = False
        self.filteredrevs = frozenset()

    def tip(self):
        """filtered version of revlog.tip"""
        for i in xrange(len(self) -1, -2, -1):
            if i not in self.filteredrevs:
                return self.node(i)

    def __contains__(self, rev):
        """filtered version of revlog.__contains__"""
        return (0 <= rev < len(self)
                and rev not in self.filteredrevs)

    def __iter__(self):
        """filtered version of revlog.__iter__"""
        if len(self.filteredrevs) == 0:
            return revlog.revlog.__iter__(self)

        def filterediter():
            for i in xrange(len(self)):
                if i not in self.filteredrevs:
                    yield i

        return filterediter()

    def revs(self, start=0, stop=None):
        """filtered version of revlog.revs"""
        for i in super(changelog, self).revs(start, stop):
            if i not in self.filteredrevs:
                yield i

    @util.propertycache
    def nodemap(self):
        # XXX need filtering too
        self.rev(self.node(0))
        return self._nodecache

    def reachableroots(self, minroot, heads, roots, includepath=False):
        return self.index.reachableroots2(minroot, heads, roots, includepath)

    def headrevs(self):
        if self.filteredrevs:
            try:
                return self.index.headrevsfiltered(self.filteredrevs)
            # AttributeError covers non-c-extension environments and
            # old c extensions without filter handling.
            except AttributeError:
                return self._headrevs()

        return super(changelog, self).headrevs()

    def strip(self, *args, **kwargs):
        # XXX make something better than assert
        # We can't expect proper strip behavior if we are filtered.
        assert not self.filteredrevs
        super(changelog, self).strip(*args, **kwargs)

    def rev(self, node):
        """filtered version of revlog.rev"""
        r = super(changelog, self).rev(node)
        if r in self.filteredrevs:
            raise error.FilteredLookupError(hex(node), self.indexfile,
                                            _('filtered node'))
        return r

    def node(self, rev):
        """filtered version of revlog.node"""
        if rev in self.filteredrevs:
            raise error.FilteredIndexError(rev)
        return super(changelog, self).node(rev)

    def linkrev(self, rev):
        """filtered version of revlog.linkrev"""
        if rev in self.filteredrevs:
            raise error.FilteredIndexError(rev)
        return super(changelog, self).linkrev(rev)

    def parentrevs(self, rev):
        """filtered version of revlog.parentrevs"""
        if rev in self.filteredrevs:
            raise error.FilteredIndexError(rev)
        return super(changelog, self).parentrevs(rev)

    def flags(self, rev):
        """filtered version of revlog.flags"""
        if rev in self.filteredrevs:
            raise error.FilteredIndexError(rev)
        return super(changelog, self).flags(rev)

    def delayupdate(self, tr):
        "delay visibility of index updates to other readers"

        if not self._delayed:
            if len(self) == 0:
                self._divert = True
                if self._realopener.exists(self.indexfile + '.a'):
                    self._realopener.unlink(self.indexfile + '.a')
                self.opener = _divertopener(self._realopener, self.indexfile)
            else:
                self._delaybuf = []
                self.opener = _delayopener(self._realopener, self.indexfile,
                                           self._delaybuf)
        self._delayed = True
        tr.addpending('cl-%i' % id(self), self._writepending)
        tr.addfinalize('cl-%i' % id(self), self._finalize)

    def _finalize(self, tr):
        "finalize index updates"
        self._delayed = False
        self.opener = self._realopener
        # move redirected index data back into place
        if self._divert:
            assert not self._delaybuf
            tmpname = self.indexfile + ".a"
            nfile = self.opener.open(tmpname)
            nfile.close()
            self.opener.rename(tmpname, self.indexfile, checkambig=True)
        elif self._delaybuf:
            fp = self.opener(self.indexfile, 'a', checkambig=True)
            fp.write("".join(self._delaybuf))
            fp.close()
            self._delaybuf = None
        self._divert = False
        # split when we're done
        self.checkinlinesize(tr)

    def readpending(self, file):
        """read index data from a "pending" file

        During a transaction, the actual changeset data is already stored in the
        main file, but not yet finalized in the on-disk index. Instead, a
        "pending" index is written by the transaction logic. If this function
        is running, we are likely in a subprocess invoked in a hook. The
        subprocess is informed that it is within a transaction and needs to
        access its content.

        This function will read all the index data out of the pending file and
        overwrite the main index."""

        if not self.opener.exists(file):
            return # no pending data for changelog
        r = revlog.revlog(self.opener, file)
        self.index = r.index
        self.nodemap = r.nodemap
        self._nodecache = r._nodecache
        self._chunkcache = r._chunkcache

    def _writepending(self, tr):
        "create a file containing the unfinalized state for pretxnchangegroup"
        if self._delaybuf:
            # make a temporary copy of the index
            fp1 = self._realopener(self.indexfile)
            pendingfilename = self.indexfile + ".a"
            # register as a temp file to ensure cleanup on failure
            tr.registertmp(pendingfilename)
            # write existing data
            fp2 = self._realopener(pendingfilename, "w")
            fp2.write(fp1.read())
            # add pending data
            fp2.write("".join(self._delaybuf))
            fp2.close()
            # switch modes so finalize can simply rename
            self._delaybuf = None
            self._divert = True
            self.opener = _divertopener(self._realopener, self.indexfile)

        if self._divert:
            return True

        return False

    def checkinlinesize(self, tr, fp=None):
        if not self._delayed:
            revlog.revlog.checkinlinesize(self, tr, fp)

    def read(self, node):
        """Obtain data from a parsed changelog revision.

        Returns a 6-tuple of:

           - manifest node in binary
           - author/user as a localstr
           - date as a 2-tuple of (time, timezone)
           - list of files
           - commit message as a localstr
           - dict of extra metadata

        Unless you need to access all fields, consider calling
        ``changelogrevision`` instead, as it is faster for partial object
        access.
        """
        c = changelogrevision(self.revision(node))
        return (
            c.manifest,
            c.user,
            c.date,
            c.files,
            c.description,
            c.extra
        )

    def changelogrevision(self, nodeorrev):
        """Obtain a ``changelogrevision`` for a node or revision."""
        return changelogrevision(self.revision(nodeorrev))

    def readfiles(self, node):
        """
        short version of read that only returns the files modified by the cset
        """
        text = self.revision(node)
        if not text:
            return []
        last = text.index("\n\n")
        l = text[:last].split('\n')
        return l[3:]

    def add(self, manifest, files, desc, transaction, p1, p2,
                  user, date=None, extra=None):
        # Convert to UTF-8 encoded bytestrings as the very first
        # thing: calling any method on a localstr object will turn it
        # into a str object and the cached UTF-8 string is thus lost.
        user, desc = encoding.fromlocal(user), encoding.fromlocal(desc)

        user = user.strip()
        # An empty username or a username with a "\n" will make the
        # revision text contain two "\n\n" sequences -> corrupt
        # repository since read cannot unpack the revision.
        if not user:
            raise error.RevlogError(_("empty username"))
        if "\n" in user:
            raise error.RevlogError(_("username %s contains a newline")
                                    % repr(user))

        desc = stripdesc(desc)

        if date:
            parseddate = "%d %d" % util.parsedate(date)
        else:
            parseddate = "%d %d" % util.makedate()
        if extra:
            branch = extra.get("branch")
            if branch in ("default", ""):
                del extra["branch"]
            elif branch in (".", "null", "tip"):
                raise error.RevlogError(_('the name \'%s\' is reserved')
                                        % branch)
        if extra:
            extra = encodeextra(extra)
            parseddate = "%s %s" % (parseddate, extra)
        l = [hex(manifest), user, parseddate] + sorted(files) + ["", desc]
        text = "\n".join(l)
        return self.addrevision(text, transaction, len(self), p1, p2)

    def branchinfo(self, rev):
        """return the branch name and open/close state of a revision

        This function exists because creating a changectx object
        just to access this is costly."""
        extra = self.read(rev)[5]
        return encoding.tolocal(extra.get("branch")), 'close' in extra