view mercurial/simplemerge.py @ 30442:41a8106789ca

util: implement zstd compression engine

Now that zstd is vendored and being built (in some configurations), we
can implement a compression engine for zstd!

The zstd engine is a little different from existing engines. Because
it may not always be present, we have to defer load the module in case
importing it fails. We facilitate this via a cached property that holds
a reference to the module or None. The "available" method is
implemented to reflect reality.

The zstd engine declares its ability to handle bundles using the
"zstd" human name and the "ZS" internal name. The latter was chosen
because internal names are 2 characters (by only convention I think)
and "ZS" seems reasonable.

The engine, like others, supports specifying the compression level.
However, there are no consumers of this API that yet pass in that
argument. I have plans to change that, so stay tuned.

Since all we need to do to support bundle generation with a new
compression engine is implement and register the compression engine,
bundle generation with zstd "just works!" Tests demonstrating this
have been added.

How does performance of zstd for bundle generation compare? On the
mozilla-unified repo, `hg bundle --all -t <engine>-v2` yields the
following on my i7-6700K on Linux:

engine        CPU time     bundle size   vs orig size   throughput
none            97.0s     4,054,405,584     100.0%       41.8 MB/s
bzip2 (l=9)    393.6s       975,343,098      24.0%       10.3 MB/s
gzip (l=6)     184.0s     1,140,533,074      28.1%       22.0 MB/s
zstd (l=1)     108.2s     1,119,434,718      27.6%       37.5 MB/s
zstd (l=2)     111.3s     1,078,328,002      26.6%       36.4 MB/s
zstd (l=3)     113.7s     1,011,823,727      25.0%       35.7 MB/s
zstd (l=4)     116.0s     1,008,965,888      24.9%       35.0 MB/s
zstd (l=5)     121.0s       977,203,148      24.1%       33.5 MB/s
zstd (l=6)     131.7s       927,360,198      22.9%       30.8 MB/s
zstd (l=7)     139.0s       912,808,505      22.5%       29.2 MB/s
zstd (l=12)    198.1s       854,527,714      21.1%       20.5 MB/s
zstd (l=18)    681.6s       789,750,690      19.5%        5.9 MB/s

On compression, zstd for bundle generation delivers:

* better compression than gzip with significantly less CPU utilization
* better than bzip2 compression ratios while still being significantly
  faster than gzip
* ability to aggressively tune compression level to achieve
  significantly smaller bundles

That last point is important. With clone bundles, a server can
pre-generate a bundle file, upload it to a static file server, and
redirect clients to transparently download it during clone. The server
could choose to produce a zstd bundle with the highest compression
settings possible. This would take a very long time - a magnitude
longer than a typical zstd bundle generation - but the result would
be hundreds of megabytes smaller! For the clone volume we do at
Mozilla, this could translate to petabytes of bandwidth savings
per year and faster clones (due to smaller transfer size).

I don't have detailed numbers to report on decompression. However,
zstd decompression is fast: >1 GB/s output throughput on this machine,
even through the Python bindings. And it can do that regardless of the
compression level of the input. By the time you have enough data to
worry about overhead of decompression, you have plenty of other things
to worry about performance wise.

zstd is wins all around. I can't wait to implement support for it
on the wire protocol and in revlogs.

# Copyright (C) 2004, 2005 Canonical Ltd
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, see <http://www.gnu.org/licenses/>.

# mbp: "you know that thing where cvs gives you conflict markers?"
# s: "i hate that."

from __future__ import absolute_import

import os

from .i18n import _
from . import (
    error,
    mdiff,
    scmutil,
    util,
)

class CantReprocessAndShowBase(Exception):
    pass

def intersect(ra, rb):
    """Given two ranges return the range where they intersect or None.

    >>> intersect((0, 10), (0, 6))
    (0, 6)
    >>> intersect((0, 10), (5, 15))
    (5, 10)
    >>> intersect((0, 10), (10, 15))
    >>> intersect((0, 9), (10, 15))
    >>> intersect((0, 9), (7, 15))
    (7, 9)
    """
    assert ra[0] <= ra[1]
    assert rb[0] <= rb[1]

    sa = max(ra[0], rb[0])
    sb = min(ra[1], rb[1])
    if sa < sb:
        return sa, sb
    else:
        return None

def compare_range(a, astart, aend, b, bstart, bend):
    """Compare a[astart:aend] == b[bstart:bend], without slicing.
    """
    if (aend - astart) != (bend - bstart):
        return False
    for ia, ib in zip(xrange(astart, aend), xrange(bstart, bend)):
        if a[ia] != b[ib]:
            return False
    else:
        return True

class Merge3Text(object):
    """3-way merge of texts.

    Given strings BASE, OTHER, THIS, tries to produce a combined text
    incorporating the changes from both BASE->OTHER and BASE->THIS."""
    def __init__(self, basetext, atext, btext, base=None, a=None, b=None):
        self.basetext = basetext
        self.atext = atext
        self.btext = btext
        if base is None:
            base = mdiff.splitnewlines(basetext)
        if a is None:
            a = mdiff.splitnewlines(atext)
        if b is None:
            b = mdiff.splitnewlines(btext)
        self.base = base
        self.a = a
        self.b = b

    def merge_lines(self,
                    name_a=None,
                    name_b=None,
                    name_base=None,
                    start_marker='<<<<<<<',
                    mid_marker='=======',
                    end_marker='>>>>>>>',
                    base_marker=None,
                    localorother=None,
                    minimize=False):
        """Return merge in cvs-like form.
        """
        self.conflicts = False
        newline = '\n'
        if len(self.a) > 0:
            if self.a[0].endswith('\r\n'):
                newline = '\r\n'
            elif self.a[0].endswith('\r'):
                newline = '\r'
        if name_a and start_marker:
            start_marker = start_marker + ' ' + name_a
        if name_b and end_marker:
            end_marker = end_marker + ' ' + name_b
        if name_base and base_marker:
            base_marker = base_marker + ' ' + name_base
        merge_regions = self.merge_regions()
        if minimize:
            merge_regions = self.minimize(merge_regions)
        for t in merge_regions:
            what = t[0]
            if what == 'unchanged':
                for i in range(t[1], t[2]):
                    yield self.base[i]
            elif what == 'a' or what == 'same':
                for i in range(t[1], t[2]):
                    yield self.a[i]
            elif what == 'b':
                for i in range(t[1], t[2]):
                    yield self.b[i]
            elif what == 'conflict':
                if localorother == 'local':
                    for i in range(t[3], t[4]):
                        yield self.a[i]
                elif localorother == 'other':
                    for i in range(t[5], t[6]):
                        yield self.b[i]
                else:
                    self.conflicts = True
                    if start_marker is not None:
                        yield start_marker + newline
                    for i in range(t[3], t[4]):
                        yield self.a[i]
                    if base_marker is not None:
                        yield base_marker + newline
                        for i in range(t[1], t[2]):
                            yield self.base[i]
                    if mid_marker is not None:
                        yield mid_marker + newline
                    for i in range(t[5], t[6]):
                        yield self.b[i]
                    if end_marker is not None:
                        yield end_marker + newline
            else:
                raise ValueError(what)

    def merge_groups(self):
        """Yield sequence of line groups.  Each one is a tuple:

        'unchanged', lines
             Lines unchanged from base

        'a', lines
             Lines taken from a

        'same', lines
             Lines taken from a (and equal to b)

        'b', lines
             Lines taken from b

        'conflict', base_lines, a_lines, b_lines
             Lines from base were changed to either a or b and conflict.
        """
        for t in self.merge_regions():
            what = t[0]
            if what == 'unchanged':
                yield what, self.base[t[1]:t[2]]
            elif what == 'a' or what == 'same':
                yield what, self.a[t[1]:t[2]]
            elif what == 'b':
                yield what, self.b[t[1]:t[2]]
            elif what == 'conflict':
                yield (what,
                       self.base[t[1]:t[2]],
                       self.a[t[3]:t[4]],
                       self.b[t[5]:t[6]])
            else:
                raise ValueError(what)

    def merge_regions(self):
        """Return sequences of matching and conflicting regions.

        This returns tuples, where the first value says what kind we
        have:

        'unchanged', start, end
             Take a region of base[start:end]

        'same', astart, aend
             b and a are different from base but give the same result

        'a', start, end
             Non-clashing insertion from a[start:end]

        'conflict', zstart, zend, astart, aend, bstart, bend
            Conflict between a and b, with z as common ancestor

        Method is as follows:

        The two sequences align only on regions which match the base
        and both descendants.  These are found by doing a two-way diff
        of each one against the base, and then finding the
        intersections between those regions.  These "sync regions"
        are by definition unchanged in both and easily dealt with.

        The regions in between can be in any of three cases:
        conflicted, or changed on only one side.
        """

        # section a[0:ia] has been disposed of, etc
        iz = ia = ib = 0

        for region in self.find_sync_regions():
            zmatch, zend, amatch, aend, bmatch, bend = region
            #print 'match base [%d:%d]' % (zmatch, zend)

            matchlen = zend - zmatch
            assert matchlen >= 0
            assert matchlen == (aend - amatch)
            assert matchlen == (bend - bmatch)

            len_a = amatch - ia
            len_b = bmatch - ib
            len_base = zmatch - iz
            assert len_a >= 0
            assert len_b >= 0
            assert len_base >= 0

            #print 'unmatched a=%d, b=%d' % (len_a, len_b)

            if len_a or len_b:
                # try to avoid actually slicing the lists
                equal_a = compare_range(self.a, ia, amatch,
                                        self.base, iz, zmatch)
                equal_b = compare_range(self.b, ib, bmatch,
                                        self.base, iz, zmatch)
                same = compare_range(self.a, ia, amatch,
                                     self.b, ib, bmatch)

                if same:
                    yield 'same', ia, amatch
                elif equal_a and not equal_b:
                    yield 'b', ib, bmatch
                elif equal_b and not equal_a:
                    yield 'a', ia, amatch
                elif not equal_a and not equal_b:
                    yield 'conflict', iz, zmatch, ia, amatch, ib, bmatch
                else:
                    raise AssertionError("can't handle a=b=base but unmatched")

                ia = amatch
                ib = bmatch
            iz = zmatch

            # if the same part of the base was deleted on both sides
            # that's OK, we can just skip it.


            if matchlen > 0:
                assert ia == amatch
                assert ib == bmatch
                assert iz == zmatch

                yield 'unchanged', zmatch, zend
                iz = zend
                ia = aend
                ib = bend

    def minimize(self, merge_regions):
        """Trim conflict regions of lines where A and B sides match.

        Lines where both A and B have made the same changes at the beginning
        or the end of each merge region are eliminated from the conflict
        region and are instead considered the same.
        """
        for region in merge_regions:
            if region[0] != "conflict":
                yield region
                continue
            issue, z1, z2, a1, a2, b1, b2 = region
            alen = a2 - a1
            blen = b2 - b1

            # find matches at the front
            ii = 0
            while ii < alen and ii < blen and \
                  self.a[a1 + ii] == self.b[b1 + ii]:
                ii += 1
            startmatches = ii

            # find matches at the end
            ii = 0
            while ii < alen and ii < blen and \
                  self.a[a2 - ii - 1] == self.b[b2 - ii - 1]:
                ii += 1
            endmatches = ii

            if startmatches > 0:
                yield 'same', a1, a1 + startmatches

            yield ('conflict', z1, z2,
                    a1 + startmatches, a2 - endmatches,
                    b1 + startmatches, b2 - endmatches)

            if endmatches > 0:
                yield 'same', a2 - endmatches, a2

    def find_sync_regions(self):
        """Return a list of sync regions, where both descendants match the base.

        Generates a list of (base1, base2, a1, a2, b1, b2).  There is
        always a zero-length sync region at the end of all the files.
        """

        ia = ib = 0
        amatches = mdiff.get_matching_blocks(self.basetext, self.atext)
        bmatches = mdiff.get_matching_blocks(self.basetext, self.btext)
        len_a = len(amatches)
        len_b = len(bmatches)

        sl = []

        while ia < len_a and ib < len_b:
            abase, amatch, alen = amatches[ia]
            bbase, bmatch, blen = bmatches[ib]

            # there is an unconflicted block at i; how long does it
            # extend?  until whichever one ends earlier.
            i = intersect((abase, abase + alen), (bbase, bbase + blen))
            if i:
                intbase = i[0]
                intend = i[1]
                intlen = intend - intbase

                # found a match of base[i[0], i[1]]; this may be less than
                # the region that matches in either one
                assert intlen <= alen
                assert intlen <= blen
                assert abase <= intbase
                assert bbase <= intbase

                asub = amatch + (intbase - abase)
                bsub = bmatch + (intbase - bbase)
                aend = asub + intlen
                bend = bsub + intlen

                assert self.base[intbase:intend] == self.a[asub:aend], \
                       (self.base[intbase:intend], self.a[asub:aend])

                assert self.base[intbase:intend] == self.b[bsub:bend]

                sl.append((intbase, intend,
                           asub, aend,
                           bsub, bend))

            # advance whichever one ends first in the base text
            if (abase + alen) < (bbase + blen):
                ia += 1
            else:
                ib += 1

        intbase = len(self.base)
        abase = len(self.a)
        bbase = len(self.b)
        sl.append((intbase, intbase, abase, abase, bbase, bbase))

        return sl

    def find_unconflicted(self):
        """Return a list of ranges in base that are not conflicted."""
        am = mdiff.get_matching_blocks(self.basetext, self.atext)
        bm = mdiff.get_matching_blocks(self.basetext, self.btext)

        unc = []

        while am and bm:
            # there is an unconflicted block at i; how long does it
            # extend?  until whichever one ends earlier.
            a1 = am[0][0]
            a2 = a1 + am[0][2]
            b1 = bm[0][0]
            b2 = b1 + bm[0][2]
            i = intersect((a1, a2), (b1, b2))
            if i:
                unc.append(i)

            if a2 < b2:
                del am[0]
            else:
                del bm[0]

        return unc

def simplemerge(ui, local, base, other, **opts):
    def readfile(filename):
        f = open(filename, "rb")
        text = f.read()
        f.close()
        if util.binary(text):
            msg = _("%s looks like a binary file.") % filename
            if not opts.get('quiet'):
                ui.warn(_('warning: %s\n') % msg)
            if not opts.get('text'):
                raise error.Abort(msg)
        return text

    mode = opts.get('mode','merge')
    if mode == 'union':
        name_a = None
        name_b = None
        name_base = None
    else:
        name_a = local
        name_b = other
        name_base = None
        labels = opts.get('label', [])
        if len(labels) > 0:
            name_a = labels[0]
        if len(labels) > 1:
            name_b = labels[1]
        if len(labels) > 2:
            name_base = labels[2]
        if len(labels) > 3:
            raise error.Abort(_("can only specify three labels."))

    try:
        localtext = readfile(local)
        basetext = readfile(base)
        othertext = readfile(other)
    except error.Abort:
        return 1

    local = os.path.realpath(local)
    if not opts.get('print'):
        opener = scmutil.opener(os.path.dirname(local))
        out = opener(os.path.basename(local), "w", atomictemp=True)
    else:
        out = ui.fout

    m3 = Merge3Text(basetext, localtext, othertext)
    extrakwargs = {
            "localorother": opts.get("localorother", None),
            'minimize': True,
        }
    if mode == 'union':
        extrakwargs['start_marker'] = None
        extrakwargs['mid_marker'] = None
        extrakwargs['end_marker'] = None
    elif name_base is not None:
        extrakwargs['base_marker'] = '|||||||'
        extrakwargs['name_base'] = name_base
        extrakwargs['minimize'] = False
    for line in m3.merge_lines(name_a=name_a, name_b=name_b, **extrakwargs):
        out.write(line)

    if not opts.get('print'):
        out.close()

    if m3.conflicts and not mode == 'union':
        return 1