hgext/convert/filemap.py

xdiff: add a preprocessing step that trims files

xdiff has a `xdl_trim_ends` step that removes common lines, unmatchable
lines. That is in theory good, but happens too late - after splitting,
hashing, and adjusting the hash values so they are unique. Those splitting,
hashing and adjusting hash values steps could have noticeable overhead.

Diffing two large files with minor (one-line-ish) changes are not uncommon.
In that case, the raw performance of those preparation steps seriously
matter. Even allocating an O(N) array and storing line offsets to it is
expensive.  Therefore my previous attempts [1] [2] cannot be good enough
since they do not remove the O(N) array assignment.

This patch adds a preprocessing step - `xdl_trim_files` that runs before
other preprocessing steps. It counts common prefix and suffix and lines in
them (needed for displaying line number), without doing anything else.

Testing with a crafted large (169MB) file, with minor change:

```
  open('a','w').write(''.join('%s\n' % (i % 100000) for i in xrange(30000000) if i != 6000000))
  open('b','w').write(''.join('%s\n' % (i % 100000) for i in xrange(30000000) if i != 6003000))
```

Running xdiff by a simple binary [3], this patch improves the xdiff perf by
more than 10x for the above case:

```
  # xdiff before this patch
  2.41s user 1.13s system 98% cpu 3.592 total
  # xdiff after this patch
  0.14s user 0.16s system 98% cpu 0.309 total
  # gnu diffutils
  0.12s user 0.15s system 98% cpu 0.272 total
  # (best of 20 runs)
```

It's still slightly slower than GNU diffutils. But it's pretty close now.

Testing with real repo data:

For the whole repo, this patch makes xdiff 25% faster:

```
  # hg perfbdiff --count 100 --alldata -c d334afc585e2 --blocks [--xdiff]
  # xdiff, after
  ! wall 0.058861 comb 0.050000 user 0.050000 sys 0.000000 (best of 100)
  # xdiff, before
  ! wall 0.077816 comb 0.080000 user 0.080000 sys 0.000000 (best of 91)
  # bdiff
  ! wall 0.117473 comb 0.120000 user 0.120000 sys 0.000000 (best of 67)
```

For files that are long (ex. commands.py), the speedup is more than 3x, very
significant:

```
  # hg perfbdiff --count 3000 --blocks commands.py.i 1 [--xdiff]
  # xdiff, after
  ! wall 0.690583 comb 0.690000 user 0.690000 sys 0.000000 (best of 12)
  # xdiff, before
  ! wall 2.240361 comb 2.210000 user 2.210000 sys 0.000000 (best of 4)
  # bdiff
  ! wall 2.469852 comb 2.440000 user 2.440000 sys 0.000000 (best of 4)
```

[1]: https://phab.mercurial-scm.org/D2631
[2]: https://phab.mercurial-scm.org/D2634
[3]:

```
// Code to run xdiff from command line. No proper error handling.
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include "mercurial/thirdparty/xdiff/xdiff.h"
#define ensure(x) if (!(x)) exit(255);
mmfile_t readfile(const char *path) {
  struct stat st; int fd = open(path, O_RDONLY);
  fstat(fd, &st); mmfile_t file = { malloc(st.st_size), st.st_size };
  ensure(read(fd, file.ptr, st.st_size) == st.st_size); close(fd);
  return file;
}
int main(int argc, char const *argv[]) {
  mmfile_t a = readfile(argv[1]), b = readfile(argv[2]);
  xpparam_t xpp = {0}; xdemitconf_t xecfg = {0}; xdemitcb_t ecb = {0};
  xdl_diff(&a, &b, &xpp, &xecfg, &ecb);
  return 0;
}
```

Differential Revision: https://phab.mercurial-scm.org/D2686

# Copyright 2007 Bryan O'Sullivan <[e-mail]>
# Copyright 2007 Alexis S. L. Carvalho <[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, print_function

import posixpath

from mercurial.i18n import _
from mercurial import (
    error,
    pycompat,
)
from . import common
SKIPREV = common.SKIPREV

def rpairs(path):
    '''Yield tuples with path split at '/', starting with the full path.
    No leading, trailing or double '/', please.
    >>> for x in rpairs(b'foo/bar/baz'): print(x)
    ('foo/bar/baz', '')
    ('foo/bar', 'baz')
    ('foo', 'bar/baz')
    ('.', 'foo/bar/baz')
    '''
    i = len(path)
    while i != -1:
        yield path[:i], path[i + 1:]
        i = path.rfind('/', 0, i)
    yield '.', path

def normalize(path):
    ''' We use posixpath.normpath to support cross-platform path format.
    However, it doesn't handle None input. So we wrap it up. '''
    if path is None:
        return None
    return posixpath.normpath(path)

class filemapper(object):
    '''Map and filter filenames when importing.
    A name can be mapped to itself, a new name, or None (omit from new
    repository).'''

    def __init__(self, ui, path=None):
        self.ui = ui
        self.include = {}
        self.exclude = {}
        self.rename = {}
        self.targetprefixes = None
        if path:
            if self.parse(path):
                raise error.Abort(_('errors in filemap'))

    def parse(self, path):
        errs = 0
        def check(name, mapping, listname):
            if not name:
                self.ui.warn(_('%s:%d: path to %s is missing\n') %
                             (lex.infile, lex.lineno, listname))
                return 1
            if name in mapping:
                self.ui.warn(_('%s:%d: %r already in %s list\n') %
                             (lex.infile, lex.lineno, name, listname))
                return 1
            if (name.startswith('/') or
                name.endswith('/') or
                '//' in name):
                self.ui.warn(_('%s:%d: superfluous / in %s %r\n') %
                             (lex.infile, lex.lineno, listname,
                              pycompat.bytestr(name)))
                return 1
            return 0
        lex = common.shlexer(
            filepath=path, wordchars='!@#$%^&*()-=+[]{}|;:,./<>?')
        cmd = lex.get_token()
        while cmd:
            if cmd == 'include':
                name = normalize(lex.get_token())
                errs += check(name, self.exclude, 'exclude')
                self.include[name] = name
            elif cmd == 'exclude':
                name = normalize(lex.get_token())
                errs += check(name, self.include, 'include')
                errs += check(name, self.rename, 'rename')
                self.exclude[name] = name
            elif cmd == 'rename':
                src = normalize(lex.get_token())
                dest = normalize(lex.get_token())
                errs += check(src, self.exclude, 'exclude')
                self.rename[src] = dest
            elif cmd == 'source':
                errs += self.parse(normalize(lex.get_token()))
            else:
                self.ui.warn(_('%s:%d: unknown directive %r\n') %
                             (lex.infile, lex.lineno, pycompat.bytestr(cmd)))
                errs += 1
            cmd = lex.get_token()
        return errs

    def lookup(self, name, mapping):
        name = normalize(name)
        for pre, suf in rpairs(name):
            try:
                return mapping[pre], pre, suf
            except KeyError:
                pass
        return '', name, ''

    def istargetfile(self, filename):
        """Return true if the given target filename is covered as a destination
        of the filemap. This is useful for identifying what parts of the target
        repo belong to the source repo and what parts don't."""
        if self.targetprefixes is None:
            self.targetprefixes = set()
            for before, after in self.rename.iteritems():
                self.targetprefixes.add(after)

        # If "." is a target, then all target files are considered from the
        # source.
        if not self.targetprefixes or '.' in self.targetprefixes:
            return True

        filename = normalize(filename)
        for pre, suf in rpairs(filename):
            # This check is imperfect since it doesn't account for the
            # include/exclude list, but it should work in filemaps that don't
            # apply include/exclude to the same source directories they are
            # renaming.
            if pre in self.targetprefixes:
                return True
        return False

    def __call__(self, name):
        if self.include:
            inc = self.lookup(name, self.include)[0]
        else:
            inc = name
        if self.exclude:
            exc = self.lookup(name, self.exclude)[0]
        else:
            exc = ''
        if (not self.include and exc) or (len(inc) <= len(exc)):
            return None
        newpre, pre, suf = self.lookup(name, self.rename)
        if newpre:
            if newpre == '.':
                return suf
            if suf:
                if newpre.endswith('/'):
                    return newpre + suf
                return newpre + '/' + suf
            return newpre
        return name

    def active(self):
        return bool(self.include or self.exclude or self.rename)

# This class does two additional things compared to a regular source:
#
# - Filter and rename files.  This is mostly wrapped by the filemapper
#   class above. We hide the original filename in the revision that is
#   returned by getchanges to be able to find things later in getfile.
#
# - Return only revisions that matter for the files we're interested in.
#   This involves rewriting the parents of the original revision to
#   create a graph that is restricted to those revisions.
#
#   This set of revisions includes not only revisions that directly
#   touch files we're interested in, but also merges that merge two
#   or more interesting revisions.

class filemap_source(common.converter_source):
    def __init__(self, ui, baseconverter, filemap):
        super(filemap_source, self).__init__(ui, baseconverter.repotype)
        self.base = baseconverter
        self.filemapper = filemapper(ui, filemap)
        self.commits = {}
        # if a revision rev has parent p in the original revision graph, then
        # rev will have parent self.parentmap[p] in the restricted graph.
        self.parentmap = {}
        # self.wantedancestors[rev] is the set of all ancestors of rev that
        # are in the restricted graph.
        self.wantedancestors = {}
        self.convertedorder = None
        self._rebuilt = False
        self.origparents = {}
        self.children = {}
        self.seenchildren = {}
        # experimental config: convert.ignoreancestorcheck
        self.ignoreancestorcheck = self.ui.configbool('convert',
                                                      'ignoreancestorcheck')

    def before(self):
        self.base.before()

    def after(self):
        self.base.after()

    def setrevmap(self, revmap):
        # rebuild our state to make things restartable
        #
        # To avoid calling getcommit for every revision that has already
        # been converted, we rebuild only the parentmap, delaying the
        # rebuild of wantedancestors until we need it (i.e. until a
        # merge).
        #
        # We assume the order argument lists the revisions in
        # topological order, so that we can infer which revisions were
        # wanted by previous runs.
        self._rebuilt = not revmap
        seen = {SKIPREV: SKIPREV}
        dummyset = set()
        converted = []
        for rev in revmap.order:
            mapped = revmap[rev]
            wanted = mapped not in seen
            if wanted:
                seen[mapped] = rev
                self.parentmap[rev] = rev
            else:
                self.parentmap[rev] = seen[mapped]
            self.wantedancestors[rev] = dummyset
            arg = seen[mapped]
            if arg == SKIPREV:
                arg = None
            converted.append((rev, wanted, arg))
        self.convertedorder = converted
        return self.base.setrevmap(revmap)

    def rebuild(self):
        if self._rebuilt:
            return True
        self._rebuilt = True
        self.parentmap.clear()
        self.wantedancestors.clear()
        self.seenchildren.clear()
        for rev, wanted, arg in self.convertedorder:
            if rev not in self.origparents:
                try:
                    self.origparents[rev] = self.getcommit(rev).parents
                except error.RepoLookupError:
                    self.ui.debug("unknown revmap source: %s\n" % rev)
                    continue
            if arg is not None:
                self.children[arg] = self.children.get(arg, 0) + 1

        for rev, wanted, arg in self.convertedorder:
            try:
                parents = self.origparents[rev]
            except KeyError:
                continue # unknown revmap source
            if wanted:
                self.mark_wanted(rev, parents)
            else:
                self.mark_not_wanted(rev, arg)
            self._discard(arg, *parents)

        return True

    def getheads(self):
        return self.base.getheads()

    def getcommit(self, rev):
        # We want to save a reference to the commit objects to be able
        # to rewrite their parents later on.
        c = self.commits[rev] = self.base.getcommit(rev)
        for p in c.parents:
            self.children[p] = self.children.get(p, 0) + 1
        return c

    def _cachedcommit(self, rev):
        if rev in self.commits:
            return self.commits[rev]
        return self.base.getcommit(rev)

    def _discard(self, *revs):
        for r in revs:
            if r is None:
                continue
            self.seenchildren[r] = self.seenchildren.get(r, 0) + 1
            if self.seenchildren[r] == self.children[r]:
                self.wantedancestors.pop(r, None)
                self.parentmap.pop(r, None)
                del self.seenchildren[r]
                if self._rebuilt:
                    del self.children[r]

    def wanted(self, rev, i):
        # Return True if we're directly interested in rev.
        #
        # i is an index selecting one of the parents of rev (if rev
        # has no parents, i is None).  getchangedfiles will give us
        # the list of files that are different in rev and in the parent
        # indicated by i.  If we're interested in any of these files,
        # we're interested in rev.
        try:
            files = self.base.getchangedfiles(rev, i)
        except NotImplementedError:
            raise error.Abort(_("source repository doesn't support --filemap"))
        for f in files:
            if self.filemapper(f):
                return True
        return False

    def mark_not_wanted(self, rev, p):
        # Mark rev as not interesting and update data structures.

        if p is None:
            # A root revision. Use SKIPREV to indicate that it doesn't
            # map to any revision in the restricted graph.  Put SKIPREV
            # in the set of wanted ancestors to simplify code elsewhere
            self.parentmap[rev] = SKIPREV
            self.wantedancestors[rev] = {SKIPREV}
            return

        # Reuse the data from our parent.
        self.parentmap[rev] = self.parentmap[p]
        self.wantedancestors[rev] = self.wantedancestors[p]

    def mark_wanted(self, rev, parents):
        # Mark rev ss wanted and update data structures.

        # rev will be in the restricted graph, so children of rev in
        # the original graph should still have rev as a parent in the
        # restricted graph.
        self.parentmap[rev] = rev

        # The set of wanted ancestors of rev is the union of the sets
        # of wanted ancestors of its parents. Plus rev itself.
        wrev = set()
        for p in parents:
            if p in self.wantedancestors:
                wrev.update(self.wantedancestors[p])
            else:
                self.ui.warn(_('warning: %s parent %s is missing\n') %
                             (rev, p))
        wrev.add(rev)
        self.wantedancestors[rev] = wrev

    def getchanges(self, rev, full):
        parents = self.commits[rev].parents
        if len(parents) > 1 and not self.ignoreancestorcheck:
            self.rebuild()

        # To decide whether we're interested in rev we:
        #
        # - calculate what parents rev will have if it turns out we're
        #   interested in it.  If it's going to have more than 1 parent,
        #   we're interested in it.
        #
        # - otherwise, we'll compare it with the single parent we found.
        #   If any of the files we're interested in is different in the
        #   the two revisions, we're interested in rev.

        # A parent p is interesting if its mapped version (self.parentmap[p]):
        # - is not SKIPREV
        # - is still not in the list of parents (we don't want duplicates)
        # - is not an ancestor of the mapped versions of the other parents or
        #   there is no parent in the same branch than the current revision.
        mparents = []
        knownparents = set()
        branch = self.commits[rev].branch
        hasbranchparent = False
        for i, p1 in enumerate(parents):
            mp1 = self.parentmap[p1]
            if mp1 == SKIPREV or mp1 in knownparents:
                continue

            isancestor = (not self.ignoreancestorcheck and
                          any(p2 for p2 in parents
                              if p1 != p2 and mp1 != self.parentmap[p2]
                                 and mp1 in self.wantedancestors[p2]))
            if not isancestor and not hasbranchparent and len(parents) > 1:
                # This could be expensive, avoid unnecessary calls.
                if self._cachedcommit(p1).branch == branch:
                    hasbranchparent = True
            mparents.append((p1, mp1, i, isancestor))
            knownparents.add(mp1)
        # Discard parents ancestors of other parents if there is a
        # non-ancestor one on the same branch than current revision.
        if hasbranchparent:
            mparents = [p for p in mparents if not p[3]]
        wp = None
        if mparents:
            wp = max(p[2] for p in mparents)
            mparents = [p[1] for p in mparents]
        elif parents:
            wp = 0

        self.origparents[rev] = parents

        closed = False
        if 'close' in self.commits[rev].extra:
            # A branch closing revision is only useful if one of its
            # parents belong to the branch being closed
            pbranches = [self._cachedcommit(p).branch for p in mparents]
            if branch in pbranches:
                closed = True

        if len(mparents) < 2 and not closed and not self.wanted(rev, wp):
            # We don't want this revision.
            # Update our state and tell the convert process to map this
            # revision to the same revision its parent as mapped to.
            p = None
            if parents:
                p = parents[wp]
            self.mark_not_wanted(rev, p)
            self.convertedorder.append((rev, False, p))
            self._discard(*parents)
            return self.parentmap[rev]

        # We want this revision.
        # Rewrite the parents of the commit object
        self.commits[rev].parents = mparents
        self.mark_wanted(rev, parents)
        self.convertedorder.append((rev, True, None))
        self._discard(*parents)

        # Get the real changes and do the filtering/mapping. To be
        # able to get the files later on in getfile, we hide the
        # original filename in the rev part of the return value.
        changes, copies, cleanp2 = self.base.getchanges(rev, full)
        files = {}
        ncleanp2 = set(cleanp2)
        for f, r in changes:
            newf = self.filemapper(f)
            if newf and (newf != f or newf not in files):
                files[newf] = (f, r)
                if newf != f:
                    ncleanp2.discard(f)
        files = sorted(files.items())

        ncopies = {}
        for c in copies:
            newc = self.filemapper(c)
            if newc:
                newsource = self.filemapper(copies[c])
                if newsource:
                    ncopies[newc] = newsource

        return files, ncopies, ncleanp2

    def targetfilebelongstosource(self, targetfilename):
        return self.filemapper.istargetfile(targetfilename)

    def getfile(self, name, rev):
        realname, realrev = rev
        return self.base.getfile(realname, realrev)

    def gettags(self):
        return self.base.gettags()

    def hasnativeorder(self):
        return self.base.hasnativeorder()

    def lookuprev(self, rev):
        return self.base.lookuprev(rev)

    def getbookmarks(self):
        return self.base.getbookmarks()

    def converted(self, rev, sinkrev):
        self.base.converted(rev, sinkrev)