hgext/record.py

Take advantage of fstat calls clustering per directory if OS support it.

util module implements two versions of statfiles function

_statfiles calls lstat per file

_statfiles_clustered takes advantage of optimizations in osutil.c, stats all
files in directory at once when new directory is hit and caches the results

util.statfiles dispatches to appropriate version during module loading

The speedup on directory tree with 2k directories and 63k files is about
factor of 1.8 (1.3s -> 0.8s for hg diff - hg startup overhead about .2s)

At this point only Win32 now benefit from this patch.
Rest of OSes use the non clustered implementation.

# record.py
#
# Copyright 2007 Bryan O'Sullivan <[e-mail]>
#
# This software may be used and distributed according to the terms of
# the GNU General Public License, incorporated herein by reference.

'''interactive change selection during commit or qrefresh'''

from mercurial.i18n import gettext, _
from mercurial import cmdutil, commands, extensions, hg, mdiff, patch
from mercurial import util
import copy, cStringIO, errno, operator, os, re, tempfile

lines_re = re.compile(r'@@ -(\d+),(\d+) \+(\d+),(\d+) @@\s*(.*)')

def scanpatch(fp):
    """like patch.iterhunks, but yield different events

    - ('file',    [header_lines + fromfile + tofile])
    - ('context', [context_lines])
    - ('hunk',    [hunk_lines])
    - ('range',   (-start,len, +start,len, diffp))
    """
    lr = patch.linereader(fp)

    def scanwhile(first, p):
        """scan lr while predicate holds"""
        lines = [first]
        while True:
            line = lr.readline()
            if not line:
                break
            if p(line):
                lines.append(line)
            else:
                lr.push(line)
                break
        return lines

    while True:
        line = lr.readline()
        if not line:
            break
        if line.startswith('diff --git a/'):
            def notheader(line):
                s = line.split(None, 1)
                return not s or s[0] not in ('---', 'diff')
            header = scanwhile(line, notheader)
            fromfile = lr.readline()
            if fromfile.startswith('---'):
                tofile = lr.readline()
                header += [fromfile, tofile]
            else:
                lr.push(fromfile)
            yield 'file', header
        elif line[0] == ' ':
            yield 'context', scanwhile(line, lambda l: l[0] in ' \\')
        elif line[0] in '-+':
            yield 'hunk', scanwhile(line, lambda l: l[0] in '-+\\')
        else:
            m = lines_re.match(line)
            if m:
                yield 'range', m.groups()
            else:
                raise patch.PatchError('unknown patch content: %r' % line)

class header(object):
    """patch header

    XXX shoudn't we move this to mercurial/patch.py ?
    """
    diff_re = re.compile('diff --git a/(.*) b/(.*)$')
    allhunks_re = re.compile('(?:index|new file|deleted file) ')
    pretty_re = re.compile('(?:new file|deleted file) ')
    special_re = re.compile('(?:index|new|deleted|copy|rename) ')

    def __init__(self, header):
        self.header = header
        self.hunks = []

    def binary(self):
        for h in self.header:
            if h.startswith('index '):
                return True

    def pretty(self, fp):
        for h in self.header:
            if h.startswith('index '):
                fp.write(_('this modifies a binary file (all or nothing)\n'))
                break
            if self.pretty_re.match(h):
                fp.write(h)
                if self.binary():
                    fp.write(_('this is a binary file\n'))
                break
            if h.startswith('---'):
                fp.write(_('%d hunks, %d lines changed\n') %
                         (len(self.hunks),
                          sum([h.added + h.removed for h in self.hunks])))
                break
            fp.write(h)

    def write(self, fp):
        fp.write(''.join(self.header))

    def allhunks(self):
        for h in self.header:
            if self.allhunks_re.match(h):
                return True

    def files(self):
        fromfile, tofile = self.diff_re.match(self.header[0]).groups()
        if fromfile == tofile:
            return [fromfile]
        return [fromfile, tofile]

    def filename(self):
        return self.files()[-1]

    def __repr__(self):
        return '<header %s>' % (' '.join(map(repr, self.files())))

    def special(self):
        for h in self.header:
            if self.special_re.match(h):
                return True

def countchanges(hunk):
    """hunk -> (n+,n-)"""
    add = len([h for h in hunk if h[0] == '+'])
    rem = len([h for h in hunk if h[0] == '-'])
    return add, rem

class hunk(object):
    """patch hunk

    XXX shouldn't we merge this with patch.hunk ?
    """
    maxcontext = 3

    def __init__(self, header, fromline, toline, proc, before, hunk, after):
        def trimcontext(number, lines):
            delta = len(lines) - self.maxcontext
            if False and delta > 0:
                return number + delta, lines[:self.maxcontext]
            return number, lines

        self.header = header
        self.fromline, self.before = trimcontext(fromline, before)
        self.toline, self.after = trimcontext(toline, after)
        self.proc = proc
        self.hunk = hunk
        self.added, self.removed = countchanges(self.hunk)

    def write(self, fp):
        delta = len(self.before) + len(self.after)
        if self.after and self.after[-1] == '\\ No newline at end of file\n':
            delta -= 1
        fromlen = delta + self.removed
        tolen = delta + self.added
        fp.write('@@ -%d,%d +%d,%d @@%s\n' %
                 (self.fromline, fromlen, self.toline, tolen,
                  self.proc and (' ' + self.proc)))
        fp.write(''.join(self.before + self.hunk + self.after))

    pretty = write

    def filename(self):
        return self.header.filename()

    def __repr__(self):
        return '<hunk %r@%d>' % (self.filename(), self.fromline)

def parsepatch(fp):
    """patch -> [] of hunks """
    class parser(object):
        """patch parsing state machine"""
        def __init__(self):
            self.fromline = 0
            self.toline = 0
            self.proc = ''
            self.header = None
            self.context = []
            self.before = []
            self.hunk = []
            self.stream = []

        def addrange(self, (fromstart, fromend, tostart, toend, proc)):
            self.fromline = int(fromstart)
            self.toline = int(tostart)
            self.proc = proc

        def addcontext(self, context):
            if self.hunk:
                h = hunk(self.header, self.fromline, self.toline, self.proc,
                         self.before, self.hunk, context)
                self.header.hunks.append(h)
                self.stream.append(h)
                self.fromline += len(self.before) + h.removed
                self.toline += len(self.before) + h.added
                self.before = []
                self.hunk = []
                self.proc = ''
            self.context = context

        def addhunk(self, hunk):
            if self.context:
                self.before = self.context
                self.context = []
            self.hunk = hunk

        def newfile(self, hdr):
            self.addcontext([])
            h = header(hdr)
            self.stream.append(h)
            self.header = h

        def finished(self):
            self.addcontext([])
            return self.stream

        transitions = {
            'file': {'context': addcontext,
                     'file': newfile,
                     'hunk': addhunk,
                     'range': addrange},
            'context': {'file': newfile,
                        'hunk': addhunk,
                        'range': addrange},
            'hunk': {'context': addcontext,
                     'file': newfile,
                     'range': addrange},
            'range': {'context': addcontext,
                      'hunk': addhunk},
            }

    p = parser()

    state = 'context'
    for newstate, data in scanpatch(fp):
        try:
            p.transitions[state][newstate](p, data)
        except KeyError:
            raise patch.PatchError('unhandled transition: %s -> %s' %
                                   (state, newstate))
        state = newstate
    return p.finished()

def filterpatch(ui, chunks):
    """Interactively filter patch chunks into applied-only chunks"""
    chunks = list(chunks)
    chunks.reverse()
    seen = {}
    def consumefile():
        """fetch next portion from chunks until a 'header' is seen
        NB: header == new-file mark
        """
        consumed = []
        while chunks:
            if isinstance(chunks[-1], header):
                break
            else:
                consumed.append(chunks.pop())
        return consumed

    resp_all = [None]   # this two are changed from inside prompt,
    resp_file = [None]  # so can't be usual variables
    applied = {}        # 'filename' -> [] of chunks
    def prompt(query):
        """prompt query, and process base inputs

        - y/n for the rest of file
        - y/n for the rest
        - ? (help)
        - q (quit)

        else, input is returned to the caller.
        """
        if resp_all[0] is not None:
            return resp_all[0]
        if resp_file[0] is not None:
            return resp_file[0]
        while True:
            choices = _('[Ynsfdaq?]')
            r = (ui.prompt("%s %s " % (query, choices), '(?i)%s?$' % choices)
                 or _('y')).lower()
            if r == _('?'):
                doc = gettext(record.__doc__)
                c = doc.find(_('y - record this change'))
                for l in doc[c:].splitlines():
                    if l: ui.write(l.strip(), '\n')
                continue
            elif r == _('s'):
                r = resp_file[0] = 'n'
            elif r == _('f'):
                r = resp_file[0] = 'y'
            elif r == _('d'):
                r = resp_all[0] = 'n'
            elif r == _('a'):
                r = resp_all[0] = 'y'
            elif r == _('q'):
                raise util.Abort(_('user quit'))
            return r
    while chunks:
        chunk = chunks.pop()
        if isinstance(chunk, header):
            # new-file mark
            resp_file = [None]
            fixoffset = 0
            hdr = ''.join(chunk.header)
            if hdr in seen:
                consumefile()
                continue
            seen[hdr] = True
            if resp_all[0] is None:
                chunk.pretty(ui)
            r = prompt(_('examine changes to %s?') %
                       _(' and ').join(map(repr, chunk.files())))
            if r == _('y'):
                applied[chunk.filename()] = [chunk]
                if chunk.allhunks():
                    applied[chunk.filename()] += consumefile()
            else:
                consumefile()
        else:
            # new hunk
            if resp_file[0] is None and resp_all[0] is None:
                chunk.pretty(ui)
            r = prompt(_('record this change to %r?') %
                       chunk.filename())
            if r == _('y'):
                if fixoffset:
                    chunk = copy.copy(chunk)
                    chunk.toline += fixoffset
                applied[chunk.filename()].append(chunk)
            else:
                fixoffset += chunk.removed - chunk.added
    return reduce(operator.add, [h for h in applied.itervalues()
                                 if h[0].special() or len(h) > 1], [])

def record(ui, repo, *pats, **opts):
    '''interactively select changes to commit

    If a list of files is omitted, all changes reported by "hg status"
    will be candidates for recording.

    See 'hg help dates' for a list of formats valid for -d/--date.

    You will be prompted for whether to record changes to each
    modified file, and for files with multiple changes, for each
    change to use.  For each query, the following responses are
    possible:

    y - record this change
    n - skip this change

    s - skip remaining changes to this file
    f - record remaining changes to this file

    d - done, skip remaining changes and files
    a - record all changes to all remaining files
    q - quit, recording no changes

    ? - display help'''

    def record_committer(ui, repo, pats, opts):
        commands.commit(ui, repo, *pats, **opts)

    dorecord(ui, repo, record_committer, *pats, **opts)


def qrecord(ui, repo, patch, *pats, **opts):
    '''interactively record a new patch

    see 'hg help qnew' & 'hg help record' for more information and usage
    '''

    try:
        mq = extensions.find('mq')
    except KeyError:
        raise util.Abort(_("'mq' extension not loaded"))

    def qrecord_committer(ui, repo, pats, opts):
        mq.new(ui, repo, patch, *pats, **opts)

    opts = opts.copy()
    opts['force'] = True    # always 'qnew -f'
    dorecord(ui, repo, qrecord_committer, *pats, **opts)


def dorecord(ui, repo, committer, *pats, **opts):
    if not ui.interactive:
        raise util.Abort(_('running non-interactively, use commit instead'))

    def recordfunc(ui, repo, message, match, opts):
        """This is generic record driver.

        It's job is to interactively filter local changes, and accordingly
        prepare working dir into a state, where the job can be delegated to
        non-interactive commit command such as 'commit' or 'qrefresh'.

        After the actual job is done by non-interactive command, working dir
        state is restored to original.

        In the end we'll record intresting changes, and everything else will be
        left in place, so the user can continue his work.
        """
        if match.files():
            changes = None
        else:
            changes = repo.status(match=match)[:3]
            modified, added, removed = changes
            match = cmdutil.matchfiles(repo, modified + added + removed)
        diffopts = mdiff.diffopts(git=True, nodates=True)
        fp = cStringIO.StringIO()
        patch.diff(repo, repo.dirstate.parents()[0], match=match,
                   changes=changes, opts=diffopts, fp=fp)
        fp.seek(0)

        # 1. filter patch, so we have intending-to apply subset of it
        chunks = filterpatch(ui, parsepatch(fp))
        del fp

        contenders = {}
        for h in chunks:
            try: contenders.update(dict.fromkeys(h.files()))
            except AttributeError: pass

        newfiles = [f for f in match.files() if f in contenders]

        if not newfiles:
            ui.status(_('no changes to record\n'))
            return 0

        if changes is None:
            match = cmdutil.matchfiles(repo, newfiles)
            changes = repo.status(match=match)
        modified = dict.fromkeys(changes[0])

        # 2. backup changed files, so we can restore them in the end
        backups = {}
        backupdir = repo.join('record-backups')
        try:
            os.mkdir(backupdir)
        except OSError, err:
            if err.errno != errno.EEXIST:
                raise
        try:
            # backup continues
            for f in newfiles:
                if f not in modified:
                    continue
                fd, tmpname = tempfile.mkstemp(prefix=f.replace('/', '_')+'.',
                                               dir=backupdir)
                os.close(fd)
                ui.debug(_('backup %r as %r\n') % (f, tmpname))
                util.copyfile(repo.wjoin(f), tmpname)
                backups[f] = tmpname

            fp = cStringIO.StringIO()
            for c in chunks:
                if c.filename() in backups:
                    c.write(fp)
            dopatch = fp.tell()
            fp.seek(0)

            # 3a. apply filtered patch to clean repo  (clean)
            if backups:
                hg.revert(repo, repo.dirstate.parents()[0], backups.has_key)

            # 3b. (apply)
            if dopatch:
                try:
                    ui.debug(_('applying patch\n'))
                    ui.debug(fp.getvalue())
                    patch.internalpatch(fp, ui, 1, repo.root)
                except patch.PatchError, err:
                    s = str(err)
                    if s:
                        raise util.Abort(s)
                    else:
                        raise util.Abort(_('patch failed to apply'))
            del fp

            # 4. We prepared working directory according to filtered patch.
            #    Now is the time to delegate the job to commit/qrefresh or the like!

            # it is important to first chdir to repo root -- we'll call a
            # highlevel command with list of pathnames relative to repo root
            cwd = os.getcwd()
            os.chdir(repo.root)
            try:
                committer(ui, repo, newfiles, opts)
            finally:
                os.chdir(cwd)

            return 0
        finally:
            # 5. finally restore backed-up files
            try:
                for realname, tmpname in backups.iteritems():
                    ui.debug(_('restoring %r to %r\n') % (tmpname, realname))
                    util.copyfile(tmpname, repo.wjoin(realname))
                    os.unlink(tmpname)
                os.rmdir(backupdir)
            except OSError:
                pass
    return cmdutil.commit(ui, repo, recordfunc, pats, opts)

cmdtable = {
    "record":
        (record,

         # add commit options
         commands.table['^commit|ci'][1],

         _('hg record [OPTION]... [FILE]...')),
}


def extsetup():
    try:
        mq = extensions.find('mq')
    except KeyError:
        return

    qcmdtable = {
    "qrecord":
        (qrecord,

         # add qnew options, except '--force'
         [opt for opt in mq.cmdtable['qnew'][1] if opt[1] != 'force'],

         _('hg qrecord [OPTION]... PATCH [FILE]...')),
    }

    cmdtable.update(qcmdtable)