view hgext3rd/topic/revset.py @ 4410:8087e9a0f902 stable

split: make sure hg split preserve the phase of splitting cset (issue6048) Before this patch, while splitting we were not taking the phase of splitting cset into account. By default new commits would have draft changes. As we know there can two possible phases i.e draft and secret in rewriting csets thing, so this adds the handling of secret phase (as default is draft) Changes in test file reflect the added behaviour.
author Sushil khanchi <sushilkhanchi97@gmail.com>
date Sat, 02 Mar 2019 01:43:14 +0530
parents 0f015fe4f71f
children 75276f858444
line wrap: on
line source

from __future__ import absolute_import

from mercurial import (
    error,
    registrar,
    revset,
    util,
)

from . import (
    destination,
    stack,
)

try:
    mkmatcher = revset._stringmatcher
except AttributeError:
    try:
        from mercurial.utils import stringutil
        mkmatcher = stringutil.stringmatcher
    except (ImportError, AttributeError):
        mkmatcher = util.stringmatcher

revsetpredicate = registrar.revsetpredicate()

def getstringstrict(x, err):
    if x and x[0] == 'string':
        return x[1]
    raise error.ParseError(err)

@revsetpredicate('topic([string or set])')
def topicset(repo, subset, x):
    """All changesets with the specified topic or the topics of the given
    changesets. Without the argument, all changesets with any topic specified.

    If `string` starts with `re:` the remainder of the name is treated
    as a regular expression.
    """
    args = revset.getargs(x, 0, 1, 'topic takes one or no arguments')

    mutable = revset._notpublic(repo, revset.fullreposet(repo), ())

    if not args:
        return (subset & mutable).filter(lambda r: bool(repo[r].topic()))

    try:
        topic = getstringstrict(args[0], '')
    except error.ParseError:
        # not a string, but another revset
        pass
    else:
        kind, pattern, matcher = mkmatcher(topic)

        if topic.startswith('literal:') and pattern not in repo.topics:
            raise error.RepoLookupError("topic '%s' does not exist" % pattern)

        def matches(r):
            topic = repo[r].topic()
            if not topic:
                return False
            return matcher(topic)

        return (subset & mutable).filter(matches)

    s = revset.getset(repo, revset.fullreposet(repo), x)
    topics = {repo[r].topic() for r in s}
    topics.discard('')

    def matches(r):
        if r in s:
            return True
        topic = repo[r].topic()
        if not topic:
            return False
        return topic in topics

    return (subset & mutable).filter(matches)

@revsetpredicate('ngtip([branch])')
def ngtipset(repo, subset, x):
    """The untopiced tip.

    Name is horrible so that people change it.
    """
    args = revset.getargs(x, 1, 1, 'ngtip takes one argument')
    # match a specific topic
    branch = revset.getstring(args[0], 'ngtip requires a string')
    if branch == '.':
        branch = repo['.'].branch()
    return subset & revset.baseset(destination.ngtip(repo, branch))

@revsetpredicate('stack()')
def stackset(repo, subset, x):
    """All relevant changes in the current topic,

    This is roughly equivalent to 'topic(.) - obsolete' with a sorting moving
    unstable changeset after there future parent (as if evolve where already
    run).
    """
    err = 'stack takes no arguments, it works on current topic'
    revset.getargs(x, 0, 0, err)
    topic = None
    branch = None
    if repo.currenttopic:
        topic = repo.currenttopic
    else:
        branch = repo[None].branch()
    return revset.baseset(stack.stack(repo, branch=branch, topic=topic)[1:]) & subset

if util.safehasattr(revset, 'subscriptrelations'):
    def stackrel(repo, subset, x, rel, n, order):
        """This is a revset-flavored implementation of stack aliases.

        The syntax is: rev#stack[n] or rev#s[n]. Plenty of logic is borrowed
        from topic._namemap, but unlike that function, which prefers to abort
        (e.g. when stack index is too high), this returns empty set to be more
        revset-friendly.
        """
        s = revset.getset(repo, revset.fullreposet(repo), x)
        if not s:
            return revset.baseset()
        revs = []
        for r in s:
            topic = repo[r].topic()
            if topic:
                st = stack.stack(repo, topic=topic)
            else:
                st = stack.stack(repo, branch=repo[r].branch())
            if n < 0:
                st = list(st)[1:]
            else:
                st = list(st)
            try:
                rev = st[n]
            except IndexError:
                continue
            if rev == -1 and n == 0:
                continue
            if rev not in revs:
                revs.append(rev)
        return subset & revset.baseset(revs)

    revset.subscriptrelations['stack'] = stackrel
    revset.subscriptrelations['s'] = stackrel

    def topicrel(repo, subset, x, rel, n, order):
        ancestors = revset._ancestors
        descendants = revset._descendants
        subset = topicset(repo, subset, x)
        if n <= 0:
            n = -n
            return ancestors(repo, subset, x, startdepth=n, stopdepth=n + 1)
        else:
            return descendants(repo, subset, x, startdepth=n, stopdepth=n + 1)

    revset.subscriptrelations['topic'] = topicrel
    revset.subscriptrelations['t'] = topicrel