annotate hgext/hbisect.py @ 5736:6e79d5e0e541

bisect: keep history of all bad revisions

- use a single state dict
- find the minimum bad revision

# bisect extension for mercurial
#
# Copyright 2005, 2006 Benoit Boissinot <[e-mail]>
# Inspired by git bisect, extension skeleton taken from mq.py.
#
# This software may be used and distributed according to the terms
# of the GNU General Public License, incorporated herein by reference.

from mercurial.i18n import _
from mercurial import hg, util, cmdutil
import os, array

class bisect(object):
    """dichotomic search in the DAG of changesets"""
    def __init__(self, ui, repo):
        self.repo = repo
        self.ui = ui
        self.state = {'good': [], 'bad': [], 'skip': []}

        p = self.repo.join("bisect.state")
        if os.path.exists(p):
            for l in self.repo.opener("bisect.state"):
                kind, node = l[:-1].split()
                node = self.repo.lookup(node)
                if kind not in self.state:
                    raise util.Abort(_("unknown bisect kind %s") % kind)
                self.state[kind].append(node)

    def write(self):
        f = self.repo.opener("bisect.state", "w")
        for kind in self.state:
            for node in self.state[kind]:
                f.write("%s %s\n" % (kind, hg.hex(node)))

    def init(self):
        """start a new bisection"""
        p = self.repo.join("bisect.state")
        if os.path.exists(p):
            os.unlink(p)

    def bisect(self):
        cl = self.repo.changelog
        clparents = cl.parentrevs
        # only the earliest bad revision matters
        badrev = min([cl.rev(n) for n in self.state['bad']])
        bad = cl.node(badrev)

        # build ancestors array
        ancestors = [[]] * (cl.count() + 1) # an extra for [-1]

        # clear good revs from array
        for node in self.state['good']:
            ancestors[cl.rev(node)] = None
        for rev in xrange(cl.count(), -1, -1):
            if ancestors[rev] is None:
                for prev in clparents(rev):
                    ancestors[prev] = None

        if ancestors[badrev] is None:
            raise util.Abort(_("Inconsistent state, %s:%s is good and bad")
                             % (badrev, hg.short(bad)))

        # accumulate ancestor lists
        for rev in xrange(badrev + 1):
            if ancestors[rev] == []:
                p1, p2 = clparents(rev)
                a1, a2 = ancestors[p1], ancestors[p2]
                if a1:
                    if a2:
                        # merge ancestor lists
                        a = dict.fromkeys(a2)
                        a.update(dict.fromkeys(a1))
                        a[rev] = None
                        ancestors[rev] = array.array("l", a.keys())
                    else:
                        ancestors[rev] = a1 + array.array("l", [rev])
                elif a2:
                    ancestors[rev] = a2 + array.array("l", [rev])
                else:
                    ancestors[rev] = array.array("l", [rev])

        if badrev not in ancestors[badrev]:
            raise util.Abort(_("Could not find the first bad revision"))

        # have we narrowed it down to one entry?
        tot = len(ancestors[badrev])
        if tot == 1:
            return (bad, 0)

        # find the best node to test
        best_rev = None
        best_len = -1
        skip = dict.fromkeys([cl.rev(n) for n in self.state['skip']])
        for n in ancestors[badrev]:
            if n in skip:
                continue
            a = len(ancestors[n]) # number of ancestors
            b = tot - a # number of non-ancestors
            value = min(a, b) # how good is this test?
            if value > best_len:
                best_len = value
                best_rev = n
        assert best_rev is not None
        best_node = cl.node(best_rev)

        return (best_node, tot)

    def next(self):
        """find and update to the next revision to test"""
        if self.state['good'] and self.state['bad']:
            node, changesets = self.bisect()

            if changesets == 0:
                self.ui.write(_("The first bad revision is:\n"))
                displayer = cmdutil.show_changeset(self.ui, self.repo, {})
                displayer.show(changenode=node)
            elif node is not None:
                # compute the approximate number of remaining tests
                tests, size = 0, 2
                while size <= changesets:
                    tests, size = tests + 1, size * 2
                rev = self.repo.changelog.rev(node)
                self.ui.write(_("Testing changeset %s:%s "
                                "(%s changesets remaining, ~%s tests)\n")
                              % (rev, hg.short(node), changesets, tests))
                cmdutil.bail_if_changed(self.repo)
                return hg.clean(self.repo, node)

    def good(self, rev=None):
        """mark revision as good and update to the next revision to test"""
        self.state['good'].append(self.repo.lookup(rev or '.'))
        self.write()
        return self.next()

    def skip(self, rev=None):
        """mark revision as skipped and update to the next revision to test"""
        self.state['skip'].append(self.repo.lookup(rev or '.'))
        self.write()
        return self.next()

    def bad(self, rev=None):
        """mark revision as bad and update to the next revision to test"""
        self.state['bad'].append(self.repo.lookup(rev or '.'))
        self.write()
        return self.next()

def bisect_run(ui, repo, node=None, extra=None,
               reset=None, good=None, bad=None, skip=None):
    """Subdivision search of changesets

This extension helps to find changesets which introduce problems.
To use, mark the earliest changeset you know exhibits the problem
as bad, then mark the latest changeset which is free from the problem
as good. Bisect will update your working directory to a revision for
testing. Once you have performed tests, mark the working directory
as bad or good and bisect will either update to another candidate
changeset or announce that it has found the bad revision.

Note: bisect expects bad revisions to be descendants of good revisions.
If you are looking for the point at which a problem was fixed, then make
the problem-free state "bad" and the problematic state "good."

    """
    # backward compatibility
    if node in "good bad reset init".split():
        ui.warn(_("(use of 'hg bisect <cmd>' is deprecated)\n"))
        cmd, node, extra = node, extra, None
        if cmd == "good":
            good = True
        elif cmd == "bad":
            bad = True
        else:
            reset = True
    elif extra or good + bad + skip + reset > 1:
        raise util.Abort("Incompatible arguments")

    b = bisect(ui, repo)
    if good:
        return b.good(node)
    elif bad:
        return b.bad(node)
    elif skip:
        return b.skip(node)
    elif reset:
        return b.init()
    else:
        return b.next()

cmdtable = {
    "bisect": (bisect_run,
               [('r', 'reset', False, _('reset bisect state')),
                ('g', 'good', False, _('mark changeset good')),
                ('b', 'bad', False, _('mark changeset bad')),
                ('s', 'skip', False, _('skip testing changeset'))],
               _("hg bisect [-gbsr] [REV]"))
}