Mercurial > evolve
view hgext3rd/topic/stack.py @ 2939:7759d040d48d
topic: provide more information when behind count cannot be computed
The behind count has been confusing for various user and the associated error
message is even worth. We try to clarify it to see if that helps.
author | Pierre-Yves David <pierre-yves.david@octobus.net> |
---|---|
date | Sun, 17 Sep 2017 19:59:25 +0200 |
parents | 9872526fc39f |
children | a61634f52742 |
line wrap: on
line source
# stack.py - code related to stack workflow # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. from mercurial.i18n import _ from mercurial import ( destutil, context, error, node, phases, util, ) from .evolvebits import builddependencies, _singlesuccessor short = node.short # TODO: compat if not util.safehasattr(context.basectx, 'orphan'): context.basectx.orphan = context.basectx.unstable if not util.safehasattr(context.basectx, 'isunstable'): context.basectx.isunstable = context.basectx.troubled class stack(object): """object represent a stack and common logic associated to it.""" def __init__(self, repo, branch=None, topic=None): self._repo = repo self.branch = branch self.topic = topic self.behinderror = None if topic is not None and branch is not None: raise error.ProgrammingError('both branch and topic specified (not defined yet)') elif topic is not None: trevs = repo.revs("topic(%s) - obsolete()", topic) elif branch is not None: trevs = repo.revs("branch(%s) - public() - obsolete() - topic()", branch) else: raise error.ProgrammingError('neither branch and topic specified (not defined yet)') self._revs = trevs def __iter__(self): return iter(self.revs) def __getitem__(self, index): return self.revs[index] def index(self, item): return self.revs.index(item) @util.propertycache def _dependencies(self): deps, rdeps = builddependencies(self._repo, self._revs) repo = self._repo srcpfunc = repo.changelog.parentrevs ### post process to skip over possible gaps in the stack # # For example in the following situation, we need to detect that "t3" # indirectly depends on t2. # # o t3 # | # o other # | # o t2 # | # o t1 pmap = {} def pfuncrev(repo, rev): """a special "parent func" that also consider successors""" parents = pmap.get(rev) if parents is None: parents = [repo[_singlesuccessor(repo, repo[p])].rev() for p in srcpfunc(rev) if 0 <= p] pmap[rev] = parents return parents revs = self._revs stackrevs = set(self._revs) for root in [r for r in revs if not deps[r]]: seen = set() stack = [root] while stack: current = stack.pop() for p in pfuncrev(repo, current): if p in seen: continue seen.add(p) if p in stackrevs: rdeps[p].add(root) deps[root].add(p) elif phases.public < repo[p].phase(): # traverse only if we did not found a proper candidate stack.append(p) return deps, rdeps @util.propertycache def revs(self): # some duplication/change from _orderrevs because we use a post # processed dependency graph. # Step 1: compute relation of revision with each other dependencies, rdependencies = self._dependencies dependencies = dependencies.copy() rdependencies = rdependencies.copy() # Step 2: Build the ordering # Remove the revisions with no dependency(A) and add them to the ordering. # Removing these revisions leads to new revisions with no dependency (the # one depending on A) that we can remove from the dependency graph and add # to the ordering. We progress in a similar fashion until the ordering is # built solvablerevs = [r for r in sorted(dependencies.keys()) if not dependencies[r]] revs = [] while solvablerevs: rev = solvablerevs.pop() for dependent in rdependencies[rev]: dependencies[dependent].remove(rev) if not dependencies[dependent]: solvablerevs.append(dependent) del dependencies[rev] revs.append(rev) revs.extend(sorted(dependencies)) # step 3: add t0 if revs: pt1 = self._repo[revs[0]].p1() else: pt1 = self._repo['.'] if pt1.obsolete(): pt1 = self._repo[_singlesuccessor(self._repo, pt1)] revs.insert(0, pt1.rev()) return revs @util.propertycache def changesetcount(self): return len(self._revs) @util.propertycache def troubledcount(self): return len([r for r in self._revs if self._repo[r].isunstable()]) @util.propertycache def heads(self): revs = self.revs[1:] deps, rdeps = self._dependencies return [r for r in revs if not rdeps[r]] @util.propertycache def behindcount(self): revs = self.revs[1:] deps, rdeps = self._dependencies if revs: minroot = [min(r for r in revs if not deps[r])] try: dest = destutil.destmerge(self._repo, action='rebase', sourceset=minroot, onheadcheck=False) return len(self._repo.revs("only(%d, %ld)", dest, minroot)) except error.NoMergeDestAbort: return 0 except error.ManyMergeDestAbort as exc: # XXX we should make it easier for upstream to provide the information self.behinderror = str(exc).split('-', 1)[0].rstrip() return -1 return 0 @util.propertycache def branches(self): branches = sorted(set(self._repo[r].branch() for r in self._revs)) if not branches: branches = set([self._repo[None].branch()]) return branches def labelsgen(prefix, labelssuffix): """ Takes a label prefix and a list of suffixes. Returns a string of the prefix formatted with each suffix separated with a space. """ return ' '.join(prefix % suffix for suffix in labelssuffix) def showstack(ui, repo, branch=None, topic=None, opts=None): if opts is None: opts = {} if topic is not None and branch is not None: msg = 'both branch and topic specified [%s]{%s}(not defined yet)' msg %= (branch, topic) raise error.ProgrammingError(msg) elif topic is not None: prefix = 't' if topic not in repo.topics: raise error.Abort(_('cannot resolve "%s": no such topic found') % topic) elif branch is not None: prefix = 'b' else: raise error.ProgrammingError('neither branch and topic specified (not defined yet)') fm = ui.formatter('topicstack', opts) prev = None entries = [] idxmap = {} label = 'topic' if topic == repo.currenttopic: label = 'topic.active' data = stackdata(repo, branch=branch, topic=topic) empty = False if data['changesetcount'] == 0: empty = True if topic is not None: fm.plain(_('### topic: %s') % ui.label(topic, label), label='topic.stack.summary.topic') if 1 < data['headcount']: fm.plain(' (') fm.plain('%d heads' % data['headcount'], label='topic.stack.summary.headcount.multiple') fm.plain(')') fm.plain('\n') fm.plain(_('### branch: %s') % '+'.join(data['branches']), # XXX handle multi branches label='topic.stack.summary.branches') if topic is None: if 1 < data['headcount']: fm.plain(' (') fm.plain('%d heads' % data['headcount'], label='topic.stack.summary.headcount.multiple') fm.plain(')') else: if data['behindcount'] == -1: fm.plain(', ') fm.plain('ambigious rebase destination - %s' % data['behinderror'], label='topic.stack.summary.behinderror') elif data['behindcount']: fm.plain(', ') fm.plain('%d behind' % data['behindcount'], label='topic.stack.summary.behindcount') fm.plain('\n') if empty: fm.plain(_("(stack is empty)\n")) for idx, r in enumerate(stack(repo, branch=branch, topic=topic), 0): ctx = repo[r] # special case for t0, b0 as it's hard to plugin into rest of the logic if idx == 0: # t0, b0 can be None if r == -1: continue entries.append((idx, False, ctx)) prev = ctx.rev() continue p1 = ctx.p1() p2 = ctx.p2() if p1.obsolete(): p1 = repo[_singlesuccessor(repo, p1)] if p2.node() != node.nullid: entries.append((idxmap.get(p1.rev()), False, p1)) entries.append((idxmap.get(p2.rev()), False, p2)) elif p1.rev() != prev and p1.node() != node.nullid: entries.append((idxmap.get(p1.rev()), False, p1)) entries.append((idx, True, ctx)) idxmap[ctx.rev()] = idx prev = r # super crude initial version for idx, isentry, ctx in entries[::-1]: states = [] iscurrentrevision = repo.revs('%d and parents()', ctx.rev()) if not isentry: symbol = '^' # "base" is kind of a "ghost" entry # skip other label for them (no current, no unstable) states = ['base'] elif ctx.orphan(): # current revision can be unstable also, so in that case show both # the states and the symbol '@' (issue5553) if iscurrentrevision: states.append('current') symbol = '@' symbol = '$' states.append('unstable') elif iscurrentrevision: states.append('current') symbol = '@' else: symbol = ':' states.append('clean') fm.startitem() fm.data(isentry=isentry) if idx is None: fm.plain(' ') if ui.verbose: fm.plain(' ') else: fm.write('topic.stack.index', '%s%%d' % prefix, idx, label='topic.stack.index ' + labelsgen('topic.stack.index.%s', states)) if ui.verbose: fm.write('topic.stack.shortnode', '(%s)', short(ctx.node()), label='topic.stack.shortnode ' + labelsgen('topic.stack.shortnode.%s', states)) fm.write('topic.stack.state.symbol', '%s', symbol, label='topic.stack.state ' + labelsgen('topic.stack.state.%s', states)) fm.plain(' ') fm.write('topic.stack.desc', '%s', ctx.description().splitlines()[0], label='topic.stack.desc ' + labelsgen('topic.stack.desc.%s', states)) fm.condwrite(states != ['clean'] and idx is not None, 'topic.stack.state', ' (%s)', fm.formatlist(states, 'topic.stack.state'), label='topic.stack.state ' + labelsgen('topic.stack.state.%s', states)) fm.plain('\n') fm.end() def stackdata(repo, branch=None, topic=None): """get various data about a stack :changesetcount: number of non-obsolete changesets in the stack :troubledcount: number on troubled changesets :headcount: number of heads on the topic :behindcount: number of changeset on rebase destination """ data = {} current = stack(repo, branch, topic) data['changesetcount'] = current.changesetcount data['troubledcount'] = current.troubledcount data['headcount'] = len(current.heads) data['behindcount'] = current.behindcount data['behinderror'] = current.behinderror data['branches'] = current.branches return data