<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Nov 13, 2017 at 3:07 PM, Augie Fackler <span dir="ltr"><<a href="mailto:raf@durin42.com" target="_blank">raf@durin42.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><span>On Sun, Nov 12, 2017 at 06:06:37PM +0100, Denis Laxalde wrote:<br>
> # HG changeset patch<br>
> # User Denis Laxalde <<a href="mailto:denis@laxalde.org" target="_blank">denis@laxalde.org</a>><br>
> # Date 1510489041 -3600<br>
> # Sun Nov 12 13:17:21 2017 +0100<br>
> # Node ID 25d1891bd300bc527cb7e878532c85<wbr>12ff5489de<br>
> # Parent 1d16e28225e01d2fac9f31ed058ae1<wbr>f4c00bb9da<br>
> # EXP-Topic rebase-obsolete<br>
> rebase: exclude obsoletes without a successor in destination (issue5300)<br></span></blockquote><div><br></div><div>If I read it right, this line should be something like "rebase: exclude *descendants of* obsoletes without..."</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><span>
<br>
</span>I took patches 1-3. I think this is right. I'm 90% sure this is<br>
right. But I'm a little tired, and so I want Martin's more careful eye<br>
on it if he has time.<br></blockquote><div><br></div><div>Thanks for asking. I think this makes sense. Thanks for working on this, Denis!</div><div><br></div><div>I think a general thing to note is that if a commit in the rebase set has become obsolete, then there's a conflict between the rebase invocation and the obsmarkers. The rebase command tells us to move the descendants onto the rebase destination, while the obsmarkers tell us to move them onto the obsolete commit's successor. Since we don't keep track of some default rebase destination (like git's upstream concept), any "hg rebase" invocation is a one-off as far as hg is concerned, so I think it makes sense to resolve the conflict in favor of the obsmarkers. That probably didn't make much sense, so let me give an example:</div><div><br></div><div><div>o B'</div><div>|</div><div>o O</div><div>|</div><div>| o U</div><div>| |</div><div>| | o C</div><div>| | |</div><div>| | x B</div><div>| |/</div><div>| .</div><div>|/</div><div>o A</div><div><br></div><div>Let's say C's upstream was somehow configured to be U here, and that B had been rebased onto O instead. If the user then runs "hg evolve -r C", you could argue that there would be a similar conflict between rebase/upstream that want to move C onto U and obsmarkers that want to move C onto B'. However, since we don't have that feature, B' is the only natural destination.</div></div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5"><br>
><br>
> In the following example, when trying to rebase 3:: onto 2, the rebase<br>
> will abort with "this rebase will cause divergence from: 4":<br>
><br>
> o 7 f<br>
> |<br>
> | o 6 e<br>
> | |<br>
> | o 5 d'<br>
> | |<br>
> x | 4 d (rewritten as 5)<br>
> |/<br>
> o 3 c<br>
> |<br>
> | o 2 x<br>
> | |<br>
> o | 1 b<br>
> |/<br>
> o 0 a<br>
><br>
> By excluding obsolete changesets without a successor in destination (4<br>
> in the example above) and their descendants, we make rebase work in this<br>
> case, thus giving:<br>
><br>
> o 11 e<br>
> |<br>
> o 10 d'<br>
> |<br>
> o 9 c<br>
> |<br>
> o 8 b<br>
> |<br>
> | o 7 f<br>
> | |<br>
> | | x 6 e (rewritten using rebase as 11)<br>
> | | |<br>
> | | x 5 d' (rewritten using rebase as 10)<br>
> | | |<br>
> | x | 4 d<br>
> | |/<br>
> | x 3 c (rewritten using rebase as 9)<br>
> | |<br>
> o | 2 x<br>
> | |<br>
> | x 1 b (rewritten using rebase as 8)<br>
> |/<br>
> o 0 a<br>
><br>
> where branch 4:: is left behind while branch 5:: is rebased as expected.<br>
><br>
> The rationale is that users may not be interested in rebasing orphan<br>
> changesets when specifying a rebase set that include them but would<br>
> still want "stable" ones to be rebased. Currently, the user is suggested<br>
> to allow divergence (but probably does not want it) or they must specify<br>
> a rebase set excluding problematic changesets (which might be a bit<br>
> cumbersome). The approach proposed here corresponds to "Option 2" in<br>
> <a href="https://www.mercurial-scm.org/wiki/CEDRebase" rel="noreferrer" target="_blank">https://www.mercurial-scm.org/<wbr>wiki/CEDRebase</a>.<br></div></div></blockquote><div><br></div><div>I think I'd still want "Option 3", but I don't think this patch prevents that from happening later.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5">
><br>
> .. feature::<br>
><br>
> Let 'hg rebase' avoid content-divergence when obsolete changesets are<br>
> present in the rebase set by skipping possible sources of divergence.<br></div></div></blockquote><div><br></div><div>Similar to what I said above, I think it would help to mention descendants here. While "possible sources of divergence" is vague enough to be correct, it would be helpful if we could clarify it.</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5">
><br>
><br></div></div></blockquote><div><br></div><div>Just to make sure, the stuff below is not part of the release notes, is it? I'm not familiar with the syntax, so I can't tell (and I'm too lazy to test it).</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5">
> We extend _computeobsoletenotrebased() so that it also return a set of<br>
> obsolete changesets in rebaseset without a successor in destination.<br>
> This 'obsoletewithoutsuccessorindes<wbr>tination' is then stored as an<br>
> attribute of rebaseruntime and used in _performrebasesubset() to:<br>
><br>
> * filter out descendants of these changesets from the revisions to<br>
> rebase;<br>
> * issue a message about these revisions being skipped.<br>
><br>
> This only occurs if 'evolution.allowdivergence' option is off and<br>
> 'rebaseskipobsolete' is on.<br>
><br>
> diff --git a/hgext/rebase.py b/hgext/rebase.py<br>
> --- a/hgext/rebase.py<br>
> +++ b/hgext/rebase.py<br>
> @@ -179,6 +179,7 @@ class rebaseruntime(object):<br>
> # other extensions<br>
> self.keepopen = opts.get('keepopen', False)<br>
> self.obsoletenotrebased = {}<br>
> + self.obsoletewithoutsuccessori<wbr>ndestination = set()<br>
><br>
> @property<br>
> def repo(self):<br>
> @@ -311,9 +312,10 @@ class rebaseruntime(object):<br>
> if not self.ui.configbool('experiment<wbr>al', 'rebaseskipobsolete'):<br>
> return<br>
> obsoleteset = set(obsoleterevs)<br>
> - self.obsoletenotrebased = _computeobsoletenotrebased(sel<wbr>f.repo,<br>
> - obsoleteset, destmap)<br>
> + self.obsoletenotrebased, self.obsoletewithoutsuccessori<wbr>ndestination = \<br>
> + _computeobsoletenotrebased(sel<wbr>f.repo, obsoleteset, destmap)<br>
> skippedset = set(self.obsoletenotrebased)<br>
> + skippedset.update(self.obsolet<wbr>ewithoutsuccessorindestination<wbr>)<br>
> _checkobsrebase(self.repo, self.ui, obsoleteset, skippedset)<br>
><br>
> def _prepareabortorcontinue(self, isabort):<br>
> @@ -419,12 +421,26 @@ class rebaseruntime(object):<br>
> def _performrebasesubset(self, tr, subset, pos, total):<br>
> repo, ui, opts = self.repo, self.ui, self.opts<br>
> sortedrevs = repo.revs('sort(%ld, -topo)', subset)<br>
> + allowdivergence = self.ui.configbool(<br>
> + 'experimental', 'evolution.allowdivergence')<br>
> + if not allowdivergence:<br>
> + sortedrevs -= repo.revs(<br>
> + 'descendants(%ld) and not %ld',<br>
> + self.obsoletewithoutsuccessori<wbr>ndestination,<br>
> + self.obsoletewithoutsuccessori<wbr>ndestination,<br>
> + )<br>
> for rev in sortedrevs:<br>
> dest = self.destmap[rev]<br>
> ctx = repo[rev]<br>
> desc = _ctxdesc(ctx)<br>
> if self.state[rev] == rev:<br>
> ui.status(_('already rebased %s\n') % desc)<br>
> + elif (not allowdivergence<br>
> + and rev in self.obsoletewithoutsuccessori<wbr>ndestination):<br>
> + msg = _('note: not rebasing %s and its descendants as '<br>
> + 'this would cause divergences\n') % desc<br>
> + repo.ui.status(msg)<br>
> + self.skipped.add(rev)<br>
> elif rev in self.obsoletenotrebased:<br>
> succ = self.obsoletenotrebased[rev]<br>
> if succ is None:<br>
> @@ -1616,11 +1632,16 @@ def _filterobsoleterevs(repo, revs):<br>
> return set(r for r in revs if repo[r].obsolete())<br>
><br>
> def _computeobsoletenotrebased(rep<wbr>o, rebaseobsrevs, destmap):<br>
> - """return a mapping obsolete => successor for all obsolete nodes to be<br>
> - rebased that have a successors in the destination<br>
> + """Return (obsoletenotrebased, obsoletewithoutsuccessorindest<wbr>ination).<br>
> +<br>
> + `obsoletenotrebased` is a mapping mapping obsolete => successor for all<br>
> + obsolete nodes to be rebased given in `rebaseobsrevs`.<br>
><br>
> - obsolete => None entries in the mapping indicate nodes with no successor"""<br>
> + `obsoletewithoutsuccessorindes<wbr>tination` is a set with obsolete revisions<br>
> + without a successor in destination.<br>
> + """<br>
> obsoletenotrebased = {}<br>
> + obsoletewithoutsuccessorindest<wbr>ination = set([])<br>
><br>
> assert repo.filtername is None<br>
> cl = repo.changelog<br>
> @@ -1641,8 +1662,17 @@ def _computeobsoletenotrebased(rep<wbr>o, reb<br>
> if cl.isancestor(succnode, destnode):<br>
> obsoletenotrebased[srcrev] = nodemap[succnode]<br>
> break<br>
> + else:<br>
> + # Unless 'srcrev' only has one successor and it's not in<br>
> + # rebaseset, we might skip it and its descendants as it has no<br>
> + # successor in destination. Otherwise, we'll issue the<br>
> + # divergence warning and abort.<br>
> + if (len(successors) > 1</div></div></blockquote><div><br></div><div>This line doesn't seem necessary, but feel free to keep if you think it helps readability (I find it slightly distracting).</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5"> </div></div></blockquote><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5">
> + and any(nodemap[s] in destmap<br>
> + for s in successors if s != srcnode)):<br>
> + obsoletewithoutsuccessorindest<wbr>ination.add(srcrev)<br>
><br>
> - return obsoletenotrebased<br>
> + return obsoletenotrebased, obsoletewithoutsuccessorindest<wbr>ination<br>
><br>
> def summaryhook(ui, repo):<br>
> if not repo.vfs.exists('rebasestate')<wbr>:<br>
> diff --git a/tests/test-rebase-obsolete.t b/tests/test-rebase-obsolete.t<br>
> --- a/tests/test-rebase-obsolete.t<br>
> +++ b/tests/test-rebase-obsolete.t<br>
> @@ -987,6 +987,208 @@ Create the changes that we will rebase<br>
> rebasing 21:7bdc8a87673d "dummy change" (tip)<br>
> $ cd ..<br>
><br>
> +Divergence cases due to obsolete changesets<br>
> +-----------------------------<wbr>--------------<br>
> +<br>
> +We should ignore branches with unstable changesets when they are based on an<br>
> +obsolete changeset which successor is in rebase set.<br>
> +<br>
> + $ hg init divergence<br>
> + $ cd divergence<br>
> + $ cat >> .hg/hgrc << EOF<br>
> + > [extensions]<br>
> + > strip =<br>
> + > [alias]<br>
> + > strip = strip --no-backup --quiet<br>
> + > [templates]<br>
> + > instabilities = '{rev}:{node|short} {desc|firstline}{if(instabilit<wbr>ies," ({instabilities})")}\n'<br>
> + > EOF<br>
> +<br>
> + $ hg debugdrawdag <<EOF<br>
> + > e f<br>
> + > | |<br>
> + > d' d # replace: d -> d'<br>
> + > \ /<br>
> + > c<br>
> + > |<br>
> + > x b<br>
> + > \|<br>
> + > a<br>
> + > EOF<br>
> + $ hg log -G -r 'a'::<br>
> + o 7:1143e9adc121 f<br>
> + |<br>
> + | o 6:d60ebfa0f1cb e<br>
> + | |<br>
> + | o 5:027ad6c5830d d'<br>
> + | |<br>
> + x | 4:76be324c128b d (rewritten using replace as 5:027ad6c5830d)<br>
> + |/<br>
> + o 3:a82ac2b38757 c<br>
> + |<br>
> + | o 2:630d7c95eff7 x<br>
> + | |<br>
> + o | 1:488e1b7e7341 b<br>
> + |/<br>
> + o 0:b173517d0057 a<br>
> +<br>
> +<br>
> +Changeset d and its descendants are excluded to avoid divergence of d, which<br>
> +would occur because the successor of d (d') is also in rebaseset. As a<br>
> +consequence f (descendant of d) is left behind.<br>
> +<br>
> + $ hg rebase -b 'e' -d 'x'<br>
> + rebasing 1:488e1b7e7341 "b" (b)<br>
> + rebasing 3:a82ac2b38757 "c" (c)<br>
> + rebasing 5:027ad6c5830d "d'" (d')<br>
> + rebasing 6:d60ebfa0f1cb "e" (e)<br>
> + note: not rebasing 4:76be324c128b "d" (d) and its descendants as this would cause divergences<br></div></div></blockquote><div><br></div><div>The plural "divergences" sounds wrong to me, but I'm not sure. Augie (or others), since you're a native English speaker, do you think it should be singular or plural?</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-m_2749144116843584576HOEnZb"><div class="gmail-m_2749144116843584576h5">
> + $ hg log -G -r 'a'::<br>
> + o 11:eb6d63fc4ed5 e<br>
> + |<br>
> + o 10:44d8c724a70c d'<br>
> + |<br>
> + o 9:d008e6b4d3fd c<br>
> + |<br>
> + o 8:67e8f4a16c49 b<br>
> + |<br>
> + | o 7:1143e9adc121 f<br>
> + | |<br>
> + | | x 6:d60ebfa0f1cb e (rewritten using rebase as 11:eb6d63fc4ed5)<br>
> + | | |<br>
> + | | x 5:027ad6c5830d d' (rewritten using rebase as 10:44d8c724a70c)<br>
> + | | |<br>
> + | x | 4:76be324c128b d (rewritten using replace as 5:027ad6c5830d)<br>
> + | |/<br>
> + | x 3:a82ac2b38757 c (rewritten using rebase as 9:d008e6b4d3fd)<br>
> + | |<br>
> + o | 2:630d7c95eff7 x<br>
> + | |<br>
> + | x 1:488e1b7e7341 b (rewritten using rebase as 8:67e8f4a16c49)<br>
> + |/<br>
> + o 0:b173517d0057 a<br>
> +<br>
> + $ hg strip -r 8:<br>
> +<br>
> +If the rebase set has an obsolete (d) with a unique successor (d') outside the<br>
> +rebase set and not in destination, we still get the divergence warning. By<br>
> +allowing divergence, we can perform the rebase.<br>
> +<br>
> + $ hg rebase -r 'c'::'f' -d 'x'<br>
> + abort: this rebase will cause divergences from: 76be324c128b<br>
> + (to force the rebase please set experimental.evolution.allowdi<wbr>vergence=True)<br>
> + [255]<br>
> + $ hg rebase --config experimental.evolution.allowdi<wbr>vergence=true -r 'c'::'f' -d 'x'<br>
> + rebasing 3:a82ac2b38757 "c" (c)<br>
> + rebasing 4:76be324c128b "d" (d)<br>
> + rebasing 7:1143e9adc121 "f" (f tip)<br>
> + $ hg log -G -r 'a':: -T instabilities<br>
> + o 10:e1744ea07510 f<br>
> + |<br>
> + o 9:e2b36ea9a0a0 d (content-divergent)<br>
> + |<br>
> + o 8:6a0376de376e c<br>
> + |<br>
> + | x 7:1143e9adc121 f<br>
> + | |<br>
> + | | o 6:d60ebfa0f1cb e (orphan)<br>
> + | | |<br>
> + | | o 5:027ad6c5830d d' (orphan content-divergent)<br>
> + | | |<br>
> + | x | 4:76be324c128b d<br>
> + | |/<br>
> + | x 3:a82ac2b38757 c<br>
> + | |<br>
> + o | 2:630d7c95eff7 x<br>
> + | |<br>
> + | o 1:488e1b7e7341 b<br>
> + |/<br>
> + o 0:b173517d0057 a<br>
> +<br>
> + $ hg strip -r 8:<br>
> +<br>
> +(Not skipping obsoletes means that divergence is allowed.)<br>
> +<br>
> + $ hg rebase --config experimental.rebaseskipobsolet<wbr>e=false -r 'c'::'f' -d 'x'<br>
> + rebasing 3:a82ac2b38757 "c" (c)<br>
> + rebasing 4:76be324c128b "d" (d)<br>
> + rebasing 7:1143e9adc121 "f" (f tip)<br>
> +<br>
> + $ hg strip -r 0:<br>
> +<br>
> +Similar test on a more complex graph<br>
> +<br>
> + $ hg debugdrawdag <<EOF<br>
> + > g<br>
> + > |<br>
> + > f e<br>
> + > | |<br>
> + > e' d # replace: e -> e'<br>
> + > \ /<br>
> + > c<br>
> + > |<br>
> + > x b<br>
> + > \|<br>
> + > a<br>
> + > EOF<br>
> + $ hg log -G -r 'a':<br>
> + o 8:2876ce66c6eb g<br>
> + |<br>
> + | o 7:3ffec603ab53 f<br>
> + | |<br>
> + x | 6:e36fae928aec e (rewritten using replace as 5:63324dc512ea)<br>
> + | |<br>
> + | o 5:63324dc512ea e'<br>
> + | |<br>
> + o | 4:76be324c128b d<br>
> + |/<br>
> + o 3:a82ac2b38757 c<br>
> + |<br>
> + | o 2:630d7c95eff7 x<br>
> + | |<br>
> + o | 1:488e1b7e7341 b<br>
> + |/<br>
> + o 0:b173517d0057 a<br>
> +<br>
> + $ hg rebase -b 'f' -d 'x'<br>
> + rebasing 1:488e1b7e7341 "b" (b)<br>
> + rebasing 3:a82ac2b38757 "c" (c)<br>
> + rebasing 5:63324dc512ea "e'" (e')<br>
> + rebasing 7:3ffec603ab53 "f" (f)<br>
> + rebasing 4:76be324c128b "d" (d)<br>
> + note: not rebasing 6:e36fae928aec "e" (e) and its descendants as this would cause divergences<br>
> + $ hg log -G -r 'a':<br>
> + o 13:a1707a5b7c2c d<br>
> + |<br>
> + | o 12:ef6251596616 f<br>
> + | |<br>
> + | o 11:b6f172e64af9 e'<br>
> + |/<br>
> + o 10:d008e6b4d3fd c<br>
> + |<br>
> + o 9:67e8f4a16c49 b<br>
> + |<br>
> + | o 8:2876ce66c6eb g<br>
> + | |<br>
> + | | x 7:3ffec603ab53 f (rewritten using rebase as 12:ef6251596616)<br>
> + | | |<br>
> + | x | 6:e36fae928aec e (rewritten using replace as 5:63324dc512ea)<br>
> + | | |<br>
> + | | x 5:63324dc512ea e' (rewritten using rebase as 11:b6f172e64af9)<br>
> + | | |<br>
> + | x | 4:76be324c128b d (rewritten using rebase as 13:a1707a5b7c2c)<br>
> + | |/<br>
> + | x 3:a82ac2b38757 c (rewritten using rebase as 10:d008e6b4d3fd)<br>
> + | |<br>
> + o | 2:630d7c95eff7 x<br>
> + | |<br>
> + | x 1:488e1b7e7341 b (rewritten using rebase as 9:67e8f4a16c49)<br>
> + |/<br>
> + o 0:b173517d0057 a<br>
> +<br>
> +<br>
> + $ cd ..<br>
> +<br>
> Rebase merge where successor of one parent is equal to destination (issue5198)<br>
><br>
> $ hg init p1-succ-is-dest<br>
> ______________________________<wbr>_________________<br>
> Mercurial-devel mailing list<br>
> <a href="mailto:Mercurial-devel@mercurial-scm.org" target="_blank">Mercurial-devel@mercurial-scm.<wbr>org</a><br>
> <a href="https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel" rel="noreferrer" target="_blank">https://www.mercurial-scm.org/<wbr>mailman/listinfo/mercurial-dev<wbr>el</a><br>
</div></div></blockquote></div><br></div></div>