Bug report

[Tobias Dammers]

Tried it out; the workflow Dennis suggested solved the issue at hand.

Still, I think this should be considered a bug - it throws an
exception instead of exiting with a regular error message. If I
understand correctly, the issue arises when converting a repository
with a file filter would result in a situation where two identical
branches get merged. However, the point where it throws is a
considerable bit down the road, and should probably be caught much
earlier.

I'm also pretty certain that I did in fact pull from a converted
version of the same repository by accident. Somehow, hg didn't detect
that the two repositories in question weren't compatible.

Anyway, my case is resolved, so thanks a lot. Excellent support.

On Sun, Nov 21, 2010 at 6:03 PM, Patrick Mézard <pmezard at gmail.com> wrote:
> Le 21/11/10 17:48, Dennis Brakhane a écrit :
>> On Sun, Nov 21, 2010 at 4:07 PM, Tobias Dammers <tdammers at gmail.com> wrote:
>>> b) I may have accidentally pulled from a converted version of the repo
>>> into a pre-convert version, or vv. I think this explanation is more
>>> likely, IIRC it went something like this: 1. convert 2. clone to
>>> bitbucket 3. pull changes back from bitbucket. And I think step 3 may
>>> have used a different local clone than step 1. I also remember vaguely
>>> that hg asked me to merge somewhere during that process (without
>>> showing any conflicts, so I shrugged and said "OK").
>>
>> That would certainly explain things.
>>
>> I'm undecided whether this should actually considered a bug of convert or not.
>>
>> On one hand, it can only be triggered when converting a repo that has
>> (at least) two branches whose roots differ only by changes to one file, while
>> all other files in that changeset as well as all other changes in all
>> other changesets
>> are identical.
>> I can see no way how such a repo might occur in real life, unless something
>> stupid (no offence) like the above was done by accident.
>>
>> On the other hand, convert provides the filemap functionality and one could
>> argue that this implies that the corner case of two branches becoming equal
>> must be handled as well.
>>
>> Opinions, anyone?
>
> That's a bug. But what bug it is is not clear to me yet.
>
> --
> Patrick Mézard
>
>