On 17.03.2014, at 18:01, Junio C Hamano <gits...@pobox.com> wrote:

> Torsten Bögershausen <tbo...@web.de> writes:
>> On 2014-03-14 23.09, Junio C Hamano wrote:
>>> * ap/remote-hg-skip-null-bookmarks (2014-01-02) 1 commit
>>> - remote-hg: do not fail on invalid bookmarks
>>> Reported to break tests ($gmane/240005)
>>> Expecting a reroll.
>> I wonder what should happen here.
>> The change breaks all the tests in test-hg-hg-git.sh
>> (And the breakage may prevent us from detecting other breakages)
>> The ideal situation would be to have an extra test case for the problem
>> which we try to fix with this patch.
>> Antoine, is there any way to make your problem reproducable ?
>> And based on that, to make a patch which passes all test cases ?
> After re-reading the thread briefly (there're just five messages)
>  http://thread.gmane.org/gmane.comp.version-control.git/239797/focus=240069

For some reason, that link does not contain all messages from that conversation 
(unfortunately, I have seen GMane do that on multiple occasions. I hence try 
not to rely on it for reviewing email history -- I just don't trust it). In 
particular, it misses this crucial post:

I call it crucial because it describes how to make a reproducible test cases 
out of this, in which a legitimate hg repository leads to a remote-hg error 
preventing the user from normal operation.

> I think the "breakage" the patch tries to fix seems to be of dubious
> nature in the first place ("I don't know how I ended-up with such a
> bookmark", Antoine says in $gmane/239800), and it has been in
> "Expecting a reroll" state in response to "I will try to come-up
> with an improved version" in $gmane/240069 but nothing has happened
> for a few months.
> At this point I think it would be OK for me to discard the topic
> (without prejudice); if the root cause of the issue (if there is
> one) and a proper fix is discovered in the future, the topic can
> come back with a fresh patch, but it appears to me that keeping the
> above patch in my tree would not help anybody.

The (or at least "a") root cause has actually been discovered. Would a patch 
that adds an xfail test case for it be acceptable?

As to the why the proposed patch causes test failures: I think this is due to 
the fact that remote-hg inserts a fake "master" branch (resp. "bookmark" in the 
hg terminology). Now, in those test cases, a hg repository gets created that 
actually contains a "null" bookmark named "master". When the proposed fix for 
the problem is added, this bookmarks gets ignored. But at that point, remote-hg 
already determined that there is a hg bookmark named "master", and adjusts how 
it works accordingly -- when we then remove that bookmarks, things go awry.

But I might be wrong here, and in any case, did not yet have time to come up 
with a proper fix. What I do have is a test case, that I could turn into an 
xfail test. As a matter of fact, I a know a few more bugs in remote-hg for 
which I could produce xfail test cases. Of course I'd prefer to put them in 
together with a fix, but I don't know when I can get to that, if ever. So, 
would such changes be welcome?

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

Reply via email to