test-mq-strip.t fails on Windows MSYS - bookmarks not updated

[Adrian Buehlmann]

On 2012-05-22 12:39, Mads wrote:
> On 22/05/12 10:38, Adrian Buehlmann wrote:
>> using d0b9ebba41e9 from default
>>
>> --- c:\users\adi\hgrepos\hg-main\tests\test-mq-strip.t
>> +++ c:\users\adi\hgrepos\hg-main\tests\test-mq-strip.t.err
>> @@ -458,6 +458,7 @@
>>     $ hg bookmarks
>>        B                         9:ff43616e5d0f
>>        delete                    6:2702dd0c91e7
>> +   * todelete                  8:d62d843c9a01
>>     $ hg strip -B delete
>>     saved backup bundle to $TESTTMP/bookmarks/.hg/strip-backup/*-backup.hg (glob)
>>     bookmark 'delete' deleted
>>
>>
>> This test failure first appears with:
>>
>>    changeset:   16718:3290e24bb3f0
>>    parent:      16716:0311a6abd38a
>>    user:        David Soria Parra<dsp at php.net>
>>    date:        Sun May 13 16:39:40 2012 +0200
>>    summary:     strip: introduce -B option to remove a bookmark
> 
> The parameter in
>      repo._writebookmarks(marks)
> is unused. That seems suspicious. Explicitly setting
>      repo._bookmarks = marks
> in mq.py before writing bookmarks do that .hg/bookmarks is updated again.
> 
> I don't know why it works on unix. Perhaps some side effect of different 
> fstats and thus different cache reloading that do that the marks are 
> modified in place and thus rewritten?
> 
> (((
> btw: Why do we have ui.write "bookmark '%s' deleted" - that is what the 
> user explicitly requested, so I would expect that Mercurial did it silently.
> 
> It also seems "interesting" that a strip of bookmark referencing a 
> non-leaf revision remove the bookmark but doesn't strip anything and 
> aborts. I would expect it to either fail completely (preferred) or just 
> remove the bookmark and return with success (less preferred).
> )))

Yeah. And mq.py has in line 2958 (function strip):

        if not revs:
            del marks[mark]
            repo._writebookmarks(mark)                        <----
            ui.write(_("bookmark '%s' deleted\n") % mark)

which was probably meant to be

            repo._writebookmarks(marks)

but since that parameter is ignored anyway, changing that doesn't help.


Finally, if I do

diff --git a/mercurial/posix.py b/mercurial/posix.py
--- a/mercurial/posix.py
+++ b/mercurial/posix.py
@@ -455,6 +455,7 @@
         self.stat = os.stat(path)

     def cacheable(self):
+        return False
         return bool(self.stat.st_ino)

     __hash__ = object.__hash__


(in class cachestat of posix.py), then the test fails on Linux in the
same way as it fails on Windows.