mercurial at 12996: 5 new changesets

[Eric Eisner]

Hmm, it appears my method of changing PATH did not propagate correctly to
the unified tests, so everything I said is a lie. I'll look into make these
things functionally pass for older versions.

As for the --date flag, should we always ignore it? should we query the git
version and change behavior based on that?

-Eric

On Fri, Dec 3, 2010 at 09:23, Eric Eisner <ede at mit.edu> wrote:

> On Fri, Dec 3, 2010 at 06:45, Thomas Arendsen Hein <thomas at intevation.de>wrote:
>
>> mercurial/subrepo.py uses "cmd += ['--date', ...", but --date was
>> added in Git 1.7.0, which is newer than the current version in
>> Debian stable (1.5.6.5).
>>
>
> I have tested with git 1.6 (and now again with 1.5.6.5), and my git
> correctly logs the date when passed. The date passed by mercurial and the
> default date in the tests are different, and the commit hashes would change
> if it didn't recognize the --date option.
>
> My results are from building git at specified versions, could Debian be
> doing something funny?
>
> This breaks test-subrepo-git.t (and of course the actual
>> functionality, but since this is new, I do not yet miss it :))
>>
>
> Unfortunately I have no idea how to make the tests pass with git older than
> 1.7. The tests pass with the newest version of git, but git's output changes
> too much before 1.7. Almost all commands are told to be quiet, but older
> versions have different ideas of what quiet means, and still some double
> quotes were changed to single quotes, etc.
>
> The tests functionally work with 1.6 (and 1.5.6.5), and major versions
> before 1.6 expect commands to be 'git-commit' instead of 'git commit' and
> should fail verbosely (although not descriptively).
>
> -Eric
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://selenic.com/pipermail/mercurial-devel/attachments/20101203/7218efdb/attachment.htm>