[csw-buildfarm] Fwd: still python problems?

[rupert THURNER]

On Sat, Apr 3, 2010 at 00:04, Mads Kiilerich <mads at kiilerich.com> wrote:
> Ben Walton wrote, On 04/02/2010 10:41 PM:
>>
>> Excerpts from Mads Kiilerich's message of Fri Apr 02 08:02:47 -0400 2010:
>>
>>
>>>
>>> In this case the problem is (IIRC) that "which" is found in /usr/bin as
>>> a csh script which doesn't set the exit code to indicate "not found".
>>> Using another shell would not help, but putting a sane "which" in $PATH
>>> would.
>>>
>>
>> Wow!  Hadn't bumped _that_ nasty before.  Afaik, the only viable
>> workaround there is, as you say, to get something decent with higher
>> precedence in the PATH.  It shouldn't be hard to whip something up,
>> local to this package build description to solve the issue.
>>
>>
>>>
>>> The tests use "#!/bin/sh", so there is no simple workaround to use
>>> another sh.
>>>
>>
>> Ok.  And looking at the way things are setup, I think it's easier for
>> the OpenCSW recipe to work around this than to alter the test suite.
>>
>> Rupert: I tested the following diff on your mercurial GAR recipe and
>> the test failure count goes down to 1 (the above noted `which`
>> deficiency).  It should be 'future proof' against new tests, etc as
>> opposed to a patch that would need to be updated constantly.
>>
>
> Yes, that could make it work.
>
> But it also raises the question: Why run the test suite at all? What is it
> you want to test? Mercurial on Solaris or Mercurial on a custom GNU user
> space? The test suite primarily tests that the logic in the code is correct,
> and these tests are run automatically as regression tests and on several
> (mostly linux and mac) platforms by developers. The test suite is not
> specially targeted at testing cross platform issues. So far it have caused
> numerous corrections to the test suite just to make it runnable on solaris,
> but it haven't found any real issues for running your Mercurial on Solaris
> (though it raised attention to a known but unsolved python issue).

we want to be as sure as possible that "the whole package" (i.e. the
solaris operating system, the csw packages necessary, and mercuriual)
does what it is expected to do.

we are really glad about the high quality mercurial has and always
had. also the little amount of work which is necessary to provide a
rock stable package is impressing (compared to subversion, and many
other tools out of the same domain). and we are really glad about how
many people out of the mercurial community help out in the rare cases
something does not work as expected.

our (trygve, and the others who worked on mercurial packaging
afterwards) policy was that we do not run the test suite automatically
but comment it in the opencsw package makefile. but we try to run it
and report things that do not work as low priority "bugs" (maybe it
does not even deserve the word bug).

its a little bit unfortunate that the mercurial test suite has a "sh"
dogma, and the only platform really testing this is solaris resp
opencsw. therefor ben's solution of replacing it with bash is
appealing. i guess we do not want to test the test suite's compliance
to its expectations :)

what do you think?

rupert.