Missing pieces for 1.6.0 on MinGW?

Missing pieces for 1.6.0 on MinGW?

[Junio C Hamano]

Just a quick question before the weekend ends and -rc2 gets tagged.  (I
lost track of that argv0 vs bin/git vs libexec/git-core/git-foo
discussion).

Are there any missing but necessary patches we need before 1.6.0 for
MinGW?

Re: Missing pieces for 1.6.0 on MinGW?

[Steffen Prohaska]

On Aug 2, 2008, at 9:05 PM, Junio C Hamano wrote:

> Just a quick question before the weekend ends and -rc2 gets tagged.   
> (I
> lost track of that argv0 vs bin/git vs libexec/git-core/git-foo
> discussion).

I haven't found time to work on this and I also haven't observed
any progress made by others.


> Are there any missing but necessary patches we need before 1.6.0 for
> MinGW?

So we don't have patches, although we still have the problem
discussed last weekend.

Unfortunately, I cannot promise that I'll find time the next couple
of days to work on the issue.  I propose you just continue the
release cycle without waiting for MinGW.

	Steffen

Re: Missing pieces for 1.6.0 on MinGW?

[Johannes Sixt]

Zitat von Junio C Hamano <gitster@pobox.com>:
> Just a quick question before the weekend ends and -rc2 gets tagged.  (I
> lost track of that argv0 vs bin/git vs libexec/git-core/git-foo
> discussion).
>
> Are there any missing but necessary patches we need before 1.6.0 for
> MinGW?

Yes, there are some open issues:

(1) git-gui was fixed for the msysgit installer, but it broke for me; but I
think I know where to fix it.

(2) the non-builtins in $(bindir) don't set argv0_path, and consequently don't
find ETC_GITCONFIG.

(3) the 'mingw_main undefined' error is still a mystery. I'm about to send a
preprocessed file to Steffen (it's a bit large, even compressed, so I'll do
that in a private mail).

Item (1) I expect to work on later today, but needs some investigation by
Steffen and perhaps Shawn. (2) is probably a minor issue. (3) is, well, a
mystery, although Steffen's patch works for me, too. Nevertheless, I'd like to
know why my original approach does not always work.

-- Hannes

Re: Missing pieces for 1.6.0 on MinGW?

[Björn Steinbrink]

On 2008.08.03 11:16:25 +0200, Johannes Sixt wrote:
> (3) the 'mingw_main undefined' error is still a mystery. I'm about to send a
> preprocessed file to Steffen (it's a bit large, even compressed, so I'll do
> that in a private mail).

FWIW, gcc 4.3 complains about such function declarations:
x.c:4: error: invalid storage class for function ‘foo’

Seems that gcc complains about that since 4.0, and linking failed with
older versions under some circumstances.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12738
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17205

Björn

Re: Missing pieces for 1.6.0 on MinGW?

[Steffen Prohaska]

On Aug 3, 2008, at 11:16 AM, Johannes Sixt wrote:

> Zitat von Junio C Hamano <gitster@pobox.com>:
>> Just a quick question before the weekend ends and -rc2 gets  
>> tagged.  (I
>> lost track of that argv0 vs bin/git vs libexec/git-core/git-foo
>> discussion).
>>
>> Are there any missing but necessary patches we need before 1.6.0 for
>> MinGW?
>
> Yes, there are some open issues:
>
> (1) git-gui was fixed for the msysgit installer, but it broke for  
> me; but I
> think I know where to fix it.

I now recognized that not all hunks that I have in 4msysgit were applied
by Shawn.  See separate mail

  http://article.gmane.org/gmane.comp.version-control.git/91221


> (2) the non-builtins in $(bindir) don't set argv0_path, and  
> consequently don't
> find ETC_GITCONFIG.

Setting argv0_path correctly wouldn't help alone.  The relative
path to $prefix is different in $prefix/bin and
$prefix/libexec/git-core.  Thus, we cannot do the same computation.
system_path() would need to check from which directory the computation
starts.


> (3) the 'mingw_main undefined' error is still a mystery. I'm about  
> to send a
> preprocessed file to Steffen (it's a bit large, even compressed, so  
> I'll do
> that in a private mail).

ok.

	Steffen

Re: Missing pieces for 1.6.0 on MinGW?

[Johannes Schindelin]

Hi,

On Sun, 3 Aug 2008, Johannes Sixt wrote:

> (2) the non-builtins in $(bindir) don't set argv0_path, and consequently 
> don't find ETC_GITCONFIG.
> 
> (3) the 'mingw_main undefined' error is still a mystery. I'm about to 
> send a preprocessed file to Steffen (it's a bit large, even compressed, 
> so I'll do that in a private mail).

For both issues, IMHO Steve's patches should be superior, as they (with my 
proposed lookup_program_in_path()) would not only make relative 
ETC_GITCONFIG work on MinGW, but everywhere else, too.  Obsoleting 
mingw_main, of course.

Ciao,
Dscho