Re: [PATCH 1/1] do not add common-main to lib

From: Christian Hesse <list@eworm.de>
Date: Aug 15, 2016, 12:21 UTC
Message-ID: <20160815142125.0ca30e0f@leda.localdomain>
In-Reply-To: <20160815120223.4lr23aiqmqzjprch@sigill.intra.peff.net>

Jeff King <peff@peff.net> on Mon, 2016/08/15 08:02:

Show 12 quoted lines

> On Mon, Aug 15, 2016 at 09:52:07AM +0200, Christian Hesse wrote:
> 
> > From: Christian Hesse <mail@eworm.de>
> > 
> > Commit 08aade70 (mingw: declare main()'s argv as const) changed
> > declaration of main function. This breaks linking external projects
> > (e.g. cgit) to libgit.a with:
> > 
> > error: Multiple definition of `main'  
> 
> I'd expect the culprit is actually 3f2e229 (add an extra level of
> indirection to main(), 2016-07-01).

Ah, probably you are right...

Show 7 quoted lines

> > So do not add common-main to lib and let projects have their own
> > main function.  
> 
> That is certainly an option, but I think it means that those projects
> are potentially buggy in the same way that some git commands were prior
> to the common-main series. Namely, the common main() may do some
> run-time setup that parts of libgit.a assume has been done.

Ok, got it.

> I would not be surprised if cgit crashes on Windows, for instance, for
> the reasons detailed in 650c449 (common-main: call
> git_extract_argv0_path(), 2016-07-01). I would also not be surprised if
> nobody actually builds cgit on Windows. :)

I never tried and probably nobody else did. :-p

> The "right" way to do it (according to the way libgit.a views the world)
> is for cgit's main to become cmd_main(), and let libgit.a do its
> run-time startup before getting there.

Looks like that does the job. I will give it some more testing.

Please ignore my patch... ;) Thanks a lot!

-- 
main(a){char*c=/*    Schoene Gruesse                         */"B?IJj;MEH"
"CX:;",b;for(a/*    Best regards             my address:    */=0;b=c[a++];)
putchar(b-1/(/*    Chris            cc -ox -xc - && ./x    */b/42*2-3)*42);}