Re: [Patch] MinGW support for Glib-Object-Introspection

["Martin Schlemmer" <Martin Schlemmer nwu ac za>]

> > > On 2012/06/23 at 05:57 PM, Torsten Schoenfeld <kaffeetisch gmx de> wrote:
> On 21.06.2012 13:50, Martin Schlemmer wrote:
> > - I am not sure how to add 'build' to PATH for a dmake Makefile (will
> > do some more research when I get time)
>
> Does cmd.exe not support the syntax that is currently used for 
> LD_LIBRARY_PATH?  That is, simply prepend "PATH=build;%PATH%" to the 
> command?

It sets the PATH (to "build;%PATH% command" amusingly), but do not execute
the command. I'll look into dmake documentation again or jerry rig it with env.exe
or sh.exe for now (MSYS).

> > - Other scattered FIXME's etc. that I am not sure of (mostly due to
> > more recent versions and G-O-I being mostly unstable on win32)
>
> Some comments:
>
> • The missing libffi stuff is actually a genuine bug. 
> gobject-introspection-1.0.pc lists libffi only as a private dependency 
> (as of a few months ago), but we make use of libffi directly in the 
> bindings.  I'm puzzled that this does not break Unix builds too.  Please 
> adjust the libffi changes so that they are done on every OS.

I think on Linux LD is more forgiving if the library is a dependency of
a linked one (libgirepository/libgobject) than on Win32, but sadly I went
from full time on Linux to only now and then these days, so I can only
speculate as Introspection.so still link without specifying libffi explicitly.

> • I recently changed the code in compile_test_libraries a little (to get 
> rid of the issue mentioned in the FIXME comment).  Can you adapt your 
> patch so that it applies to what is in git master?

Will do.

> • Your rewritten compile_test_libraries will throw an exception when the 
> compilation fails, aborting the Makefile.PL run.  The intention is that 
> these failures are non-fatal.  Please change the code so that it simply 
> returns a false value on failure.

The die's are in the eval block so it should not be an issue? I did that only
to print $@ on failure during debugging, but I'll change it to return only.

> • In t/inc/setup.pl you have disabled the skip_all branch for MSWin32. 
> But the tests still do need to be skipped if PATH is not set up 
> correctly, right?

Yes, sorry, was while I was trying to get it to work - should not have crept
into the patch.


Regards,
Martin



Vrywaringsklousule / Disclaimer:  http://www.nwu.ac.za/it/gov-man/disclaimer.html