Re: Troubleshooting openSUSE's packaging process

[Alexander Leithner]

Am 04.04.25 um 15:44 schrieb Alexandre Julliard:
> Alexander Leithner  writes:
> > Shouldn't GCC provide support for -mabi=ms out of the box?
>
> Yes, there must be something else going on. Is the config.log file
> available?

OK, I've managed to extract config.log from a local OBS build. The 
section relevant to -mabi=ms reads:

configure:20674: checking for working -mabi=ms
configure:20693: gcc -m64 -c -O2 -Wall -U_FORTIFY_SOURCE 
-D_FORTIFY_SOURCE=3 -fstack-protector-strong -funwind-tables 
-fasynchronous-unwind-tables -fstack-clash-protection 
-Werror=return-type  -mabi=ms  conftest.c >&5
conftest.c: In function 'a':
conftest.c:165:79: error: control reaches end of non-void function 
[-Werror=return-type]
   165 | int a(int b, ...) { __builtin_ms_va_list list; 
__builtin_ms_va_start(list,b); }
       | 
               ^
cc1: some warnings being treated as errors
configure:20693: $? = 1

Indeed, the build script on OBS exports a bunch of additional flags 
before configuring Wine as seen above, including setting 
-Werror=return-type.

Running configure with CFLAGS='-Werror=return-type ' also is enough to 
reproduce OBS' config.log output locally on my machine. Is this a bug in 
-mabi=ms detection or should OBS not set -Werror=return-type?

As a side note, OBS also enables Fortify. Isn't this subsequently 
disabled by configure anyway? If so, I imagine OBS could just not enable 
it in the first place, right?

Thanks for your help!

Best,
Alexander Leithner

Troubleshooting openSUSE's packaging process

[Alexandre Julliard]

Alexander Leithner  writes:
> Am 04.04.25 um 15:44 schrieb Alexandre Julliard:
> > Alexander Leithner  writes:
> > > Shouldn't GCC provide support for -mabi=ms out of the box?
> >
> > Yes, there must be something else going on. Is the config.log file
> > available?
>
> OK, I've managed to extract config.log from a local OBS build. The 
> section relevant to -mabi=ms reads:
>
> configure:20674: checking for working -mabi=ms
> configure:20693: gcc -m64 -c -O2 -Wall -U_FORTIFY_SOURCE 
> -D_FORTIFY_SOURCE=3 -fstack-protector-strong -funwind-tables 
> -fasynchronous-unwind-tables -fstack-clash-protection 
> -Werror=return-type  -mabi=ms  conftest.c >&5
> conftest.c: In function 'a':
> conftest.c:165:79: error: control reaches end of non-void function 
> [-Werror=return-type]
>    165 | int a(int b, ...) { __builtin_ms_va_list list; 
> __builtin_ms_va_start(list,b); }
>        | 
>                ^
> cc1: some warnings being treated as errors
> configure:20693: $? = 1
>
> Indeed, the build script on OBS exports a bunch of additional flags 
> before configuring Wine as seen above, including setting 
> -Werror=return-type.
>
> Running configure with CFLAGS='-Werror=return-type ' also is enough to 
> reproduce OBS' config.log output locally on my machine. Is this a bug in 
> -mabi=ms detection or should OBS not set -Werror=return-type?

We can work around this specific failure, but yes, setting -Werror
options is a bad idea, particularly during configure.
> As a side note, OBS also enables Fortify. Isn't this subsequently 
> disabled by configure anyway? If so, I imagine OBS could just not enable 
> it in the first place, right?

Yes, it's useless.

Troubleshooting openSUSE's packaging process

[Alexander Leithner]

Am 05.04.25 um 14:52 schrieb Alexandre Julliard:
> Alexander Leithner  writes:
> > Indeed, the build script on OBS exports a bunch of additional flags
> > before configuring Wine as seen above, including setting
> > -Werror=return-type.
> >
> > Running configure with CFLAGS='-Werror=return-type ' also is enough to
> > reproduce OBS' config.log output locally on my machine. Is this a bug in
> > -mabi=ms detection or should OBS not set -Werror=return-type?
>
> We can work around this specific failure, but yes, setting -Werror
> options is a bad idea, particularly during configure.
> > As a side note, OBS also enables Fortify. Isn't this subsequently
> > disabled by configure anyway? If so, I imagine OBS could just not enable
> > it in the first place, right?
>
> Yes, it's useless.

Thanks for all the help; I've submitted a change request to openSUSE's 
packaging on OBS. For the time being, exporting -Wno-error=return-type 
in CFLAGS and CXXFLAGS before configure is called does the trick.

Let's see if it'll be accepted.

This could be removed if/when the configure test is modified to work in 
case -Werror=return-type is set.

Best,
Alexander Leithner