wine FTBFS with mingw64 gcc 11: undefined reference to `sincos'

[Alex Xu (Hello71)]

Hi,

I tried building wine 6.8 with mingw. All fine, just one problem, it 
reports error "undefined reference to `sincos'". Why, you may ask, since 
wine never calls sincos function? Well, it seems to be because mingw gcc 
11 enables sincos optimization: [0]. But, wine uses -nodefaultlibs, 
intending to provide libc functions with msvcrt; however, wine msvcrt 
doesn't implement any sincos. So, there is a mismatch: gcc accepts mingw 
claim to implement sincos (ok since mingwex implements it), then 
generates call to sincos, then fails at link time (because msvcrt does 
not implement it).

After adding -fno-builtin-{sin,cos}{,f} to CFLAGS, wine compiled and ran 
OK.

This issue seems to be worked around for winelib builds by adding 
-fno-builtin, so that gcc will not generate sincos (or most other 
problematic functions).

I think this issue could be worked around by adding those -fno-builtin-* 
flags for mingw builds. However, I think the sin/cos->sincos 
optimization is a good one, so maybe we could instead add sincos 
forwarder in msvcrt (calling sin+cos separately in case sincos is not 
implemented in libc?).

CCed [e-mail] due to potential relevance 
and [e-mail] due to maintaining wine and mingw-gcc on 
Arch. You may need to subscribe in order to reply all. I look forward to 
hearing your thoughts.

Regards,
Alex.

[0] https://github.com/gcc-mirror/gcc/commit/4f48f31bbfc10697296ff004a92614d9249...

[Jacek Caban]

Hi Alex,

On 5/14/21 3:01 AM, Alex Xu (Hello71) wrote:
> Hi,
>
> I tried building wine 6.8 with mingw. All fine, just one problem, it
> reports error "undefined reference to `sincos'". Why, you may ask, since
> wine never calls sincos function? Well, it seems to be because mingw gcc
> 11 enables sincos optimization: [0]. But, wine uses -nodefaultlibs,
> intending to provide libc functions with msvcrt; however, wine msvcrt
> doesn't implement any sincos. So, there is a mismatch: gcc accepts mingw
> claim to implement sincos (ok since mingwex implements it), then
> generates call to sincos, then fails at link time (because msvcrt does
> not implement it).
>
> After adding -fno-builtin-{sin,cos}{,f} to CFLAGS, wine compiled and ran
> OK.
>
> This issue seems to be worked around for winelib builds by adding
> -fno-builtin, so that gcc will not generate sincos (or most other
> problematic functions).
>
> I think this issue could be worked around by adding those -fno-builtin-*
> flags for mingw builds. However, I think the sin/cos->sincos
> optimization is a good one, so maybe we could instead add sincos
> forwarder in msvcrt (calling sin+cos separately in case sincos is not
> implemented in libc?).
>
> CCed [e-mail] due to potential relevance
> and [e-mail] due to maintaining wine and mingw-gcc on
> Arch. You may need to subscribe in order to reply all. I look forward to
> hearing your thoughts.

I think that the decision was unfortunate on GCC side, but there is 
little we can do. We will probably need to provide it in msvcrt 
importlibs. Please try the attached patch, it should help.

Thanks,

Jacek

[Dean Greer]

Hi Jacek

I’ve tested compiling wine-6.8 with mingw-gcc-11.1.0 (stock) with the
provided patch applied and the compile finished without issue.

Thanks,

Dean aka Gcenx

On Fri, May 14, 2021 at 1:28 PM Jacek Caban  wrote:
> Hi Alex,
>
> On 5/14/21 3:01 AM, Alex Xu (Hello71) wrote:
> > Hi,
> >
> > I tried building wine 6.8 with mingw. All fine, just one problem, it
> > reports error "undefined reference to `sincos'". Why, you may ask, since
> > wine never calls sincos function? Well, it seems to be because mingw gcc
> > 11 enables sincos optimization: [0]. But, wine uses -nodefaultlibs,
> > intending to provide libc functions with msvcrt; however, wine msvcrt
> > doesn't implement any sincos. So, there is a mismatch: gcc accepts mingw
> > claim to implement sincos (ok since mingwex implements it), then
> > generates call to sincos, then fails at link time (because msvcrt does
> > not implement it).
> >
> > After adding -fno-builtin-{sin,cos}{,f} to CFLAGS, wine compiled and ran
> > OK.
> >
> > This issue seems to be worked around for winelib builds by adding
> > -fno-builtin, so that gcc will not generate sincos (or most other
> > problematic functions).
> >
> > I think this issue could be worked around by adding those -fno-builtin-*
> > flags for mingw builds. However, I think the sin/cos->sincos
> > optimization is a good one, so maybe we could instead add sincos
> > forwarder in msvcrt (calling sin+cos separately in case sincos is not
> > implemented in libc?).
> >
> > CCed [e-mail] due to potential relevance
> > and [e-mail] due to maintaining wine and mingw-gcc on
> > Arch. You may need to subscribe in order to reply all. I look forward to
> > hearing your thoughts.
>
> I think that the decision was unfortunate on GCC side, but there is
> little we can do. We will probably need to provide it in msvcrt
> importlibs. Please try the attached patch, it should help.
>
> Thanks,
>
> Jacek

[Mingw-w64-public] wine FTBFS with mingw64 gcc 11: undefined reference to `sincos'

[lhmouse]

在 5/15/21 1:27 AM, Jacek Caban 写道:
> I think that the decision was unfortunate on GCC side, but there is little we can do. We will 
> probably need to provide it in msvcrt importlibs. Please try the attached patch, it should help.

Doesn't GCC transform such pair of calls to `sincos()` again and result in an infinite recursion?

[Mingw-w64-public] wine FTBFS with mingw64 gcc 11: undefined reference to `sincos'

[Jacek Caban]

On 15/05/2021 04:24, lhmouse wrote:
> 在 5/15/21 1:27 AM, Jacek Caban 写道:
> > I think that the decision was unfortunate on GCC side, but there is little we can do. We will
> > probably need to provide it in msvcrt importlibs. Please try the attached patch, it should help.
>
> Doesn't GCC transform such pair of calls to `sincos()` again and result in an infinite recursion?

The patched file is built with -fno-builtin, so it should be fine.

Jacek