Re: gcc 14+ and m68k build failure

On Sun, Jun 2, 2024, at 8:58 AM, Joel Sherrill wrote:
> Hi
>
> GCC 14+ fail to build a few newlib targets. I am looking into m68k-rtems
> (m68k-elf) first. If there is a common thread, it appears to be related to
> targets which do not have long double in all multilibs.

The same thing is happening on SH for the m4-single-only multilib.

> First, how do you see more than "CC..file.c"?

Add `V=1' to you make invocation.
 
> The error for m68k starts with this:
>
> ake[3]: Entering directory
> '/home/joel/test-gcc/b-m68k-rtems6-gcc/m68k-rtems6/newlib'
>   CC       libm/complex/libm_a-ccoshl.o
> ../../../gcc/newlib/libm/complex/ccoshl.c: In function 'ccoshl':
> ../../../gcc/newlib/libm/complex/ccoshl.c:43:13: error: implicit
> declaration of function 'coshl'; did you mean 'coshf'?
> [-Wimplicit-function-declaration]
>    43 |         w = coshl(x) * cosl(y) + (sinhl(x) * sinl(y)) * I;
>       |             ^~~~~
>       |             coshf
> ../../../gcc/newlib/libm/complex/ccoshl.c:43:24: error: implicit
> declaration of function 'cosl'; did you mean 'cosf'?
> [-Wimplicit-function-declaration]
>    43 |         w = coshl(x) * cosl(y) + (sinhl(x) * sinl(y)) * I;
>
>
> I suspect that since only one multilib for m68k has multilib and that is
> picked at compile time, the multilib here isn't enabling the prototypes for
> long double. That idea seems right but the code doesn't seem to always
> follow up to avoid trying to compile files like the above which assume long
> double.

I think I understand what's going on. `_LDBL_EQ_DBL' is not being defined for that multilib. So, functions like `coshl' aren't compiled (see newlib/libm/common/coshl.c), but the complex versions of the functions don't have the `#ifdef _LDBL_EQ_DBL' guard around them (see: newlib/libm/complex/ccoshl.c). And GCC 14 changed the lack of a function declaration from a warning into an error.

> Ideas appreciated.

I'll try to post a patch adding the ifdefs to add long double complex functions. 

> --joel

pietro