https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118133
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
Last reconfirmed| |2024-12-19
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to ktkachov from comment #1)
> I agree with the need to find a new scheme for these .def files.
> This reminds me a bit of the days in the arm backend when we used ML scripts
> to generate the arm_neon.h header...
>
> I'm a bit hesitant about M4. I know there are many .m4 files in the GCC
> source so it's somewhat established, but I also feel those files are rarely
> touched in normal development.
>
> But for frequently-extended intrinsics definitions it feels like Python is a
> more accessible syntax to most developers that work on the AArch64 backend
> these days.
Python has an extra problem on top of m4, it is not currently required at all
even for generated files. Yes python is required for riscv target with
with-arch but not for aarch64 building currently.
Plus I feel python is way too heavy weight for this and is less accessible
syntax for myself.
May I suggest awk? It is currently required and used for .opt files and even
used to process the passes.def file. Its syntax is easier to handle than python
in my view too.
