greenlet 3.0.3 fails to build on ppc64el (and riscv64)

[Frank Heimes]

greenlet 3.0.3 fails to build on riscv64 and ppc64el:

complete build logs at
https://launchpad.net/ubuntu/+source/python-greenlet/3.0.3-0ubuntu2

riscv64-linux-gnu-gcc -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -g -O2 -fno-omit-frame-pointer -ffile-prefix-map=/<>=. -fstack-protector-strong -Wformat -Werror=format-security -fdebug-prefix-map=/<>=/usr/src/python-greenlet-3.0.3-0ubuntu2 -Wdate-time -D_FORTIFY_SOURCE=3 -fPIC -I/usr/include/python3.11 -c /<>/src/greenlet/greenlet.cpp -o build/temp.linux-riscv64-cpython-311/<>/src/greenlet/greenlet.o -Os

So the reason is that recent Linux distros decided to build with -fno-omit-frame-pointer by default. Apparently both the riscv64 and the ppc64el implementations cannot cope with that.

The issue can be reproduce by adding -fno-omit-frame-pointer to the build flags.

https://launchpad.net/ubuntu/+source/python-greenlet/3.0.3-0ubuntu4
_____

There is also this upstream github issue:
https://github.com/python-greenlet/greenlet/issues/395

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-09 11:22 EDT-------
According to https://github.com/python-greenlet/greenlet/issues/395 this is the source with asm affected:

https://github.com/python-greenlet/greenlet/blob/master/src/greenlet/platform/switch_ppc64_linux.h

It is likely that r31 is reserved when the frame pointer is required.

Adding one of the powerpc toolchain developers to confirm.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-09 12:50 EDT-------
(In reply to comment #4)
> According to https://github.com/python-greenlet/greenlet/issues/395 this is
> the source with asm affected:
>
> https://github.com/python-greenlet/greenlet/blob/master/src/greenlet/
> platform/switch_ppc64_linux.h
>
> It is likely that r31 is reserved when the frame pointer is required.
>
> Adding one of the powerpc toolchain developers to confirm.

I've asked in the upstream issue for the preprocessed source file that I can recreate and debug with.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-09 13:07 EDT-------
(In reply to comment #5)
> I've asked in the upstream issue for the preprocessed source file that I can
> recreate and debug with.

I was able to create a small reproducer myself. I'll have a look.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-09 22:30 EDT-------
I opened an external GCC bug to track this.

https://gcc.gnu.org/PR114664

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-10 10:18 EDT-------
The consensus in the GCC bugzilla is that the inline asm used in greenlet is itself buggy and this is not a compiler bug. A different solution to their inline asm will need to be used.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2024-04-10 10:36 EDT-------
May I ask why the distro build is forcing the use of -fno-omit-frame-pointer on ppc64le? It just leads to less efficient code on ppc64le.