Sorry, I forgot about ubsan32 case. I'll resolve this error later.

Note that as for all32 (-m32) builds, test-freestanding skips its actual test.

Could you tell me why freestanding (or just its tests?) is not compatible with 32-bit ?

-m32 generates i386 code but os_syscall1() uses rax and 64-bits registers (and its own stack frame).

Also, due to lack of my experience with freestanding Linux environment, I just avoid to #include almost everything. As a result, the following line causes incompatibility:

Because each platform has its own numbers for syscall

x64 (exit = 60):
https://github.com/torvalds/linux/blob/v4.17/arch/x86/entry/syscalls/syscall_64.tbl#L71

x86 (exit = 1):
https://github.com/torvalds/linux/blob/v4.17/arch/x86/entry/syscalls/syscall_32.tbl#L15

We may need to include <sys/syscall.h> for portability.

Thanks for the clear explanation @t-mat !

We may need to include <sys/syscall.h> for portability.

Isn't the idea of including anything going against the very concept of freestanding ?

edit :
OK, sorry, now I understand that these specificities happen in the test program tests/freestanding.c,
not in the lz4 freestanding library.

In which case, of course, the test itself can be specific to certain platforms.

Maybe this could be called out explicitly in the unit name ?

these specificities happen in the test program tests/freestanding.c, not in the lz4 freestanding library.

Right. If we can successfully run lz4 freestanding library without standard library (libc, etc) but only depends platform's own syscall, we can say it conforms the concept of the freestanding (bare bone C environment).

Maybe this could be called out explicitly in the unit name ?

Yes. It'd be nice to have explicit / specific name (and maybe test) for them.

Hi @t-mat ,

it seems I completely lost track of this PR.

If I understand correctly, all it does is enable freestanding test automatically as part of make all and make test,
as well as ensuring it's setup properly for 32-bit x86 target.

Is it still valid as is ?
Could it rebased ?
(I see there was a ubsan test failure, presumably due to some external flaky condition)

Yeah, I also suspect this PR is outdated. I've tried to implement wider support of freestanding test. But it seems it's been failed.

Please close this PR for now.
When I write better test for other platforms/architectures, I'll send another PR.