Libm and quad tester rely on Unix specific sys/wait.h for pipes. This limits testing capacity on Windows, e.g. with msys2/mingw, and other non-unix based OS-es.
Fixing this would allow to fully test SLEEF on Windows (x86 and aarch64), and not just DFT (see #540).
Opening this issue to discuss potential alternatives:
@shibatch suggests to base a new tester1 engine on tlfloat (https://github.com/shibatch/tlfloat). This would remove the need for pipe-ing (wait.h) while still enabling multi-threading. It would also remove the dependency in mpfr. A first problem here is probably linking SLEEF and tlfloat, as it might not be entirely straightforward.
use a port of wait.h for Windows, e.g. , not sure if Windows on Arm is supported.
Brutally, remove the calls to wait(pid). Is it even possible?
Libm and quad tester rely on Unix specific sys/wait.h for pipes. This limits testing capacity on Windows, e.g. with msys2/mingw, and other non-unix based OS-es. Fixing this would allow to fully test SLEEF on Windows (x86 and aarch64), and not just DFT (see #540).
Opening this issue to discuss potential alternatives: