Fix fenv_t initializer

[000lbh]

I have no idea why previous versions compile on GitHub actions, but they don't on my loongarch64 machine, and there was also someone reported on aarch64 architecture #278 . Initializer of a struct should be a brace-enclosed list. Btw, my gcc version is 14.2.1, it seems that cross-compile gcc version in GitHub actions is old.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 39.03%. Comparing base (c1e8462) to head (6404565). Report is 9 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #308   +/-   ##
=======================================
  Coverage   39.02%   39.03%           
=======================================
  Files         233      233           
  Lines        6144     6151    +7     
  Branches     1608     1608           
=======================================
+ Hits         2398     2401    +3     
- Misses       3393     3397    +4     
  Partials      353      353           

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[yuyichao]

It seems that the definition of fenv_t in this repo is not a struct except on x86 (both 32 and 64bit). However, the system one does seem to be a struct. Which one is intended to be used?

(The include guard is such that if one version of fenv is included, the other one won't, at least on Linux)

[000lbh]

Thanks for you comment. I didn't notice there are some openlibm_fenv_ARCH.h in the repo before. It is strange that fenv.c of some architectures like arm, amd64 and aarch64 includes openlibm_fenv.h, which uses project's fenv.h, but others like loongarch64 and riscv64 directly includes system's fenv.h

Maybe I should update my PR, to change these two architectures to use openlibm_fenv.h or just revert my changes to other architectures? or should I fix this inconsistence issue?

[000lbh]

Maybe this version is ok