The Bizarre Case of Zstd's Slow Performance on Arch Linux

zstd normally uses timespec_get() for measuring wall-clock time. But that's available only since C11. For older standards, it uses clock(); but that returns CPU time, which is wrong.

https://github.com/facebook/zstd/pull/3165

So -std=c99 inserted by cmake was the culprit, whilst the default in current GCC is gnu17. So please people, don't use silly build systems. autoconf doesn't insert this sillyness, when the compiler does better.

https://gcc.gnu.org/onlinedocs/gcc/Standards.html

> The default, if no C language dialect options are given, is -std=gnu17.

I’m sure autoconf will insert this if you explicitly tell it to [1].

1: https://github.com/facebook/zstd/blob/b7b7edb3a3017ac8e16d7e...

I have AC_PROG_CC_C99 in all my configure.ac's, and autoconf probes if the compiler needs that flag. Most compilers don't need it. So, no.

It is not possible to make autoconf set the compiler mode to exactly C99?

Following the GitHub issue, it seems like this is actually just a bug in the timing code, and that in actual use it's not any slower.

For TL;DR, "having that C99 standard specified is what ultimately was found to cause this large performance difference"

mmm