Remove usage of __has_include

[msprotz]

@franziskuskiefer reported that our usage of __has_include was yet one more oddity that made support across compiler versions difficult.

I believe this used to be a an issue but with a recent change from Natalia, there is now a preprocessor guard that doesn't trigger __has_include unless we know we have it.

But simplifying this logic is good, and so I would like to remove our usage of __has_include. (There is one left for debugging purposes in libintvector.h that I believe is innocuous.)

There would need to be further changes, for instance:

• fixing the local Makefiles to copy a dummy development config.h to enable all the features
• making sure no dist relies on the absence of config.h (I know Mozilla has one now, but maybe election guard needs one...?)

Therefore, I'm leaving this as a draft PR for further discussion.

[msprotz]

Unsurprisingly, this breaks windows where indeed, our CI does not generate a config.h (it should!).

[franziskuskiefer]

I believe this used to be a an issue but with a recent change from Natalia, there is now a preprocessor guard that doesn't trigger __has_include unless we know we have it.

__has_include doesn't exist in older GCCs. So any GCC 4.8 build for example will not pick up the config.h.

Unsurprisingly, this breaks windows where indeed, our CI does not generate a config.h (it should!).

I'd suggest removing the CI. There's no point in trying to keep this broken build system around.

[msprotz]

I think there is still some benefit to this, namely that we check that the C code generated by krml is syntactically correct for MSVC, so I'd like to keep this if we can. It's better than nothing, until we have more comprehensive Windows CI.

[franziskuskiefer]

Thanks!
It appears to work fine on Windows as well: https://github.com/cryspen/hacl-packages/actions/runs/2581290156

[msprotz]

Thanks, I needed a couple more tweaks, hopefully this time is the right time.

[msprotz]

So now make will error out if you don't have config.h and Makefile.config. You can still write these by hand (Mozilla does), but of course the suggestion is to run ./configure before invoking make.

The CI scripts have been updated to reflect this.

[franziskuskiefer]

lgtm, but some changes in gcc-compatible look unrelated?

[msprotz]

Yeah that's because the nightly job for refreshing dist/ isn't running yet. So this PR contains a refresh of dist/ as well.

Currently blocked on a review, then I'll merge this.

[Frosne]

If you don't mind, I will first check that the modification does not break anything in our code :)
I should be back with a definitive answer on Monday. In any case, I support any code simplifications.

[msprotz]

Thanks, if this looks good for you on Monday, please go ahead and merge (I'll be off on vacation).

[Frosne]

as promised, I am back.

It seems that everything works, but I would like to take more time to look at it.
I will be away as well, until 13th of July. As soon as I am back, it will be the first thing I will address :)

[msprotz]

@Frosne can you confirm this looks good for you? I think we've pretty much committed to always having config.h now (even if it's hand-written), so __has_include can go I think.

[Frosne]

IIUC, we (dist/mozilla) also should have a config.h file? Could we have it empty?

[Frosne]

I would prefer not to have any predefined values there, because we do it in our build system.

[msprotz]

Ok. Until now there was a config.h under version control for you guys, in dist/mozilla.

Are you saying you're ok to remove dist/mozilla/config.h?

[msprotz]

Sorry, here: https://github.com/hacl-star/hacl-star/blob/main/dist/config.mozilla.h

[Frosne]

We do not use it, so yes, I think it is ok :)