GMP and clang bugginess

[Niels Möller]

tg at gmplib.org (Torbjörn Granlund) writes:

> The clang on FreeBSD 10 miscompiles GMP on for some x86 CPU subtypes.
> Apparently Intel Haswell is one of these; this is currently not
> exercised by our tests setup.

But the bugs are likely to be exposed if a user runs make check on an
affected platform?

My impression is that clang has been particularly broken on non-x86
platforms and in earlier versions. Which indicates poor testing, or too
optimistic advertising, but not that clang is unlikely to mature for any
particular platform.

I think that requiring an --enable-clang to be able to build with any
version of clang on any platform whatsoever is a bit harsh. On the other
hand, a blacklisting of known broken compilers (by version and where
relevant also by platform), and a --disable-compiler-blacklist option to
override it, would be helpful to both users and us.

If a user has the latest and greatest clang on a very vanilla x86 system
where it actually works fine, the user is going to be pretty annoyed by
an --enable-clang option, and possibly make it a habit to always use
that option.

Ideally, the blacklist in configure should automatically turn red
entries to orange on the list of test results. And unexpected successes
with a blacklisted compiler could also be marked in some happy color
distinct from the usual green.

It would also be useful to have an overview of known compiler issues
somewhere on the web-page, and a pointer to that info in the
bug-reporting chapter in the manual. A bit like the freebsd problems
listed in recent NEWS.

>   BTW, it would be nice with an alternative test result page with
>   problematic platforms only, or sorted with failing platforms at the top.
>   
> Not sure what you mean.  Just problematic confs, with the problematic
> confs on top?  Or do you mean to put clang test results on a separate
> page?

Either some way to sort the current list so that builds with problems
are at the top, followed by the all-green entries.

Or a separate page, where the all-green entries are omitted.

With the current list, quite a lot of scrolling is needed to identify
the set of failed builds.

Regards,
/Niels

-- 
Niels Möller. PGP-encrypted email is preferred. Keyid C0B98E26.
Internet email is subject to wholesale government surveillance.