Re: fixes for Clang builtins when compiling Emacs on Fedora

[Bruno Haible <bruno@clisp.org>]

Hi Paul,

> I had a couple of problems with the recent Clang-related fixes on Fedora 31 when 
> building Emacs, and worked around them by installing the attached patches

Yes, things are a bit more complicated in practice than in theory... I have just
completed the testing on FreeBSD 12, and will test on OpenBSD and macOS next.

> +	libc-config: avoid Clang’s __diagnose_if__
> +	* lib/cdefs.h (__warndecl, __warnattr, __errordecl):
> +	For now, do not use __diagnose_if__ here, as this fails
> +	on Fedora 31 with Clang 9.0.1, with diagnostic
> +	"/usr/include/bits/stdio2.h:263:9: error: fgets called with bigger
> +	size than length of destination buffer

This is a pity, because these user-defined diagnostics are useful, and
clang does not have __attribute__((__warning__ ...)) that GCC has.

This piece of code in stdio2.h looks like this:

extern char *__REDIRECT (__fgets_chk_warn,
                         (char *__restrict __s, size_t __size, int __n,
                          FILE *__restrict __stream), __fgets_chk)
     __wur __warnattr ("fgets called with bigger size than length "
                       "of destination buffer");

__fortify_function __wur char *
fgets (char *__restrict __s, int __n, FILE *__restrict __stream)
{
  if (__bos (__s) != (size_t) -1)
    {
      if (!__builtin_constant_p (__n) || __n <= 0)
        return __fgets_chk (__s, __bos (__s), __n, __stream);

      if ((size_t) __n > __bos (__s))
        return __fgets_chk_warn (__s, __bos (__s), __n, __stream);  // <=== line 263
    }
  return __fgets_alias (__s, __n, __stream);
}

Most likely the cause is that __bos (__s) currently evaluates to 0,
because I did not port the __bos macro, because the corresponding
attribute works differently in clang than in GCC
<https://clang.llvm.org/docs/AttributeReference.html#alloc-size>.

If you have time to finish this corner of clang support, please go
ahead!

> +   Avoid Clang’s __builtin_assume, as clang 9.0.1 -Wassume can
> +   generate a bogus diagnostic "the argument to '__builtin_assume' has
> +   side effects that will be discarded" even when the argument has no
> +   side effects.  */

Do you have a test case, that we could check on clang 10 and on future
clang versions?

Bruno