Re: __builtin_add_overflow on clang

[Paul Eggert]

Mattias Engdegård wrote:
> On macOS (and *BSD, probably), compiler-rt is linked by default so the problem 
> never arises there. Much as I dislike them in general, perhaps a configure test 
> would be appropriate?

I've been avoiding configure-time tests for intprops.h, since we haven't needed 
them yet and intprops.h is occasionally used outside Gnulib. And strictly 
speaking we don't need the tests here, as the fallback code works correctly for 
Clang (albeit most likely not as efficiently).

The reason I patched Gnulib in 2017 to work around the Clang bug:

https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=04441fd8156978cfc51d6c203fd43c23b64f95df

was in response to a private bug report for GNU 'sed' builds from Nelson H.F. 
Beebe, who reported finding the Clang bug on these platforms:

>     alpine3           dragonflybsd47    mageia6           salix142
>     alpine343         fedorarh          mint18            slackware142
>     alpine35          fedorarh          mint181           slackware142x
>     beta              gentoo            openbsd60         solus12
>     clearos7          grousecreek       opensusetw        ubuntu1610
>     dragonflybsd46    kali2             pclinuxos         ubunturr
>     dragonflybsd461   mageia6           pclinuxosb        void64
>     dragonflybsd46b

These failures occurred with ordinary integer types, as sed doesn't use 128-bit 
types.

Beebe also wrote:

> I fear that others might hit the problem, and because Mac OS X
> and recent FreeBSD have switched to clang as the default C compiler,
> the same problem might start showing up on those platforms.

Given the prevalence of this Clang bug, and the iffiness of whether the bug 
might occur on macOS or FreeBSD, I was cautious and stuck with the fallback code 
for Clang. However, now that you mention it this problem seems limited to 
multiplication and it's easy to change intprops.h to use Clang's 
__builtin_add_overflow and __builtin_sub_overflow even if its 
__builtin_mul_overflow is buggy. So I installed the attached patch into Gnulib 
to do that.

If we definitely know that this bug cannot occur on FreeBSD and/or macOS, I 
suppose we could add further tests on the appropriate preprocessor macros. But 
that would require further investigation as to what packages are available in 
these platforms and whether the relevant clang runtimes can be installed separately.

0001-intprops.h-verify.h-port-better-to-clang.txt
Description: Text document