Thomas Munro <thomas.munro@enterprisedb.com> writes:
> On Tue, Sep 4, 2018 at 9:39 AM Andrew Gierth
> <andrew@tao11.riddles.org.uk> wrote:
>> I also confirmed that without #define isinf(x) __builtin_isinf(x), on
>> both 32bit and 64bit fbsd isinf() compiles as a function call, so the
>> OP's proposed change would not be desirable.
Presumably what we are looking at here is a compiler codegen bug for
__builtin_isinf(). The proposed change dodges it precisely by
substituting the library function instead --- at a performance penalty.
I agree that this isn't a real desirable fix, and we definitely ought
not cause it to happen on platforms that haven't got the bug.
> I installed FreeBSD 11.2 i386 on a virtual machine. I couldn't
> reproduce the problem with either the base cc (clang 6.0.0) or clang38
> (clang 3.8.1) installed via pkg.
> The OP reported clang 3.8.0, so a minor version behind what I tested.
I tried to reproduce the problem, without success, on a few different
FreeBSD images I had laying about:
FreeBSD 11.0/x86_64, clang version 3.8.0
(this confirms OP's report that x86_64 is OK)
FreeBSD 10.3/ppc, gcc 4.2.1
FreeBSD 12.0-CURRENT (from around mid-May)/arm64, clang version 6.0.0
(I was testing PG HEAD, not the 11 branch, but I don't see a reason
to think that'd make a difference.)
I also looked for evidence of a bug of this sort in the clang bugzilla.
I couldn't find anything, but it's possible that "isinf" isn't what
I should have searched on.
Anyway, my estimation is that this is a compiler bug that's been
repaired, and it probably isn't widespread enough to justify our
inserting some klugy workaround.
regards, tom lane