[Bug middle-end/50724] isnan broken by -ffinite-math-only in g++

["marc.glisse at normalesup dot org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50724

--- Comment #13 from Marc Glisse <marc.glisse at normalesup dot org> 2011-10-15 09:05:57 UTC ---
(In reply to comment #11)
> I'm guessing (and apologies if this is inaccurate) that this might boil down to
> saying that you want to interpret an end user setting -ff-m-o as an opportunity
> to skip validating their input or skip doing assertions during its processing,

Yes, exactly.

> which could be a reasonable thing to do, but that's a choice I'd rather leave
> to individual developers, e.g. can also wrap code with #if
> __FINITE_MATH_ONLY__==0 or such...

Not very portable afaik.

> Or in other words, it's only a missed optimization if you wind up with
> classification calls,

-ffinite-math-only exists for the sole purpose of allowing extra optimizations
at the expanse of correctness if the contract is broken.

> whereas it's a full-fledged execution error when NaN gets
> past validation.

A full-fledged user error to have used this flag ;-)

Actually, I can understand why you'ld want to be able to write
assert(!isnan(x)) and get a useful result. If on the other hand you want
if(isnan(x))f();else g(); then g should be in a separate translation unit that
can use -ffinite-math-only (and using LTO may require a lot of prayers in that
case) or it could use the optimize attribute/pragma. This could work for the
assertion as well: have a validate_input function compiled without the flag.

I am not sure if it is possible to have it both ways: be able to validate the
input and still be able to optimize headers.