[parisc-linux] glibc/gcc bug -> perl/gcc bug?

[Carlos O'Donell <carlos@baldric.uwo.ca>]

> perl-5.8 fails to build on hppa because of a test failure. The problem
> can be seen very easily:
> 
> tausq@gsyprf11:~/build/perl-5.8.0$ cat t/t.pl
> print log "A";
> tausq@gsyprf11:~/build/perl-5.8.0$ ./perl t/t.pl
> Can't take log of 2.75773e-308 at t/t.pl line 1.
> 
> that should have said:
> tausq@gsyprf11:~/build/perl-5.8.0$ /usr/bin/perl t/t.pl
> Can't take log of 0 at t/t.pl line 1.
> 
> interestingly, a debug build of perl (using -g and no -O flags) doesn't
> have the same problem, so this looks like some kind of compiler
> optimization bug. I've tried this with both 3.2.3 and 3.3... same
> problems.
> 
> would anyone like to try to look at this some more and see if they can
> isolate the C code/a small test case that is causing the problem?

Glibc's math tests are still failing even under gcc-3.3 (not to mention
the threading problems I'm in the middle of fixing).

Though I'm seeing more:

Failure: Real part of: cacos (NaN + inf i) == NaN - inf i: Exception "Invalid operation" set
Failure: Real part of: cacos (NaN - inf i) == NaN + inf i: Exception "Invalid operation" set
Failure: Real part of: cacos (NaN + NaN i) == NaN + NaN i: Exception "Invalid operation" set

Or:
Failure: Test: Imaginary part of: ctanh (NaN - 0 i) == NaN - 0 i
Result:
 is:          nan   nan
 should be:  -0.00000000000000000000e+00  -0x0.00000000000000000000p+0

Even with a new ulps, the tests were still failing.

c.