hughperman / pure-lang

Automatically exported from code.google.com/p/pure-lang
0 stars 0 forks source link

Pure 0.55 test020 fails on Fedora's Rawhide with LLVM 3.1 #72

Closed GoogleCodeExporter closed 8 years ago

GoogleCodeExporter commented 8 years ago
What steps will reproduce the problem?
1. Install mock
2. Recompile the F17 source RPM 
(http://kojipkgs.fedoraproject.org/packages/pure/0.55/1.fc17/src/pure-0.55-1.fc1
7.src.rpm) on Rawhide: mock -r fedora-rawhide-x86_64 ./pure-*.rpm

What is the expected output? What do you see instead?
test020 passes as in F17 / LLVM 3.0 -- instead it fails. Some +ve infinities 
become negative instead.

What version of the product are you using? On what operating system?
Pure 0.55, LLVM 3.1, GCC 4.7.0 (compiling with GCC instead of clang)

Please provide any additional information below.

Original issue reported on code.google.com by Michael....@gmail.com on 5 Jun 2012 at 8:06

Attachments:

GoogleCodeExporter commented 8 years ago
Pure uses the C pow() function to calculate this value. According to my copy of 
ISO/IEC 9899:TC3, Annex F.9.4.4, pow(0, -INFINITY) in C and thus 0^(-inf) in 
Pure must return inf (and on my system it sure does).

Can you please execute the attached C program? It should print:

pow(0.0, -INFINITY) = inf (positive? 1)

Original comment by aggraef@gmail.com on 5 Jun 2012 at 10:05

Attachments:

GoogleCodeExporter commented 8 years ago
Looks like it's another regression in glibc -- reported on our bug tracker, 
thanks.

https://bugzilla.redhat.com/show_bug.cgi?id=828980

Original comment by Michael....@gmail.com on 5 Jun 2012 at 5:15

GoogleCodeExporter commented 8 years ago
Pure shaking the bugs out of bleeding edge glibc, that's nice. :)

Original comment by aggraef@gmail.com on 5 Jun 2012 at 6:09