Slight rounding difference in core tests

[dbr]

The change on b7dd8d8 breaks the tests for me on OS X 10.6/10.7:

.../src/core/FileFormatHDL.cpp:1449:
FAILED: OCIO::pystring::strip(osvec[i]) == OCIO::pystring::strip(resvec[i])
    values were '0.732058' and '0.732057'

No idea why it would cause the slight difference.. Only vaguely useful thing I can think to say is I'm using i686-apple-darwin11-llvm-g++-4.2, and the executable was built with the following command (excluding a bunch of .o files):

/usr/bin/c++   -Wall -Wextra -Wshadow -Wconversion -Wcast-qual -Wformat=2 -msse2 -O2 -g -arch x86_64 -isysroot /Developer/SDKs/MacOSX10.7.sdk -Wl,-search_paths_first -Wl,-headerpad_max_install_names  [....]  -o ocio_core_tests  ../../ext/dist/lib/libtinyxml.a ../../ext/dist/lib/libyaml-cpp.a

[jeremyselan]

Thanks for the heads up. Yah, I see that on OSX too. Not sure what the cause it. Probably wont have time to look into it in the short-term, but if you care to investigate it further please fill me in on what you find.

[dbr]

Interesting, the value that differs is 4.0/9.0 ** (1/2.6), and:

$ uname -s
Darwin
$ gcc -v
Using built-in specs.
Target: i686-apple-darwin11
Configured with: /private/var/tmp/llvmgcc42/llvmgcc42-2335.15~25/src/configure --disable-checking --enable-werror --prefix=/Developer/usr/llvm-gcc-4.2 --mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --program-prefix=llvm- --program-transform-name=/^[cg][^.-]*$/s/$/-4.2/ --with-slibdir=/usr/lib --build=i686-apple-darwin11 --enable-llvm=/private/var/tmp/llvmgcc42/llvmgcc42-2335.15~25/dst-llvmCore/Developer/usr/local --program-prefix=i686-apple-darwin11- --host=x86_64-apple-darwin11 --target=i686-apple-darwin11 --with-gxx-include-dir=/usr/include/c++/4.2.1
Thread model: posix
gcc version 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)

$ cat roundtest.c 
#include<iostream>
#include<math.h>
int main(){
    std::cout << powf(4.0/9.0, (1.0/2.6)) << "\n";
}
$ g++ roundtest.c && ./a.out
0.732058
$ python2.7 -c 'print "%.06f" % (4.0/9.0)**(1/2.6)'
0.732057

Where the expected value on Linux is 0.732057

Wolfram|Alpha agrees with Linux, http://www.wolframalpha.com/input/?i=%284.0%2F9.0%29**%281%2F2.6%29

[dbr]

Oh, it's a gcc version thing:

$ uname -s
Linux
$ run in gcc/4.1.2 : g++ roundtest.c && ./a.out
0.732057
$ run in gcc/4.4.0 : g++ roundtest.c && ./a.out
0.732058

[dbr]

Asked about this on stackoverflow,

http://stackoverflow.com/questions/8802831/gcc-rounding-difference-between-versions

[jeremyselan]

Interesting result (that it changes between gcc versions), and +1 for the nice stackoverflow answers.