Bug 198829 (wfmath)

- rpmlint output clean
- package named according to package naming guidelines
- spec filename matches base package %{name}
- package meets packaging guidelines
- package licensed with open source compatible license
- license in spec file matches actual license
- license contained in %doc
- spec file written in American english
- spec file legible
- sources match upstream
6a14f7de9d467d7b72b37da5ca5f92c5  wfmath-0.3.4.tar.gz
- package successfully compiles and builds on x86_64 FC-5
- All BuildRequires listed
- no locales
- ldconfig called in %post/%postun
- package is not relocatable
- package owns all directories it creates
- package contains no duplicate files
- file permissions set properly
- contains proper %clean section
- macro usage is consistant
- package contains permissible content
- no large documentation
- %doc does not affect runtime of application
- header files are in devel package
- pkgconfig .pc file in devel
- .so files w/o suffix in devel
- devel package requires base package
- package does not contain .la files
- not a GUI app needing a .desktop file
- package does not own files or directories owned by other packages


==== MUST ====
- Add Requires: pkgconfg to devel package
- package does not contain a %check section
 - make check fails on 1 test, investigate why (FAIL: intstring_test)

(In reply to comment #1)
> ==== MUST ====
> - Add Requires: pkgconfg to devel package
> - package does not contain a %check section
>  - make check fails on 1 test, investigate why (FAIL: intstring_test)

I'll contact upstream and see if they have a fix or explanation for this test
failure.  It seems to only fail on x86_64, not i386.  I'll make a new package
once the test failure is resolved.

I saw the long list of review request for worldforge, so I thought I would give
a hand. But as usual you guys (Chris this time) have beaten my to it.

So I saw this 64bit problem and I thought I would/could give a hand. The test
fails because of 2 problems

1) wfmath/MersenneTwister.h, line 77 says:
 typedef unsigned long uint32;  // unsigned integer type, at least 32 bits
 Which is strange, because you would expect a uint32 to be 32 bits period,
 I've reviewed the rest of the code and it seems to assume 32bits everywhere/
 never return random values larger then 32 bit, so I believe this should be 
 changed to:
 typedef unsigned int uint32;
 This makes things work out as the name suggests and will also be faster since
 32 bit values are faster then 64 bit (even on x86_64).
2) wfmath/intstring_test.cpp, second test uses a long to restore the unsigned
 long value of the random function. The idea here is to cause negative numbers
 for 2^31 unsigned long numbers to test negative conversion however > 2 ^ 31 
 will fit fine in the long on 64 bits, without becoming a negative number.
 Next this long is passed to IntToString, who correctly converts this to a
 string holding a large (> 2 ^ 31) positive number, after which the string
 gets passed to atoi. However atoi returns an int which cannot represent this
 bug a positive number > caboom!

 Fix use (unsigned) int 's in the test instead of long's.

Thanks for the test investigation, Hans!

I built a patch to fix the test results.  Here's the new package that includes
the patch and the review MUSTFIX items:

http://www.kobold.org/~wart/fedora/wfmath.spec
http://www.kobold.org/~wart/fedora/wfmath-0.3.4-2.src.rpm

Only other comment is make check should use smpflags.  APPROVED.

Imported, but the build failed on ppc due to a single test failure.  I'll follow
up with upstream on this.

I'll defer closing this ticket until the test issue is resolved.

I disabled 'make check' on ppc until upstream can provide a fix.  Now that this
has been imported and built on devel I'm closing the ticket.