[boost] Nitpicking Win32 Regression Test Results

From: Douglas Gregor (gregod_at_[hidden])
Date: 2003-12-07 18:51:59

I'm picking through the Win32 regression test results to see where we stand. 
Here are some concerns I have (on a per-library basis):

Filesystem: operations_test is failing on Borland but not anything else, even 
though most of the compilers use the same code as Borland.

Format: Total failure on CodeWarrior 8.3, Intel 7.0, and MSVC 6.0; we should 
probably either support these libraries or add notes into 
win32-notes-map.txt. (It looks like an Intel 7.0/MSVC 6.0 library bug; 
potentially a name lookup bug in CW 8.3).

IO: Same failures as with format under Intel 7.0 and MSVC 6.0; standard 
library bug?

Lambda: totally broken on Borland, MSVC 6.0, and Intel 7.0; I've marked these 
as compiler bugs, because Lambda won't ever work on these compilers.

MPL: Untested (?)

Multi-array: Totally broken on Borland; I think we can work around this, 
though.

numeric/interval: Broken on Borland (looks like a standard library bug); 
Broken on MSVC 6.0 (looks like it thinks float/double are equivalent?)

numeric/ublas: Failures on CodeWarrior 8.3; I _think_ that the line it's 
pointing out is missing a "typename" (there are other occurrences of the same 
construct).

regex: Library build failure on CodeWarrior 8.3

spirit: Untested (?)

type_traits: 
        alignment_of_test/type_with_alignment_test on CodeWarrior 8.3: I can't make 
heads or tails of this. We compute the alignment of a pointer-to-member 
function one way and get 8; then compute it a different way and get 12. I 
haven't a clue which is right, but this is _strange_.

        remove_xxx: Borland fails most of these; is this a compiler limitation? MSVC 
6.0 and 7.0 fail all of them (no partial specialization support), so I've 
noted this in win32-notes-map.txt.

utility: enable_if_* tests fail on Borland, CW 8.3, MSVC 6.0, and MSVC 7.0; 
I've noted that they're all due to compiler bugs.

In a perfect world, I'd love to see all of the tests either passing or noted 
as a compiler/platform/standard library bug for a release. Not holding my 
breath, of course :)

        Doug