Boost Testing :
From: Robert Ramey (ramey_at_[hidden])
Date: 2005-07-19 14:35:56
Thats's more or less correct. Due the to way borland handles const in
templates, I had to tweak something for borland that has these test failures
as a side effect. For borland that leaves only the following:
a) test_no_rtti - inexplicable to me. I've come to believe for a couple of
reasons that my no_rtti facility isn't really quite ready for prime time. I
suspect this has never been an issue as I doubt if anyone uses it.
b) On my machine test_variant fails for borland - but it doesn't show in the
test matrix. I suspect there are couple of different versions of borland
c) On one or both versions of borland - I had failures when attempting to
test in release mode due to failures in the Boost Test. I would have hoped
that this might be corrected
Anyway, I'm amazed that I got (almost) everything to work with borland -
much better than msvc 6.5 after all.
I'll mark the 3 tests above as expected failures per your suggestion.
> Robert Ramey wrote:
>> As I write this ,I have no more corrections to make to the
>> serialization library. There are a number of test failures, but I
>> cannot resolve them for various reasons.
> I think the 4 compile-fail tests that fail (i.e. compile) with Borland
> 0x564 can be marked expected fail, although that is gut instinct
> rather than reducing to a proven test case that no workaround is
> possible. It is the nature of the compiler to accept certain kinds
> of badly formed code, especially wrt constness, and I suspect that
> may be what is happening here?
> I am still investigating the other failure
> (and hopefully the Boost Test failure for floating point too)