|
|
Boost Testing : |
From: Anthony Williams (anthony_w.geo_at_[hidden])
Date: 2006-11-16 03:14:07
"AlisdairM" <alisdair.meredith_at_[hidden]> writes:
> Anthony Williams wrote:
>
>> One thing possibly worth noting: I'm running these tests with
>> bjam_options="release".
>
> That could well turn up a new set of errors on the Borland compiler.
> Exception handling has been known to play havok with their
> code-generator when you allow function inlining.
It seems to be turning up a couple of new errors on all compilers.
This seems problematic to me.
Firstly, if a compiler is known to have a bug in the code generator with
particular options, we shouldn't be enabling those options in bjam, and we
should document that these options shouldn't be enabled.
Secondly, I would expect most applications to be compiled with what amounts to
"release" options --- full optimization, no debug symbols, etc. If boost code
doesn't work under these circumstances, that's problematic. For some of my
apps I've seen an order of magnitude difference in performance between debug
and release builds. If a boost library doesn't work in release builds, it
means I can't use it in that application.
Thirdly, given the above, why aren't there more people running "release"
tests?
I'm seeing failures in bind, graph, iterator, numeric-interval, python,
random, range, regex, spirit and test on msvc-8.0express that aren't showing
up in the RudbekAssociates-V2 tests.
Anthony
-- Anthony Williams Software Developer Just Software Solutions Ltd http://www.justsoftwaresolutions.co.uk