From: Gennaro Prota (gennaro_prota_at_[hidden])
Date: 2006-07-20 21:10:40
On Thu, 20 Jul 2006 19:54:47 +0200, Joaquín Mª López Muñoz
>I'm not so sure about those reasons you propose. Having a cursory inspection
>of the regression tests landscape, I'd say the main sources of red and yellow
>cells are, roughly ordered in decreasing importance measured as no of
>1. More or less exotic compilers for which nobody cares to try to fix things or,
>at least, mark whole libs as unusable on them.
>2. Toolsets with some kind of configuration problem.
>3. Abandoned libs.
>4. Libs which are strictly not abandoned (the author hangs around the lists)
>but need some more attention: even if the codebase does not change,
>problems in 3rd party libs can result in regressions here.
You may be right. It would probably help to be explicit and tell what
libraries belong to what category. I don't think there's any reason
for authors/maintainers to be offended. We are all here for the same
purpose, and this can be a good way to improve our (way to) work.
>Beman Dawes proposed a very attractive maintenance and release
>scheme designed to avoid this kind of situations.
I could only have a quick look, so I'll not comment on it yet.
As to our regression reports I have noticed more "oversights" compared
to earlier releases, e.g. forgetting to build libraries (a while ago
almost all STLport 5 tests were failing because the iostream part of
the library hadn't been built) and introduction of new compilers for
which the library author was unprepared (as Dave has pointed out too,
in French :-)). I also seize the occasion to say that some tests don't
have to be "fixed" at all costs. I've just added a note about an old
failure concerning dynamic_bitset references in conjunction with
STLport 4.5.3: I could have probably found a workaround for it (though
admittedly that wouldn't have been an easy task) but I didn't even
try, as the error depend on a _serious_ flaw of STLport which can (and
will) show up independently from dynamic_bitset. This is a typical
case where an annotated failure is much more informative than a
-- [ Gennaro Prota, C++ developer for hire ]
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk