Re: [boost] [Report] 2749 regressions on RC_1_34_0 (2007-01-24)

From: Joaquín Mª López Muñoz (joaquin_at_[hidden])
Date: 2007-01-25 10:12:40

Arkadiy Vertleyb ha escrito:

> "Stefan Seefeld" <seefeld_at_[hidden]> wrote
>
> > If the rate regressions were fixed over the last couple of weeks is any
> > indication, getting from this high a number down towards zero will take
> > another year or three... :-(
>
> What happened?  I thought we were down to 6 failures at some point...  Are
> all these new failures just indication of a problem in some low-level
> library?

AFAIK things aren't as bad as they look:

* Scores of tests from the "Martin Wille x86_64 V2" regression runner are
bogus entries due to an as of yet unresolved problem with the tool used
to run the regression procedure, see
http://lists.boost.org/boost-testing/2007/01/3853.php

* The "speedsnail" regression runner has some  peculiarities:
1. It builds the tests in release mode (although it gives no indication of
this in the toolset names used) This seems to cause problems with some
facilities from Boost.Test.
2. It fails to use Spirit 1.6 to build Boost.Serialization, which is a requirement

for Borland and older MSVC compilers, hence all serialization tests (1583)
and some more depending on serialization are guaranteed to fail.
3. Boost.Regex fails to build because it refers to a seemingly nonexsistent
function stdext::unchecked_equal: my hunch is that this is also related to the
fact builds are in release mode, but this should be investigated. Again, this
cascades to dozens of errors.

If you factor out these spurious problems, things should look more or less
like they did a month ago: basically, the source code has been pretty stable
during the last weeks.

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo