Re: [boost] stdcxx-tests on release branch?

From: Beman Dawes (bdawes_at_[hidden])
Date: 2008-02-07 08:47:39

John Maddock wrote:
> Without wanting to be too pedantic (!), do we have a policy about which 
> toolsets are tested on the release branch?  I'm asking because we currently 
> have a large number of failures for the "stdcxx" (Apache std lib) tests. 
> Whilst we certainly should be testing this on the trunk going forward, it's 
> not previously been a required release platform, hence the large number of 
> failures.

The policy is to test our *high priority* platforms only. (Note that we 
agreed to use that term in preference to *required*, to avoid 
prejudicial views of the other platforms.)

> 
> There are also one set of results (for msvc-8.0-stdcxx-4.2.1) that appear to 
> failing almost everything and isn't even compiling the "hello world" 
> config_info program: 
> http://beta.boost.org/development/tests/release/developer/output/msvc-8-0-stdcxx-4-2-1-boost-bin-v2-status-config_info-test-msvc-8-0-debug-stdlib-stdcxx-4-2-1-threading-multi.html
> 
> IMO this failure is made worse because the toolset used is simply called 
> "msvc-8.0" so it shows up as a *required* platform, rather than a 
> currently-unsupported one.  Likewise the Sunpro and gcc tests with the 
> apache std lib are simply called "gcc-4.2.1" and "sun".

I'm totally ignoring the results all of those stdcxx tests as far as the 
release  testing goes.

> Apologies to the test runner for the gripe: we really do need you and 
> appreciate the effort!  It's just that we also need to be able to clearly 
> focus on what is and is not required for *this next release*.

Right. These tests are appropriate and useful for the trunk, but just 
confuse people and get in the way for the release branch.

--Beman