running 4.4 libstdc++ testsuite using 4.5's libstdc++ shared library

[Matthias Klose]

On 26.06.2010 13:16, Paolo Carlini wrote:
> Hi,
>
>> running 4.4's libstdc++ testsuite using 4.5's libstdc++ shared library shows some additional test failures, compared to running 4.4 libstdc++ testsuite using the libstdc++ from the same build.
>>
>> For two regressions found (running the binaries built with 4.4 on a system with 4.5's libstdc++ shared library) I filed PR44679 and PR44680,
>
> I will look into those later today, but frankly I don't understand the rationale of this kind of testing: if I fix a bug in 4.5, which is part of the code in the .so, and at the same time I adjust only the 4.5 testsuite to match, of course the 4.4 testsuite will show spurious regressions. And since we are not removing testcases from the testsuite, I don't see what genuine problems we are hoping to catch with that kind of obscure testing... Bah

Thanks.

How would I differentiate between a genuine problem and a problem in the 
testsuite?  When a distribution starts an update from 4.4 to 4.5, not everything 
is rebuilt, so you end up with the new shared libstdc++ and packages still built 
with 4.4.  Same thing if a third party distributes software linked against 4.4's 
libstdc++, and trying to run on a system with 4.5's libstdc++.  The distribution 
upgrade scenario is a real issue for, I don't know if the third party case is 
relevant.

   Matthias