static C++ ABI testing

[Benjamin Kosnik]

>The LSB would like to be able to reuse GCC's libstdc++ DejaGnu tests for
>static C++ ABI testing similar to the way the LSB reuses the POSIX libc
>tests for ABI conformance testing.  First, if we built the libstc++
>tests on one Linux distribution and ran them on another system without
>recompiling, then will that tell us anything?  

I'm sure it will tell you something. I'm interested in this approach,
and am curious to see what you find with it. Note Janis (IBM) was doing
some kind of ABI testing along these lines: you might want to talk to
her about this.

> Would the GCC team
>object to us trying to port the libstc++ tests from the DejaGnu
>build/test environment so that we can create these prebuilt images?

I wouldn't. There are two parts to this: the test harness, and the
tests.

There exists a love/hate (hate/hate?) relationship with DejaGnu: that's
why the bash script is also available. At some point in the near future
I'd love to see support for QMtest. I'm not a super big fan of LSB's use
of TET, but hey. If that's what you want to use.... use it.

I don't speak for the GCC team but I think what the LSB is trying to do
is the right idea. You will have to comply with the availability of
sources bits, so people can reproduce what the LSB is distributing, but
I don't think that will really be a problem, since they are part of the
gcc distro and theoretically you could just point at these sources. Do
you anticipate problems?

>http://freestandards.org/pipermail/lsb-cxx/2003-October/000031.html

Be forwarned that mainline gcc has around 2335 libstdc++ test cases at
the moment. Some of these don't make sense or do not apply to the
gcc-3.4 codebase. This message talks about gcc-3.3.2, so maybe you have
another round of surprises!

It would be best for the LSB testing group to just plan on periodic
merges with released libstdc++ versions for testsuite refreshes. I think
you can count on a strictly increasing number of tests. Hopefully the
major rearrangements will not re-occur.

>Lastly, does anyone see any problems with GCC's libstdc++ tests being
>GPLed instead of LGPLed?

No. The actual license, FWIW, is GPL plus exception. See the libstdc++ FAQ.

-benjamin