Re: testsuite failures on solaris

[Benjamin Kosnik <bkoz at redhat dot com>]

I don't know what to tell you. 

In summary, the problem is the following: configuration time happens
before the C++ includes are "constructed." Thus, a test of the things
that are going to be used is not possible, so instead, a simpler test is
used. This simple test passes. This flips the configuration bit and
everything charges ahead as if the more complex thing will, in fact,
work. 

It doesn't. 

I this case, it doesn't really mean that the built library is wrong, and
actually installing and using the compiler/library will be fine, as long
as the precompiled header isn't used (it's not by default, so it's not
an issue.) You can verify this by either running 'make check' of the
library with testsuite_flags configured to not include the PCHFLAGS, or
removing the generated PCH files in the stdc++.h.gch directory and
touching them with a zero-size file (ie, what is done for stdc++.h
itself).

The real fix to this issue would be to find a configure test that is
more complex than the simple test done currently, one that fails in your
case, and use it. I don't know what that test would be, sadly. 

If you feel motivated, you might want to try to experiment. Start by
trying to compile more than math.h in the test (and maybe include some
actual c++ code?) 

-benjamin