Re: G++ test failures

[Mike Stump <mrs at windriver dot com>]

> Date: Wed, 22 Nov 2000 11:51:55 -0800 (PST)
> From: Benjamin Kosnik <bkoz@redhat.com>
> To: Mark Mitchell <mark@codesourcery.com>

> I would suggest that the g++ testcases use non-deprecated parts of
> the library, thus side-stepping this issue.

The right way to do this, is to run the testsuite after a library
change, and if a portion of the testsuite is wrong, then to fix the
testcases in question to be more compatible with the letter and intent
of the Standard.  As that happens, it might be worthwhile to add
documentation on how to fix old, broken code to more fully conform to
the Standard.  This way, we ease the transition for users of our
software.

These types of changes are best made by the same person that changes
the library, as they are authoritative for the behavior.