changing testcases (don't.)

[Geoff Keating]

I'd like to ask contributors _not_ to change existing testcases except
to remove bugs in the testcases.  It causes trouble like this:

> From: Jan Hubicka <hubicka@atrey.karlin.mff.cuni.cz>

> The cast-qual mentioned in the email don't seems to be caused by me 
> (I believe) and it has passed (according to the logs) in my make check.

Now, the reason Jan is confused is that his version of
testsuite/gcc.dg/cast-qual-1.c does in fact pass.  The testcase
was changed, and now it doesn't pass.  Is this a real regression
caused by Jan's patch?  There's no way to know; I'd have to revert the
testcase change (or the patch) to find out.

The problem is even worse when it's a program trying to interpret the
testcase results.  I think it's not unreasonable to assume that if a
particular testcase used to pass, and now it fails, then that
represents a new bug in gcc.

Happily, I think this testcase change was in fact wrong anyway; so
I'll probably just revert it.  I'd like to avoid this happening in
future, though.
-- 
- Geoffrey Keating <geoffk@cygnus.com>