Re: patch/RFC/RFT detect ICE in libjava testsuite

[Janis Johnson <janis187 at us dot ibm dot com>]

On Mon, May 15, 2006 at 07:26:19PM -0600, Tom Tromey wrote:
> >>>>> "Janis" == Janis Johnson <janis187@us.ibm.com> writes:
> 
> Janis> The practical part of this is that an internal compiler error
> Janis> in a test that is already XFAILed is likely to go unnoticed.
> 
> Yes, and in fact we already have this problem with jacks -- jacks
> doesn't notice ICEs and so it incorrectly reports some passes that it
> should not.
> 
> If we really need this functionality, it needs a new flag.
> Overloading 'xfail-gcj' or whatever may hide regressions.

In response to the similar patch for gcc/testsuite tests, Mark Mitchell
responded in http://gcc.gnu.org/ml/gcc-patches/2006-05/msg00613.html:

    I agree that these limitations are OK, given that the goal is zero fails
    in the testsuite anyhow.  The only reasons for XFAILs are either
    regressions we mean to fix or failures we can live with; an ICE should
    always be in the first category.

The limitations are that you can't XFAIL an ICE failure, and there might
be additional FAIL messages for an ICE failure because the messages aren't
removed from the compiler output before other processing.

Janis