Re: [committed] Skip ICEing -frtl-abstract-sequences tests for MIPS

[Mark Mitchell <mark at codesourcery dot com>]

Richard Sandiford wrote:

> TBH, I still support your original stance.  XFAILing can often be
> a long-term thing that lasts for several releases.  This particular
> failure is not something we want to leave in its current state for
> the next release

I still think the bug-tracker is for bug-tracking, while the testsuite 
is for validation.  The primary point of XFAILs is so that someone 
testing can know whether or not their toolchain meets our expectations 
for quality.  If something causes an ICE, then there ought to be an 
entry in the bug tracker -- possibly with quite high priority -- and an 
XFAIL -- since we know the test fails.

> If there are strong objections to using skips this way, I'd rather
> make -frtl-abstract-sequences trigger a sorry() for MIPS and PowerPC
> instead of adding support for XFAILing ICEs.

I think it's reasonable to argue for the sorry, but that we should still 
have the ability to XFAIL ICEs.  If we expect a test case to cause an 
ICE in the current toolchain, then I think that meets every requirement 
of an XFAIL.  We can XFAIL wrong code generation, and that's often a 
worse failure than an ICE.

--
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713