Re: PATCH: XFAIL tests that aren't regressions

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

On Monday 30 April 2001 18:24, Mark Mitchell wrote:
> >>>>> "Franz" == Franz Sirl <Franz.Sirl-kernel@lauterbach.com> writes:
>
>     Franz> The test belongs into the ieee subdir because it relies on
>     Franz> correct IEEE math, it's just that ieee.exp shoudn't use
>     Franz> -ffloat-store unconditionally on all platforms.
>
> I guess I don't quite understand this.  Take it one level further,
> please. :-)

Well, ppc for example passes this test without any special options (in fact 
this testcase is directly from the math code in glibc). That means that on 
ppc all tests in ieee/ should pass even without -ffloat-store, and a quick 
test shows it actually passes all tests without.

So I think ieee.exp should add -ffloat-store only for some blacklisted 
platforms, otherwise we might just hide ieee problems on conformant platforms.

So, my suggestion is to move the test tto the ieee dir on both mainline and 
branch first (removing the XFAIL), and add a blacklist in ieee.exp in the 
mainline. Does this sound reasonable?

>     Franz> So I would say move the test on both mainline and branch,
>     Franz> and put fixing ieee.exp on the mainline TODO list.
>
> What benefit to we get out of moving the test on the branch?  Will we
> test things better, or will users see fewer unexpected failures?

Well, the test passes fine on PPC and other platforms, and I checked it 
passes on x86 when in the ieee dir. The only thing I don't know for sure is 
if the bug still would trigger with -ffloat-store on PPC, but as far as I 
remember the bug, it was about handling UNKNOWN comparison codes and that 
should be independent of -ffloat-sttore.

BTW, dumb question, is the website with the database of all testresults 
posted no longer online/active/uptodate? It should be easy to find a tighter 
check than xfail *-*-* in there in the first place.

Franz.