Re: PATCH: XFAIL tests that aren't regressions

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

On Monday 30 April 2001 16:43, Mark Mitchell wrote:
> These tests fail with GCC 2.95.x as well as with the current sources,
> so there is no point in XFAILing them on the branch.
>
> Tested on i686-pc-linux-gnu, installed on the branch.
>
> --
> Mark Mitchell                   mark@codesourcery.com
> CodeSourcery, LLC               http://www.codesourcery.com
>
> 2001-04-30  Mark Mitchell  <mark@codesourcery.com>
>
> 	* gcc.c-torture/execute/20010114-2.x: XFAIL it.
> 	* gcc.c-torture/execute/20010123-1.x: Likewise.
> 	* gcc.c-torture/execute/loop-2b.x: Likewise.
>
> Index: gcc/testsuite/gcc.c-torture/execute/20010114-2.x
> ===================================================================
> RCS file: 20010114-2.x
> diff -N 20010114-2.x
> *** /dev/null	Tue May  5 13:32:27 1998
> --- 20010114-2.x	Mon Apr 30 07:37:34 2001
> ***************
> *** 0 ****
> --- 1,2 ----
> + set torture_execute_xfail "*-*-*"
> + return 0

This test should be moved into the ieee subdir and already discussed that 
with Geoff and Joern. But then Geoff wondered why ieee.exp unconditionally 
uses -ffloat-store on all platforms and never OK'ed the move.

> Index: gcc/testsuite/gcc.c-torture/execute/20010123-1.x
> ===================================================================
> RCS file: 20010123-1.x
> diff -N 20010123-1.x
> *** /dev/null	Tue May  5 13:32:27 1998
> --- 20010123-1.x	Mon Apr 30 07:37:34 2001
> ***************
> *** 0 ****
> --- 1,2 ----
> + set torture_execute_xfail "*-*-*"
> + return 0

Just for your records, this is an unimplemented C99 feature AFAIK. It badly 
affects embedded platforms, cause writeable data may end up in .rodata and 
so booting such  code from ROM doesn't work.

> Index: gcc/testsuite/gcc.c-torture/execute/loop-2b.x
> ===================================================================
> RCS file: loop-2b.x
> diff -N loop-2b.x
> *** /dev/null	Tue May  5 13:32:27 1998
> --- loop-2b.x	Mon Apr 30 07:37:35 2001
> ***************
> *** 0 ****
> --- 1,2 ----
> + set torture_execute_xfail "*-*-*"
> + return 0

This is a regression from egcs-1.1.2, should be easy to fix, it seems to be a 
missing check for possible overflow in the loop code. I think I requested 
help on the list to fix that one.

I also requested comments on c/2100, but nobody cared yet :-/.

FYI, on powerpc-linux-gnu I have the following showstopper bugs:
- C: recent -fpic regressions on PPC
- C: excessive stack usage with inlined DImode values
- C++: FAIL: g++.pt/static11.C, a new-abi regression on PPC
- C++: FAIL: g++.abi/vtable2.C with -O2
- C++: artsd of KDE2 crashes on startup (maybe a threading problem of 
libstdc++-v3?)

In my 3.0 patch queue are:
- a PPC specific fix for a "unrecognizeable insn" ICE (will submit this week)
- a small CSE fix (will submit this week)
- better handling for weak symbols (will submit this week, nobody commented 
on the warning question, so I'll post the minimum version that just disables 
optimizations)
- objc vs. libgcc_s (submitted, waiting for approval)

Franz.