check problems with -g/-m64

[IainS]

On 10 Jun 2008, at 14:40, Tobias Burnus wrote:

> IainS wrote:
>> make check with  -m64 on
>> powerpc-apple-darwin8 is throwing our a huge number of extra  
>> errors when checking gfortran (c.f. default, m32).
> I think this is a known problem, which won't be fixed.

thanks, you hit the nail right on the head.

> If I read http://gcc.gnu.org/ml/gcc-patches/2007-02/msg00735.html
> (plus referred/follow up emails) correctly,
> one can replace -g by -gdwarf-2 - then it works

I had a misunderstanding that this was the bug that "-pipe" was  
supposed to work round.

> For the test suite, one could do so, e.g., in
> gcc/testsuite/lib/fortran-torture.exp and .../gcc-dg.exp.

I did this (in all the places that torture tests are declared) ; and,  
for the record this is effect
[powerpc-apple-darwin8 host/build/target, current 4.4.0trunk]  :

-m32  [ -g as per current tree ]
		=== gfortran Summary ===

# of expected passes		26084
# of unexpected failures	17
# of expected failures		40
# of unsupported tests		132


-m32 [ -gdwarf-2 change ]

		=== gfortran Summary ===

# of expected passes		26084
# of unexpected failures	17
# of expected failures		40
# of unsupported tests		132

-m64 [ -gdwarf-2 change ]

		=== gfortran Summary ===

# of expected passes		26231
# of unexpected failures	18
# of expected failures		46
# of unsupported tests		59

(the one extra failure is  gfortran.dg/equiv_7.f90  -Os  execution  
test - unrelated to the dwarf change, AFAICT).

=== some additional questions.

1/ Are tests with changes such as these acceptable on gcc- 
testresults@gcc.gnu.org ?

At present the torture "-O3 -g" statements are in apparently global  
scopes [to a given test set].
2/ is there a way I can over-ride those specific to *-apple-darwin8    
[without editing every test case?]

I seem to recall reading that the sequence  "-O3 -g" is there so that  
the -g preserves symbols that might, otherwise, be removed by O3.
(I can't find the reference now I look - but I also seem to recall  
that this assumed dwarf).

if that's the case, then making -gdwarf specific in the tests would  
be possibly useful?

thanks
Iain