GSL and MinGW with the GCC 4.4.0

GSL and MinGW with the GCC 4.4.0

[Massimo Gaspari]

Dear all,

Platform : Windows XP Sp3 , MinGW latest binutils,make,runtime GCC 4.4.0

MinGW team has just release and official version of the new GCC compiler 
(4.4.0). I used it to compile (./configure --disable-shared) GSL 1.12 . 
Using CFLAGS = -fexception the checking (make check) completed successfully.

But using optimization (CFLAGS = -O2 -fexceptions) the testing fails at 
(first error)

make[2]: Leaving directory `/home/Massimo/gsl-1.12/integration'
make  check-TESTS
make[2]: Entering directory `/home/Massimo/gsl-1.12/integration'
FAIL: qawc(f459) elist (1.8329242496630777e-015 observed vs 
1.8330829482071535e-015 expected) [744]
FAIL: test.exe

I compiled the source code for GSL 1.12 successfully (well, disregarding 
the well known montecarlo optimizer bug) with GCC 3.4.5 and GCC 4.2.1. 
So it seems a new bug of the new GCC compiler (optimizer) in MinGW. I 
don't know if the same bug is also present in other platforms.

I also tested the TDM GCC 4.4.0 (R2 available at 
http://www.tdragon.net/recentgcc/ )  and I got the same failure using 
optimization.

Is there a way to complete testing instead of stopping at first error?



Hop this helps.

Re: GSL and MinGW with the GCC 4.4.0

[Brian Gough]

At Tue, 23 Jun 2009 23:37:10 +0200,
Massimo Gaspari wrote:
> 
> Is there a way to complete testing instead of stopping at first error?

Use "make -k check" to continue running the remaining tests in the
event of failures. To see the complete output, set the environment
variable GSL_TEST_VERBOSE=1.

-- 
Brian Gough

Network Theory Ltd,
Publishing Free Software Manuals --- http://www.network-theory.co.uk/

Re: GSL and MinGW with the GCC 4.4.0

[Massimo Gaspari]

Brian Gough wrote:
> At Tue, 23 Jun 2009 23:37:10 +0200,
> Massimo Gaspari wrote:
>   
>> Is there a way to complete testing instead of stopping at first error?
>>     
>
> Use "make -k check" to continue running the remaining tests in the
> event of failures. To see the complete output, set the environment
> variable GSL_TEST_VERBOSE=1.
>
>   

Thanks for the info. So I had the old bug in montecarlo (vegas)  plus 
the new one in the integration routines. Using -ffloat-store both of 
them disappear as usual.

Max