Segfaults on 'hard' conversions.

Hdf-forum archives
1

Hi

while conducting some extensive tests on 1.8.4, due to the long list of warnings
at building time, I found what follows on Debian unstable with arch x86. It would
be nice understanding what those 'hard' conversions are and why they fails.
I'm available to conduct some deep analysis, just in case...

Testing hard unsigned long long -> long double conversions PASSED
Testing hard normalized float -> signed char conversions PASSED
Testing hard normalized double -> signed char conversions PASSED
Testing hard normalized float -> unsigned char conversions PASSED
Testing hard normalized double -> unsigned char conversions PASSED
Testing hard normalized float -> short conversions PASSED
Testing hard normalized double -> short conversions PASSED
Testing hard normalized float -> unsigned short conversions PASSED
Testing hard normalized double -> unsigned short conversions PASSED
Testing hard normalized float -> int conversions PASSED
Testing hard normalized double -> int conversions PASSED
Testing hard normalized float -> unsigned int conversions PASSED
Testing hard normalized double -> unsigned int conversions PASSED
Testing hard normalized float -> long long conversions PASSED
Testing hard normalized double -> long long conversions PASSED
Testing hard float -> unsigned long long conversions -SKIP-
    Test skipped due to hardware conversion error.
Testing hard double -> unsigned long long conversions -SKIP-
    Test skipped due to hardware conversion error.
Testing hard normalized long double -> signed char conversions Command terminated by signal 11
0.31user 0.04system 0:01.64elapsed 21%CPU (0avgtext+0avgdata 19936maxresident)k
824inputs+360outputs (6major+14996minor)pagefaults 0swaps
make[4]: *** [dt_arith.chkexe_] Error 1
make[4]: Leaving directory `/home/frankie/debian/debian-gis/supported/hdf5/hdf5-1.8.4/debian/build-serial/test'
make[3]: *** [build-check-s] Error 2
make[3]: Leaving directory `/home/frankie/debian/debian-gis/supported/hdf5/hdf5-1.8.4/debian/build-serial/test'
make[2]: *** [test] Error 2
make[2]: Leaving directory `/home/frankie/debian/debian-gis/supported/hdf5/hdf5-1.8.4/debian/build-serial/test'
make[1]: *** [check-am] Error 2
make[1]: Leaving directory `/home/frankie/debian/debian-gis/supported/hdf5/hdf5-1.8.4/debian/build-serial/test'
make: *** [check-recursive] Error 1

···

--
Francesco P. Lovergine

2

Sorry, I noted that the debian build-system partially override the -O0 flag and
that probably causes the problem with gcc 4.4...

···

On Tue, Feb 02, 2010 at 11:15:17AM +0100, Francesco P. Lovergine wrote:

Hi

while conducting some extensive tests on 1.8.4, due to the long list of warnings
at building time, I found what follows on Debian unstable with arch x86. It would
be nice understanding what those 'hard' conversions are and why they fails.
I'm available to conduct some deep analysis, just in case...

--
Francesco P. Lovergine

3

A Tuesday 02 February 2010 11:15:17 Francesco P. Lovergine escrigué:

Hi

while conducting some extensive tests on 1.8.4, due to the long list of
warnings at building time, I found what follows on Debian unstable with
arch x86. It would be nice understanding what those 'hard' conversions are
and why they fails. I'm available to conduct some deep analysis, just in
case...

[...]

This is most probably due to the use of the optimization flags in gcc
triggering a bug in the HDF5 library. Until this is fixed, the HDF group
recommends not using optimization at all. For details, see the thread:

http://mail.hdfgroup.org/pipermail/hdf-forum_hdfgroup.org/2009-
October/001995.html

···

--
Francesc Alted

4

Yup, that's the problem. I'm working on fixing it for the 1.8.5 release (~May 2010).

  Quincey

···

On Feb 2, 2010, at 5:20 AM, Francesco P. Lovergine wrote:

On Tue, Feb 02, 2010 at 11:15:17AM +0100, Francesco P. Lovergine wrote:

Hi

while conducting some extensive tests on 1.8.4, due to the long list of warnings
at building time, I found what follows on Debian unstable with arch x86. It would
be nice understanding what those 'hard' conversions are and why they fails.
I'm available to conduct some deep analysis, just in case...

Sorry, I noted that the debian build-system partially override the -O0 flag and
that probably causes the problem with gcc 4.4...