If you have time to test this release candidate, we would greatly appreciate it. We try our best to test HDF5 on a variety of platforms and with multiple compilers, but there is always a system that we couldn’t test on. Your feedback is critical.
For those who use autoconf to build HDF5, please notice that the default configuration mode for this distribution is “development". Use the "—enable-production" configure flag if you want to build HDF5 in the production mode. The default will be changed to “production" in the next release candidate and the release itself.
Thank you and happy testing and reporting!
Elena
···
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1) For a Maintenance only why a ABI bump ?
HDF5 is the only package, I am aware of, that in averages
bump the ABI every 2 releases. This time two bumps in the row.
2) I wasted some time to discover this change
## Disable shared libraries on CYGWIN. (LK - 04/16/15)
## A number of tests run by "make check" fail on CYGWIN, so for HDF5 v1.8.15
## we will change the default for shared libraries to disabled.
...
This seems, in my humble opinion, a questionable approach.
I am running tests with shared libraries re-enabled and
I will look on these new failures.
Thank you and happy testing and reporting!
Elena
Marco Atzeri
(cygwin package maintainer for HDF5 )
The 1.8.15 release candidate builds and passes all tests (deflate test is
skipped on my setup) on Debian 8 RC2 gcc (Debian 4.9.2-10) 4.9.2
OS identifies from uname -a as
3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt7-1 (2015-03-01) x86_64 GNU/Linux
Best regards,
Karl Hoover
Senior Software Engineer
Geometrics
Could someone clarify why "shared libraries" where
disabled for cygwin in 1.8.15pre3 ?
Regards
Marco
···
On 4/21/2015 3:08 PM, Marco Atzeri wrote:
On 4/21/2015 12:05 AM, Elena Pourmal wrote:
2) I wasted some time to discover this change
## Disable shared libraries on CYGWIN. (LK - 04/16/15)
## A number of tests run by "make check" fail on CYGWIN, so for HDF5
v1.8.15
## we will change the default for shared libraries to disabled.
...
This seems, in my humble opinion, a questionable approach.
I am running tests with shared libraries re-enabled and
I will look on these new failures.
1) For a Maintenance only why a ABI bump ?
HDF5 is the only package, I am aware of, that in averages
bump the ABI every 2 releases. This time two bumps in the row.
There were changes in the C public headers and symbols were added/removed for C++.
## Disable shared libraries on CYGWIN. (LK - 04/16/15)
## A number of tests run by "make check" fail on CYGWIN, so for HDF5 v1.8.15
## we will change the default for shared libraries to disabled.
...
This seems, in my humble opinion, a questionable approach.
I am running tests with shared libraries re-enabled and
I will look on these new failures.
dt_arith test fails with the shared libraries. If you need to build shared, you should be able to use the —enable-shared configure option.
Thank you!
Elena
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Thank you and happy testing and reporting!
Elena
Marco Atzeri
(cygwin package maintainer for HDF5 )
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The 1.8.15 release candidate builds and passes all tests (deflate test is
skipped on my setup) on Debian 8 RC2 gcc (Debian 4.9.2-10) 4.9.2
OS identifies from uname -a as
3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt7-1 (2015-03-01) x86_64 GNU/Linux
Best regards,
Karl Hoover
Senior Software Engineer
Geometrics
Thank you for the report! And yes, please, try Intel and PGI compilers if you have time.
Elena
···
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
That test fails in general. As I reported years ago, it performs several undefined behaviours, so you should expect failure. I just ran it again with UBSan and:
13: Test command: /Users/builder/external/hdf5_1_8_cmake-clang-dbg-x86_64-static/bin/dt_arith
13: Test timeout computed to be: 1200
13: Testing non-aligned conversions (ALIGNMENT=1)....
13: Testing query functions of compiler conversion PASSED
13: Testing user-define and query functions of floating-point types PASSED
13: Testing user-define and query functions of integer types PASSED
13: (/Users/builder/external/hdf5_1_8_cmake-clang-dbg-x86_64-static/bin/dt_arith+0x1000065fe): runtime error: value -8.28403e+90 is outside the range of representable values of type 'float'
13: Testing noop float -> float conversions PASSED
13: Testing noop double -> double conversions PASSED
13: Testing noop long double -> long double conversions PASSED
13: Testing hard normalized float -> double conversions PASSED
13: (/Users/builder/external/hdf5_1_8_cmake-clang-dbg-x86_64-static/bin/dt_arith+0x100020f05): runtime error: value 2.22507e+40 is outside the range of representable values of type 'float'
13: Testing hard normalized double -> float conversions Testing hard normalized float -> long double conversions PASSED
13: Testing hard normalized double -> long double conversions PASSED
13: (/Users/builder/external/hdf5_1_8_cmake-clang-dbg-x86_64-static/bin/dt_arith+0x1000212a9): runtime error: value 3.3621e+44 is outside the range of representable values of type 'float'
13: (/Users/builder/external/hdf5_1_8_cmake-clang-dbg-x86_64-static/bin/dt_arith+0x100021616): runtime error: value 3.3621e+308 is outside the range of representable values of type 'double'
13: Testing hard normalized long double -> float conversions Testing hard normalized long double -> double conversions Testing hard denormalized float -> double conversions PASSED
13: Testing hard denormalized double -> float conversions PASSED
13: Testing hard denormalized float -> long double conversions PASSED
13: Testing hard denormalized double -> long double conversions PASSED
13: Testing hard denormalized long double -> float conversions PASSED
13: Testing hard denormalized long double -> double conversions PASSED
13: Testing hard special float -> double conversions PASSED
13: Testing hard special double -> float conversions PASSED
13: Testing hard special float -> long double conversions PASSED
13: Testing hard special double -> long double conversions PASSED
13: Testing hard special long double -> float conversions PASSED
13: Testing hard special long double -> double conversions PASSED
13: /Users/builder/external/hdf5_1_8_cmake/test/dt_arith.c:1786:9: runtime error: left shift of negative value -1
13: Testing hard signed char -> unsigned char conversions
1/1 Test #13: H5TEST-dt_arith ..................***Failed 1.11 sec
Cheers,
···
On Wed, 22 Apr 2015 12:01:42 +0000, Elena Pourmal said:
dt_arith test fails with the shared libraries
--
____________________________________________________________
Sean McBride, B. Eng sean@rogue-research.com
Rogue Research www.rogue-research.com
Mac Software Developer Montréal, Québec, Canada
## Disable shared libraries on CYGWIN. (LK - 04/16/15)
## A number of tests run by "make check" fail on CYGWIN, so for HDF5
v1.8.15
## we will change the default for shared libraries to disabled.
...
This seems, in my humble opinion, a questionable approach.
I am running tests with shared libraries re-enabled and
I will look on these new failures.
dt_arith test fails with the shared libraries. If you need to build
shared, you should be able to use the �enable-shared configure option.
-------------------------------------------------------
Testing non-aligned conversions (ALIGNMENT=1)....
Testing query functions of compiler conversion *FAILED*
Can't query conversion function
-------------------------------------------------------
Intel 15.0.2 (icc, ifort and icpc) using the same configure line pass all tests.
PGI (pgcc, pgfortran and pgc++) once again using the same configure line compiles fine, however fails on the "Subtest: File Creation I/O" test. Note, this is the same test that 1.8.14 fails on with PGI too. I'm attaching the testhdf5.chklog.
yes, we confirmed this too. Will fix in the next release candidate.
Thank you!
Elena
···
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
On 4/22/2015 2:01 PM, Elena Pourmal wrote:
Hi Marco,
2) I wasted some time to discover this change
## Disable shared libraries on CYGWIN. (LK - 04/16/15)
## A number of tests run by "make check" fail on CYGWIN, so for HDF5
v1.8.15
## we will change the default for shared libraries to disabled.
...
This seems, in my humble opinion, a questionable approach.
I am running tests with shared libraries re-enabled and
I will look on these new failures.
dt_arith test fails with the shared libraries. If you need to build
shared, you should be able to use the —enable-shared configure option.
Elena,
Please look at the code.
It was "—enable-shared" until 1.8.14 (with --enable-unsupported)
It was removed in the 1.8.15-pre3.
Attached patch I used to re-instate the 1.8.14 behavior.
About dt_arith:
I see only 1 failure
-------------------------------------------------------
Testing non-aligned conversions (ALIGNMENT=1)....
Testing query functions of compiler conversion *FAILED*
Can't query conversion function
-------------------------------------------------------
We didn’t have a chance to test with the new PGI compilers. We will definitely look into the problem, but I am afraid that the fix (if any) will not make into the release.
Is C++ the only test that fails? Did you run make check with -i to run through all tests?
Thank you!
Elena
···
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Intel 15.0.2 (icc, ifort and icpc) using the same configure line pass all tests.
PGI (pgcc, pgfortran and pgc++) once again using the same configure line compiles fine, however fails on the "Subtest: File Creation I/O" test. Note, this is the same test that 1.8.14 fails on with PGI too. I'm attaching the testhdf5.chklog.
Thank you for the report! And yes, please, try Intel and PGI compilers if you have time.
Elena
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
