Re: HDF5 library mismatch (switching to the maintainers list)

[Michael D Godfrey]

On 02/11/2011 07:00 AM, Przemek Klosowski wrote:

On 02/10/2011 06:00 PM, Jordi Gutiérrez Hermoso wrote:

On 10 February 2011 16:01, Przemek Klosowski<address@hidden>  wrote:

because Octave was compiled against HDF5 v.1.8.4 and in the meantime HDF was
updated because the dependency in the package is specificed loosely. The
abort is actually a preventive measure of the incompatibility testing code
in Octave; it's possible to disable the warning by controlling Octave with
an environment variable mentioned in the warning:

Weren't we discussing in the maintainers mailing list how the Octave
rpm packages need some love? Can you do any of that? Do you need help
with it?

There are two issues here: packaging,  and the actual technical issue with HDF. As to the latter, which I believe is relevant to all distributions of Octave, the code implements a version check which on the surface appears to be overly strict: essentially it warns of the possibility of a crash and then proceeds to abort the program itself. There is an environment variable that avoids that abort, but it has to be set before starting Octave, so it's an usability issue.

I guess the thinking is that it's better to abort the program than to possibly return incorrect data. If people agree that this is the correct thing to do, I think we should change packaging so that Octave 'requires' the version of HDF5 that it was built against. I think Fedora packaging system would trigger a rebuild of Octave in this case. Jussi Lehtola suggested a packaging fix in my bugzilla entry for this problem, https://bugzilla.redhat.com/show_bug.cgi?id=676707

Alternatively, from the usability point of view, I belive it would make sense to NOT abort octave, but issue the warning and either return nothing from the 'load' (overridable by a runtime configuration option), or proceed anyway (if we crash, we crash, and if the data is bad, well, there was a warning :).


About the general packaging issue, Octave RPM packages are managed by Jussi Lehtola, Alex Lan(?), mmahut (?) @Redhat and Rakesh Pandit. I try to help by submitting bugs: in bugzilla when it looks like a packaging issue, and to Octave's tracker otherwise. I wouldn't mind applying to become a co-packager for Octave, if people here felt that it would help.
Could you point me to the list of RPM packaging issues that need attention?

It would be wonderful if you could help with rpm building.  The RedHat and Fedora rpm building and updating has not
kept up to date.  RedHat 5.6 is on 3.0.5 and Fedora 14 is on 3.2.4 (close to up-to-date).   So, it appears that they
could use some help.  I suspect that they try to keep up, but that they also have lots of other things to do.  I do
not think that there are major technical problems.  Octave 3.4 builds cleanly and passes all the make check
tests without any special tweaks. (Just "./autogen.sh ; ./configure ; make" builds correctly)  All the dependencies
are resolved by packages in the Fedora repositories.
(To be precise, I should recheck that, but I am pretty sure.)  I routinely use 3.4 (or the current development system)
on Fedora 14 and plan to build it for friends who use RHEL 5.6 as soon as I have time.

Michael