Re: [moose-generic] hdf5 version, etc.

[Subhasis R. <ray...@gm...>]

Hi Alex,
 Thank you for reporting the issues. I think Niraj' response did not go to
the mailing list. So I'll put it on record and add my input regarding the
bugs.

On Tue, Nov 20, 2012 at 10:47 PM, Alexander Kozlov <ak...@na...>wrote:

> 1) Moose in the Debian package 'moose_2.0.0_i386_python2.7.deb'
> is built against libhdf5.so.6 (not libhdf5.so) and can't work
> in the current Debian 'sid' environment which installs
> libhdf5.so.7
>
> 2) Link to release notes on NCBS server doesn't work
> http://moose.sourceforge.net/downloads/RELEASE_NOTES_MOOSE_2.0.0.html
>

This has been fixed now. The correct link should have been:
http://moose.sourceforge.net/downloads/RELEASE_NOTES_MOOSE_2.0.0.txt


> 3) help(moose) crashes with 'Segmentation fault' message
>     help(moose.connect) and moose.doc(moose.Compartment) work fine
>

This is a bug. Till now I found that it appears only in builds with full
optimization (gcc -O3), e.g., the release build but does not occur in debug
builds (or with release build with "-O3" option removed from the Makefile).
This may be due to some subtle memory leak in the python/C extension. If we
reduce the number of exposed MOOSE classes, then the problem does not
appear. gdb tells us that there is some issue with Python trying to get
some string length.

The only solution I could find till now makes it impossible to use "from
moose import *" and use the MOOSE classes. This is very common practice in
spite of being frowned upon by the Python style guides. So we need more
work to track it down.


> This is after creating link libhdf5.so.6->libhdf5.so; same happens
> on another machine which runs source-build moose and doesn't need
> that workaround.
>

Is that a release or a debug build?

Best,
  Subha