VMD-L Mailing List

From: Hart, David Blaine (dbhart_at_sandia.gov)
Date: Tue Jul 14 2015 - 14:52:02 CDT

It took me a while, but I was able to finally solve this :-) You are right
on the cause - it was not linking with the HDF5 libraries. That was
complicated by the fact that, despite overriding the HDF5 library location
with the 32-bit binaries (to be compatible with VMD), Cmake was not
recognizing that override and was still trying to link with 64-bit libs ¡ª
hence why they weren¡¯t linked in properly. I did also change to build with
32-bit clang, not gcc (CFLAGS=LDFLAGS=-m32). So if anyone needs a Mac
10.10 H5MD-molfile binary, I have one!

Thank you for the help!

David Hart

P.S. making sure .dylib is renamed (or soft linked) to a .so file was also
necessary for the plugin loader to recognize the plugin at all - I assume
this is true for any Mac-built plugins.

On 7/14/15, 1:45 PM, "John Stone" <johns_at_ks.uiuc.edu> wrote:

>Hi,
> I think that the issue with loading the h5md plugin might be
>caused by either an unresovled dependency (from the perspective of
>the h5mdplugin.so shared object at the point when VMD calls dlopen()),
>or by the MacOS X runtime not searching the directory where the
>libraries it depends on are located. To get VMD to be completely
>self-contained, we have to override the normal MacOS X search path
>by using "otool" to set the search path for things like Tcl so that
>they are relative to the VMD install folder. If you have compiled
>h5mdplugin.so against other libraries, MacOS X may or may not do
>the right thing with resolving dependencies if the libraries it needs
>aren't already in your framework or dyld search paths.
>
>Yet another potential problem might be caused by mixing code compiled
>via clang vs. code compiled with GNU gcc/g++. All of the existing
>shipping versions of VMD were compiled using gcc/g++ rather than clang
>since this produces binaries that will run on any later version of
>MacOS X. It is possible that clang and gcc are incompatible in various
>ways and this might explain why h5mdplugin.so doesn't load if you used
>clang to compile it.
>
>Cheers,
> John Stone
> vmd_at_ks.uiuc.edu
>
>
>On Mon, Jul 06, 2015 at 10:26:13PM +0000, Hart, David Blaine wrote:
>> Hi John,
>>
>> On 7/6/15, 3:42 PM, "John Stone" <johns_at_ks.uiuc.edu> wrote:
>>
>> >Hi,
>> > I'm fairly certain that the issue here is that there are unresolved
>> >dependencies in the H5MD when VMD calls dlopen()/dlsym() on the
>> >H5MD plugin shared library file, resulting in dlopen()/dlsym()
>>returning
>> >an error, which VMD then reports as a plugin load failure.
>> >
>> >What output do you get if you run this command?:
>> > otool -L h5mdplugin.so
>>
>> h5mdplugin.so:
>> h5mdplugin.so (compatibility version 0.0.0, current version 0.0.0)
>> libh5md.so (compatibility version 0.0.0, current version 0.0.0)
>> /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version
>> 1213.0.0)
>>
>>
>> >
>> >Perhaps we should include the H5MD plugin authors on this email
>> >exchange...
>>
>> I believe Jonas is the author, and I believe he is on this list;
>>
>>
>> If I understand correctly, you are suggesting that the
>>h5mdplugin/libh5md
>> (both created by the plugin) aren©öt properly linked to the HDF5-1.5
>> libraries?
>>
>> Thanks,
>>
>> David
>>
>
>--
>NIH Center for Macromolecular Modeling and Bioinformatics
>Beckman Institute for Advanced Science and Technology
>University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
>http://www.ks.uiuc.edu/~johns/ Phone: 217-244-3349
>http://www.ks.uiuc.edu/Research/vmd/