I hadn't seen this thread.
There are two variants of each release; one is BSD and it doesn't contain
When I checked out libmesh from github, it appeared to me at least like
the standard model was to build into the source code tree those libraries
it used for I/O purposes. It seemed to be carrying around multiple
versions of netCDF and Exodus (not to mention gmv, tecplot, xdr, tetgen
etc. -- though I admit those are much smaller pieces of code due primarily
to a) ascii and b) serial-only). So, I concluded (without asking I admit)
that handling Silo similarly was the right way to go. That said, Silo can
use HDF5 instead of its native PDB (Protabl Database not Protien Database)
and I did not opt to add HDF5 in the same way netCDF apparently was for
I am happy to remove that and just have Silo treated as an external lib
you can 'point-to' when building libmesh. Though I guess I would like to
see if there is some traction for having libmesh actually down the
download and install of silo if it is requested by a user. If that makes
more sense, I can certain do it that way. And, knowing that probably
would've made my life a lot easier because I think I spent a lot more time
integrating Silo with libmesh' build system than I did on the actualy I/O
Finally, I am not "done" with the I/O routines; just at a point where it
made sense to try to push something useful back to the main repository; it
can write serial ucd meshes with nodal data.
My intention is to add ability to write in parallel too.
On 11/17/13 10:56 AM, "John Peterson" <firstname.lastname@example.org> wrote:
>On Sun, Nov 17, 2013 at 9:48 AM, Derek Gaston <email@example.com> wrote:
>> Just a heads up that I'm helping Mark Miller at LLNL get SILO support
>> I just cleaned up his patches and created a branch that he's going to
>>branch off of in the future here:
>> I would really appreciate it if you guys could look at what he's done
>>and give us some comments...
>What is the license?
>Wikipedia (http://en.wikipedia.org/wiki/Silo_(software)) takes me
>here: http://www.nevercenter.com/ which says you can buy it for
>Second point is that it appears to be a massive piece of software.
>The first patch says:
>"Showing 340 changed files with 259,038 additions and 9,219 deletions."
>so I'm pretty hesitant to distribute this with libmesh at all.
>Netcdf/Exodus is a pain to deal with, but it's worth it to distribute
>it because it provides a key I/O capability. SILO could become that
>some day, but in the meantime I don't think we want to saddle every
>libmesh user with needing to download/build yet another library they
>may not ever use.