[Debichem-devel] Abinit (Was: r5051 - unstable/abinit/debian)
Michael Banck
mbanck at debian.org
Sun Apr 20 10:11:47 UTC 2014
Hi,
On Sun, Apr 20, 2014 at 08:22:48AM +0200, Andreas Tille wrote:
> On Sat, Apr 19, 2014 at 07:28:36PM +0200, Michael Banck wrote:
> >
> > So I propose the following:
> >
> > 1. We install the pseudopotentials from tests/Psps_for_tests to
> > /usr/share/abinit/psp
By the way, I think there are two TODOs here:
1. Teach Abinit to read in gzip'd psp data. Some of the
pseudopotentials are rather large, as they are generated numerically,
not analytically.
2. Modify Abinit to look in /usr/share/abinit/psp when a pseudopotential
in the input is not found and/or no specific path is specified.
> > 2. We install all files from tests/tuto*/Input/ into
> > /usr/share/doc/abinit/examples.
>
> > 3. We patch/modify the respective .files files to use
> > /usr/share/abinit/psp as prefix, not ../../Psps_for_tests or so. This
> > will make it possible to relocate the .in files (to some degree) and to
> > run them (I hope).
I did the above now.
> > 5. We either don't ship the testsuite inputs/outputs for now, or at
> > least gzip all the reference outputs, and make the testsuite setup deal
> > with it. I suggest we do the latter as TODO, but don't delay an upload
> > due to it.
>
> In the Debian Med team we try to run anything that could serve as a test
> suite in autopkgtest. So I'm definitely for including the data and I
> would consider providing a test would be a good thing to do. However,
> if it might turn out as complex this should be no reason for a further
> delay.
I think that's useful, and should be done in general. It's just that
right now the size of the reference output dwarves the rest of the
package, so I think it only makes sense to include those when (i) we
actually have autopkgtest in place (I need to look at that in general)
and (ii) we can gzip them (they should compress quite well).
> > As for the packages, I propose we introduce an abinit-data package which
> > includes the pseudopotentials (and maybe later the testsuite
> > input/output). I think at least mid-term (jessie+1 or so) it makes
> > sense to move the contents of -doc over to -data and deprecate the -doc
> > package. Or do it already now, or never, I don't care either way.
>
> I usually prefer modularised packaging and see no real profit in merging
> the *-doc into the *.data package. But I'm not very strongly opinionated
> here if you would like to drop the -doc package.
Let's keep the -doc package then for now. But I think a -data package
at least for the pseudopotentials and (later on) the testsuite
input/output makes sense.
Thanks for the feedback,
Michael
More information about the Debichem-devel
mailing list