Bugs and tests

Axel akoh... at gmail.com
Sun Mar 23 00:17:01 UTC 2008


hi carlo and teo,

[...]

> > a few days ago I compiled cp2k and runned an input for AlN bulk
> > everything was ok (input.orig)
>
> > I recompiled the code today and numbers were wrong (see out1)

these things can happen... this is the risk of using a code
that is under development.

> > I tried it and it worked so that the output
> > (out_cured) was as ok as the original one.
>
> > I strongly suggest that you reinforce the test that have to be
> > passed before modifications are submitted to the CVS.

> That's true.. unfortunately it is very difficult to cover all
> possibilities.. At the moment we'have something like ~ 900 regtests..
> and all
> of them are tested after every modification done.. as I said to cover
> all the potentialities of cp2k not even the double of regtests would
> be enough..

not to mention the time it costs to run those tests...
if you use the up-to-date CVS version, you have to be
aware that some oversights can always happen. after all,
do you want developers to spend their time developing code
or running tests? in the CVS trees of other packages you
can run across the case that the code does not even compile.
in that respect, cp2k does quite well and in my experience
over the last couple of years most developers are quite careful.
also, particularly for people new to this style of software
developement (i.e. have a lot of people follow your changes
on a daily basis instead of giving them a copy every once
in a while) it takes a while to get used to how to avoid
mistakes and program "defensively".


> The most important thing would be to have a release... This would
> avoid these kind of high risk activities, separating
> a production release from a development release..
>
> Axel, did you hear it? ;-) looks like *we* have to push the release
> stuff ;-)


yes. i hear it. right now i'm stuck with finalizing the release
of a different package (little problems keep creeping up... as usual),
teaching, and setting up the ACS QM/MM symposium and the CMM workshop
in philly (ACS is mostly done. just submitted a tentative
schedule...).

[...]

> > it is not a problem if cards evolve (and usually it is a good thing)
> > but it is dangerous that an input that was producing a correct result
> > tunrs out in wrong results.
>
> Ooooooooooohhhhhhhhhhh.. finally!!!
> Finally someone that does not complain that keywords change ;-) I was
> sure that clever people were appreciating the evolution ;-)
>
> As I said for this specific problem Matt found the bug fix (give me 1
> hour or so..)
> This is another important reason why I always ask people to update
> the code frequently (at least as soon as we won't have the release)..
> This helps identifying bugs introduced and not catched from the
> regtests.

yes, and it also helps to _improve_ the regtest library.
and even more so, in order to help sorting out potential
problems with an upcoming release version, it would be
particularly helpful if people here would update as often
as possible over the next couple of months and tests as many
features as possible (and report problems), so that by the
time the release is branched off, we can start from a solid
and reliable version (at least as far as commonly used features
are concerned).

thanks and cheers,
    axel.

>
> Ciao and thanks again for the report!
> Teo
>
>
>
> > Please find enclosed the file aln.tar
>
> > Ciao
>
> > Carlo
>
> > <aln.tar>


More information about the CP2K-user mailing list