dolfin team mailing list archive

[Matthew Knepley <knepley@xxxxxxxxxxx>]

Faheem Mitha <faheem@xxxxxxxxxxxxx> writes:

> Of course, the source needs to be available etc. That is hardly the issue.

  I think it is exactly the issue.

> However, I'm talking about compiling library B against library A. In this case, the standard Unix model (nothing to do
> with Debian or Gentoo) is that B should not need anything from A except the library files themselves (static and/or
> dynamic) and the header files. Again, this has nothing to do with Debian or Gentoo. If you don't believe me, please ask
> on any Debian or Gentoo development list, or anywhere else, for that matter.

  That is one model. I don't like it.

> The only relevant difference between Gentoo and Debian is that Gentoo generally sets things up so that people can easily
> do recursive builds
>> From source, and customize the source as necessary. In Debian it is not so
> easy, but can be done. This has nothing to do with the matter under discussion, which is whether one should compile
> library B against the source of library A.
>
> For debugging, one usually compiles the library with debugging symbols.

  That is not the point. Try debugging with only symbols and no source.

> Please carefully consider the standard models before discarding them, since that way lies chaos. If everyone does
> something different, then putting together an operating system becomes an impossibility.

  I see it differently. Following a standard merely because it exists is foolish consistency.
Without critical examination of popular models, we would never make any progress. PETSc is
a research project and should thus be free to challenge the existing orthodoxy.

> Please also bear in mind space issues, and the fact that (generally) inclusion in a OS like Gentoo or Debian ensures
> wider distribution.

  The price per MB has plummetted exponentially. I think space issues are the last thing anyone
should think about. Usability is a much more cost-effective consideration, and also ensures a
wide distribution. I don't mind having Debian or Gentoo packages, but I do not think I'd be willing
to break our own model to get it.

  Some more generic criticism of the no-source model. This does not deal well with scripting
languages like Python, which we deal with all the time. Furthermore, we are trying to remove the
artifical distinction between compile and run so that runtime code generation can be used to
solve some presistent scientific computing bottlenecks. I believe our insistence on distributing
source will be born out in the long run.

         Matt
-- 
"Failure has a thousand explanations. Success doesn't need one" -- Sir Alec Guiness