Re: Metis

[John W. Eaton]

On 16-Oct-2007, David Bateman wrote:

| John W. Eaton wrote:
| > Is it possible to modify the Octave configure script to not use Metis
| > by default?  If so, and we modify configure so that it doesn't look
| > for Metis, then I assume the checks for SuiteSparse will fail if it
| > has been built in a way that requires Metis, correct?  In that case,
| > could the user specify libraries in a way that would allow building to
| > continue, but we would not be encouraging people to use Metis?
| 
| Not really, as Octave doesn't in fact link to Metis.. It links to
| Cholmod that then links to Metis. If the user built Cholmod to use
| metis, then our only choices are to link to metis or to not link to
| cholmod at all.
| 
| >From my point of view, its perfectly alright to link to Metis, its the
| distribution of the resulting binary that is the issue. This is a
| similar situation to GOTO blas, we should prevent it being used, but
| should enforce the fact that the binary resulting from linking to it
| shouldn't be distributed..

I think there is a slight difference with the BLAS.  In that case,
there are multiple implementations and some have licenses that are
compatible with the GPL.  For Metis, I don't know of any
GPL-compatible replacment.

In any case, I'm not saying prevent linking entirely, but make it so
that Octave's configure script looks for cholmod without metis only.
If people want to specify something else explicitly when configuring I
can't prevent that, but I think the default should be to link with
ufsparse without Metis.

Did we ever request a change in the Metis license to make it GPL
compatible?

jwe