[Rivet] GENSER meeting minutes ...

Andy Buckley andy.buckley at durham.ac.uk
Thu Feb 7 16:57:10 GMT 2008


James Monk wrote:
> 
> On 7 Feb 2008, at 11:56, Andy Buckley wrote:
> 
>>> Apart of that (and James might correct me if I'm wrong) it didn't look
>>> like there would already be some active effort.
> 
> They did seem to say that they would patch the existing configure
> scripts to address some of our issues in the short term (e.g. bash
> instead of sh, dylibs on OSX)

Okay, that's good. Add the $(FC) -> $(CXX) problem to that list (I just
today managed to make a flimsy work-around the problems that causes) and
we're happy for the short-term.

> but the migration to autotools might be a
> while.  Their biggest problem with autotools is relocation of libs
> during grid installation, which I don't understand since surely you can
> use autotools to install on a grid node too?

I assume they mean deployment on to grid nodes without re-building on
each one. The issues associated with that aren't autotools-specific:
they're related to the architecture of ELF libraries, which contain path
references to their dependencies.

Anyway, the evidence of several million Linux installations with binary
package systems shows that this is entirely compatible with autotools.
But note that Linux system packaging systems use packages which are only
designed to install into the same location as the autotools installation
during the package build. If Genser want to be able to produce packages
which can be arbitrarily re-located, then there are more problems in
store. There may be lessons to be learned from autopackage, which is the
only packaging system I know of that's designed to make all packages
relocatable.

Anyway, it's a solvable problem, if they're clear about what they want
to achieve.

Andy


More information about the Rivet mailing list