instant team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Jan 02, 2008 at 11:32:22AM +0100, Kent-Andre Mardal wrote:
> ons, 02.01.2008 kl. 00.53 +0100, skrev Anders Logg:
> > On Tue, Jan 01, 2008 at 11:04:31PM +0100, Anders Logg wrote:
> > > It seems Instant does not ignore the name of the module when
> > > looking for previously built modules in the cache.
> > > 
> > > Here's a simple example:
> > > 
> > >   foo = create_extension(wrap_headers=["foo.h"], module="foo", ...)
> > >   bar = create_extension(wrap_headers=["foo.h"], module="bar", ...)
> > > 
> > > Here, I want Instant to realize that the code for both modules is the
> > > same (the file foo.h) so it should only need to build the module once.
> > > 
> > > This does not seem to be the case. Can it be fixed?
> > > 
> > > This is important when running time-dependent problems where the form
> > > (and the code) remains the same but the system needs to be assembled
> > > in each time step.
> > 
> > Never mind. The FFC JIT compiler now uses it's own cache, since we can
> > then use the md5sum of the signature of the form (not the generated
> > code) which is faster (the code does not need to be generated).
> > 
> 
> ok, but is the signature of the form unique ? 
> 
> Kent

I thought so. But now when I think about it, it's not enough to just
look at the signature of the forms since the spaces may be different.
So I need to include the signatures of the spaces when computing the
checksum. That should fix it.

-- 
Anders