On Sat, Aug 30, 2008 at 08:17:34PM -0400, walker@xxxxxxxxxxxx wrote:
On Sat, Aug 30, 2008 at 09:43:52AM -0400, walker@xxxxxxxxxxxx wrote:
No, it doesn't exist. I imagine we could have something like either
It doesn't exist! man!
<dofmap>
<dofs size="...">
<cell index="...">
<dof local="0" global="..."/>
<dof local="1" global="..."/>
...
</cell>
...
</dofs>
</dofmap>
or
<dofmap>
<signature="FFC dof map for ..."/>
</dofmap>
depending on whether we store the entire dofmap or the signature.
I just did the signature so far.
ok, nice.
noticed that the routines that are there now for creating a Function
with
a given dofmap, require a Form input as well. I am a little confused on
how this would be done here. We are not really defining a variational
form for higher order mesh geometry. If there are any demos/files I
should look at, please tell me.
It's also possible to create a (Discrete)Function from a finite
element signature and dofmap signature. See lines 167-171 in
XMLFile.cpp.
Actually, this is what I did already. I thought this would be easier.
But you commented before about having the explicit dofmap. Do you really
think having the explicit dofmap is necessary? We could always have a
conversion routine later for past meshes. Or better yet, have a more
standard format, then have a conversion script to turn it into an FFC
compliant thing.
I think it's necessary for two reasons: (i) DOLFIN should not depend
explicitly on FFC, one should be able to use it with other form
compilers, and (ii) the mesh numbering could change if someone decides
to add some optimization in TopologyComputation.cpp.
We might ignore the first reason since we wouldn't really depend on
FFC, just deciding to use the same numbering as FFC uses in this
particular case.
Anyway, we can start out with just having the dofmap signature and
then add the explicit dofs later.
I hope you can find your way around the code. I'm very happy you are
working on this. I'll have very little time in to work on it myself
the coming weeks but I'll try to help out as much as I can.
Yeah, I don't have much time either. The semester starts soon, so
progress will be slow. I'm still a little paranoid about modifying the
code, but it's seeming less bad now.
Just go ahead an edit. It's just a C++ code and it's there to be
modified.
Does anyone have a suggestion for a good code editing platform? I have
nice editor that does highlighting and stuff, but is there anything
extraordinary out there? In other words, what do FENICS people use for
modifying?
I just use emacs + xterm, typically on different desktops (I usually keep
around 6x3 desktops).
--
Anders
PS: Use emacs with -fn 10x20 for nicer fonts.
