compile_subdomains does not complain to non-existent variable

[Jan Blechta]

This code

domain = compile_subdomains("x[0]<L")

passes without error. This is quite dangerous as user is not warned that L is not initialized by possible variable L from python context.

[Anders Logg (Chalmers)]

This came up in a question posted on StackExchange: http://scicomp.stackexchange.com/questions/7232/poiseuille-flow.

[Prof Garth Wells]

This might be a left-over from the past, kept for backwards compatibility. It isn't pretty.

[Johan Hake]

Yes this is a left over. A lot of work has gone in to make compiled expressions robust but not much love has gone into compiled subdomains.

Johan

[Johan Hake]

I suggest removing the "feature" of compiling several subdomains in one go, which is used in demo/pde/hyperelasticity/python:

left, right = compile_subdomains(["(std::abs(x[0])       < DOLFIN_EPS) && on_boundary",
                                  "(std::abs(x[0] - 1.0) < DOLFIN_EPS) && on_boundary"])

so we can more easily check for given parameters. This might break some user code, but I think it is worth it. Any comments?

[Prof Garth Wells]

Removing the 'feature' sounds sensible to me.

[Anders Logg (Chalmers)]

Yes, and then rename the function. Keep the old with an error message.

[Johan Hake]

To compile_subdomain or maybe CompiledSubdomain?

[Anders Logg (Chalmers)]

Hmm... Both are attractive, but CompiledSubDomain (note capital 'D') seems to be the most consistent with Expression. Or if you want to be really consistent and want to have some fun with more metaclasses, then it should be just SubDomain... ;-)

[Johan Hake]

Ok, CompiledSubDomain it is. There was a reason I did not mention SubDomain...

[Johan Hake]

Change naming of compile_subdomains to CompiledSubDomain. -- Make CompiledSubDomain more robust, by enforcing setting default arguments (fixing issue 32)

→ <<cset bae2f7038d59>>