Do not set domains of all active coefficients in domain_integral_type

[pbrubeck]

Description

Revert #4775

Validates the domains in a multidomain Form at compile time and throws a MismatchingDomainError if the domains are topologically unrelated or if the domains were not included in the integral Measure.

[connorjward]

Note that some tests are still failing (see the complex run).

[pbrubeck]

Note that some tests are still failing (see the complex run).

There are some conflicts with #4763.

It seems tricky to resolve, mainly because we sometimes allow coefficients for which domain_integral_type_map[coefficient_domain] = None

[connorjward]

Note that some tests are still failing (see the complex run).

There are some conflicts with #4763.

It seems tricky to resolve, mainly because we sometimes allow coefficients for which domain_integral_type_map[coefficient_domain] = None

I looked into this a little with @leo-collins. Setting things to None here I think is just a hack to avoid key errors later on.

Note that some tests are still failing (see the complex run).

There are some conflicts with #4763.

It seems tricky to resolve, mainly because we sometimes allow coefficients for which domain_integral_type_map[coefficient_domain] = None

Is it sufficient to just remove this line:

    domain_integral_type_map.update(dict.fromkeys(coefficient_meshes, "cell"))

This is the one that I think is properly wrong.

[pbrubeck]

Is it sufficient to just remove this line:

    domain_integral_type_map.update(dict.fromkeys(coefficient_meshes, "cell"))

This is the one that I think is properly wrong.

That's what the PR is doing.

[connorjward]

Is it sufficient to just remove this line:

    domain_integral_type_map.update(dict.fromkeys(coefficient_meshes, "cell"))

This is the one that I think is properly wrong.

That's what the PR is doing.

Ah OK. I didn't quite follow the logic.

[pbrubeck]

I changed ValueError to NotImplementedError and complex tests are now passing. Real CI is giving a timeout, but this does not seem to be a problem introduced here.

[ksagiyam]

#4775 basically allowed users to use illegal forms. If reverting #4775 causes conflicts with #4763, it indicates that #4763 relied on #4775 and used illegal forms in the tests; please see my comments in #4763. You don't want to manipulate types of exception just to make those tests pass. NotImplementedError is not the right type of error.

I would revert both #4763 and #4775, restore the original code, and restart from there.

You should ask @dham to have a look at these PRs.

[dham]

Custom exception type here.