Tests in types_graph_building_test.py

[turbo]

Hi there,

I'm reading through the various type inference modules to get a better understanding. I'm having trouble running the tests in types_graph_building_test.py and wanted to know if that's a bug on my side.

For example, test_when_connection_with_other_predicates fails because the expected case lists 3 Equalitys, but the edges result holds:

• those 3 Equalitys, and
• 2 PredicateArguments

Other tests in the same module fail for the same reason. All tests in type_inference_test.py on the other hand, succeed.

See:

[turbo]

This is the change (line 122) that broke these tests. The new propagation wasn't considered in the tests: 7a0ea40#diff-371895e22cf3ce93a5cf4d22a70c5171007a02e68679a267bf4436cd724e0a59

[turbo]

Actually, this test has another issue. The vertex order on all Equality edges Equality edges with a matching predicate addressing edge are the wrong way around. This isn't detected because the tests only test the edge set cardinality.

So for the single test mentioned above, the real edge set is actually:

PRED_ARG Num.logica_value -> Num.col0
PRED_ARG T.logica_value -> T.col0

EQUALITY Q.col0 -> x
EQUALITY T.col0 -> x
EQUALITY Num.col0 -> x

[EvgSkv]

Oh thanks for brining it up. It was quite an adventure to get type inference working, we had a couple back and forths and we should clean up the codebase.

All relevant tests are executed from run_all_tests.py. Type inference tests are here: https://github.com/EvgSkv/logica/blob/main/type_inference/research/integration_tests/run_tests.py

Tests that you point to seem to be testing dead code. I should delete it.
Sorry for this confusion caused!

Please let me know if you have further questions!

[turbo]

Ah, thanks!

Question something in reference_algebra:

This fails at runtime, but I think is actually supposed to return false:

IsFullyDefined([TypeReference('Any')])

It fails because IsFullyDefined gets confused by a dangling ID suffix.

[EvgSkv]

Surely we need more comments in type inference code :-)

The failure is actually expected there, as IsFullyDefined is allowed only to be called on "very concrete" types.

"Very concrete" in this context means that it involves no type references.

So the call should be done as

IsFullyDefined(VeryConcreteType([TypeReference('Any')]))

Example colab: https://colab.research.google.com/drive/1dD23lpksDPBc5UenT8HKqJEpv0v4U8PW?usp=sharing

Let me know if you have further questions!

[turbo]

Aha! Yes, that makes sense, thanks for the detailed answer.

BTW: Are discussions a good way to ask these questions? I'm working on a native compiler for Logica, so I may have a few additional stupid questions along the way 😄 (I'm fully done with the parser, about 70% through the type inference, and about 10% done with the SQL generator).

[EvgSkv]

I'm working on a native compiler for Logica

Oh, that's very interesting! I was thinking of implementing a C++ version, but didn't get to this.

Asking questions here is great, as someone may have the same question later :-) But in case you have very narrow questions, you are welcome to ask them by email - I reached out to email that you have in your profile.