Feat(sqlmesh_dbt): Select based on dbt name, not sqlmesh name

[erindru]

This builds on the dbt node info exposed in #5412 to enable its usage in the selector engine.

Previously, to use sqlmesh_dbt --select, you had to use the sqlmesh-native model names.

This PR:

• Adds dbt_mode flag to the Context to pass to the Selector engine
• Sets this flag in the sqlmesh_dbt CLI when constructing the Context
• Uses it to indicate that dbt fqn's should be matched against instead of the sqlmesh model name

The selectors still resolve to SQLMesh native fqn's so everything continues to match up downstream, they just match based on dbt fqn instead of sqlmesh model name

[erindru]

I called this generically dbt_mode instead of something more specific to selectors incase we discovered other behaviour that we wanted to attach to it

[themisvaltinos]

I wonder if we need the extra flag instead of using the project_type, for when self._project_type == c.DBT

[themisvaltinos]

actually never mind I realise this is for when someone runs sqlmesh_dbt rather than a dbt project with the native sqlmesh commands so this is indeed needed

[izeigerman]

I don't like this flag at all. Can we instead have a base Selector class and 2 implementations and then we can configure the implementation used when creating a Context instance.

[erindru]

Sure, i've refactored Selector to be an abstract base class and then implemented it as NativeSelector and DbtSelector

[themisvaltinos]

I see the wildcard matching happens outside _pattern_to_model_fqns which seems straightforward and should work, but could you add a test for this as well to ensure nothing breaks in the future if someone modifies this? unless if I missed it and there is a wildcard test for dbt model names already

[erindru]

Well spotted, I added some tests to exercise this codepath based on the dbt docs.

Note that the patterns currently need a * in them to trigger this branch (even though ?, [ and ] could be used as well) so this will need more work in future if we find people using these in the wild

[themisvaltinos]

lrgtm! I was wondering of ways for the _pattern_to_model_fqns logic to be reduced to a regex, but it will look complicated and I do like your approach a lot better, it is more readable and makes it clear what is happening

[izeigerman]

Thanks!

[github-christophe-oudar]

Just noticed it implemented #3583 which was quite painful! Thanks 🙌