fix: Raise AttributeError for non-existing meshing objects after switch_to solver.

[prmukherj]

Raise AttributeError for non-existing meshing objects after switch_to solver.

Earlier instead of raising an Attribute error (general python behaviour), it was returning None. This has been corrected now.

There is an "is_active" method associated with session objects to check it's state.

[seanpearsonuk]

Hi @prmukherj,

Thanks for the improvements here — I agree that raising AttributeError on access to a switched meshing session is more idiomatic and aligns better with general Python expectations around invalid states.

That said, I wanted to highlight a few related areas that may warrant attention to ensure consistent and intuitive behavior:

1. dir(<mesh_session>) still raises a low-level exception, both before and after this change, due to access to internal attributes during __dir__ resolution. This could be surprising and uninformative to users. We may need to explicitly handle the switched state in __dir__() and return a helpful message or an empty list.

2. help(<mesh_session>) now presents stale info after the switch (which is a regression). It gives the impression the interface is still usable. This could mislead users who turn to help() for post-switch diagnostics. If there’s a way to influence help() output (e.g., overriding __doc__ or __getattribute__ with a custom response), we might consider doing so for clarity.

3. No public API to check if the session is active. At the moment, the only way to determine whether the meshing session is still valid is to access the private _switched attribute. It would be helpful to expose a method like is_active() or has_been_switched() to allow users to make this check in a clean and forward-compatible way.

Happy to discuss any of these further — let me know what you think.

[prmukherj]

Hi @prmukherj,

Thanks for the improvements here — I agree that raising AttributeError on access to a switched meshing session is more idiomatic and aligns better with general Python expectations around invalid states.

That said, I wanted to highlight a few related areas that may warrant attention to ensure consistent and intuitive behavior:

1. dir(<mesh_session>) still raises a low-level exception, both before and after this change, due to access to internal attributes during __dir__ resolution. This could be surprising and uninformative to users. We may need to explicitly handle the switched state in __dir__() and return a helpful message or an empty list.
2. help(<mesh_session>) now presents stale info after the switch (which is a regression). It gives the impression the interface is still usable. This could mislead users who turn to help() for post-switch diagnostics. If there’s a way to influence help() output (e.g., overriding __doc__ or __getattribute__ with a custom response), we might consider doing so for clarity.
3. No public API to check if the session is active. At the moment, the only way to determine whether the meshing session is still valid is to access the private _switched attribute. It would be helpful to expose a method like is_active() or has_been_switched() to allow users to make this check in a clean and forward-compatible way.

Happy to discuss any of these further — let me know what you think.

Yes @seanpearsonuk, I agree with this proposal. These changes are necessary. Others please put in your views as well.
Thank you.

[mkundu1]

Can you please explain the *_wo_exit changes in tests?

[prmukherj]

Can you please explain the *_wo_exit changes in tests?

Yes, these tests have to be exited manually as they are meshing sessions switching to solver in the test body.

[mkundu1]

The private _switched attribute is probably no longer required now.

[prmukherj]

The private _switched attribute is probably no longer required now.

Agreed, Removed now. Will need to update PyConsole accordingly in the next release.

[seanpearsonuk]

@prmukherj Is there a test for help(<meshing_session>)?

[prmukherj]

@prmukherj Is there a test for help(<meshing_session>)?

No, the help behavior was not updated yet. Will update it as part of this PR only then. Converting it to draft for further testing.