Investigate environment detection in our overall environment UX to see if there are scenarios where discovery is unnecessary

[karrtikr]

Related: #17498 #18647 #18671

Users can use the API or the refresh icon for triggering it if needed:

Check #19102 (comment) for investigation results.

[karrtikr]

Auto-triggering discovery is probably not needed. Only a couple places need it:

• When auto-selecting interpreters for a new workspace.
• Triggering it for every scope exactly once (across sessions).

[karrtikr]

Question: Is triggering it once in the first session even needed, if an interpreter is already selected?

Perhaps not from a user perspective (#18647), but yes, the discovery API is only considered to be ready for query once refresh has been triggered.

Reasons for doing it in the first session:

• Jupyter is going to do that anyways via the API to fetch kernels.
• Internally it is assumed discovery is triggered atleast once, changing which might not be worth it for now. For example, Auto-selection (which triggers discovery) is always triggered even if an interpreter is explicitly selected, so that user has an auto-selected interpreter to fallback on if the selected interpreter is cleared. Another example: Investigate environment detection in our overall environment UX to see if there are scenarios where discovery is unnecessary #19102 (comment)
• Select Interpreter section in the getting started section already has environments populated.
• Rest of sessions should be good.

[karrtikr]

We also need to make sure legacy jupyter integration APIs still work.

[karrtikr]

Proposed approach, only auto-trigger discovery if:

• A user attempts to choose a different interpreter
• When a particular scope (workspace or global) is opened for the first time
• When jupyter queries legacy discovery APIs for envs (Jupyter currently queries this on startup)
• Environment cache is empty

Note we also need to make sure that the selected environment has the latest version, now that discovery isn't run by default.

[karrtikr]

Verification steps:

If Jupyter is not installed or disabled, verify we auto-trigger discovery only if:

• A user attempts to choose a different interpreter when quickpick list is opened the first time.
• When a particular scope (workspace or global) is opened for the first time
• No Python is installed

If Jupyter is installed, discovery is always triggered.

[karrtikr]

Certain components (like diagnostics) call for hasInterpreters() API, which requires that discovery is triggered atleast once for it to work. See if we can avoid calling it and make sure information returned by it is reliable enough.

[rchiodo]

/verified?

It does trigger discovery just when opening 'Select interpreters'? Is this expected?

Otherwise behaves as you outlined.

[karrtikr]

A user attempts to choose a different interpreter when quickpick list is opened the first time

Yep, that's what the above point means in #19102 (comment), only happens the first time though.