Support REPL `import` completion (needs investigation first)

[kfatyuip]

Feature or enhancement

Proposal:

press TAB after

import o

shows

object()  oct(      open(     or        ord(

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

### Tasks
- [ ] Design the infrastructure for discovering importable modules
- [ ] Benchmark the cost of adding `import` completion (how long it takes to create the cache, how fluid it is in the end)

[picnixz]

I'd be glad to have this but I'm afraid it will be inefficient. It's possible to autocomplete attributes because the namespace already exists at runtime. However, if one writes import x, we need to discover all modules that start with x (and those in site-packages!) and that's probably a no-go.

cc @pablogsal

[pablogsal]

I'd be glad to have this but I'm afraid it will be inefficient. It's possible to autocomplete attributes because the namespace already exists at runtime. However, if one writes import x, we need to discover all modules that start with x (and those in site-packages!) and that's probably a no-go.

cc @pablogsal

This is the case but is not unheard of. For example, IPython (and many other REPL) does it:

https://github.com/ipython/ipython/blob/364e544a76c590ff161d9fffacf35533db478324/IPython/core/completerlib.py#L108-L152

We could profile it and if we see is not too bad we can try to add support for it

[picnixz]

Maybe we could also make this feature on demand? that way we wouldn't penalize current users if they don't need that feature. But I would definitely be interested in knowing whether it's feasible for us or not (and would be happy to review such PR if we find a good solution)

[pablogsal]

Maybe we could also make this feature on demand? that way we wouldn't penalize current users if they don't need that feature. But I would definitely be interested in knowing whether it's feasible for us or not (and would be happy to review such PR if we find a good solution)

We should first benchmark to know what is the cost. My guess is that is close to be negligible. We can also try to back off if there are too many packages or add some other stopgaps to control the impact

[picnixz]

I've added two tasks for users that want to work on that. We need a PoC first and it's better for anyone wanting to help NOT to create any PR first but link their own fork in this issue as well as their results. We'll decide on how to proceed afterwards.

[vadmium]

See #69605 for my attempt ten years ago

[picnixz]

Oh there was a precedent. Maybe it's better to close this issue then (it seems the IPython case was also mentioned in the original issue).

EDIT: Considering that #117312 was closed as duplicate, I'll also close this one as a duplicate.