Create a NamedTuple instance for NamedTuples

[DanielNoord]

Steps

• For new features or bug fixes, add a ChangeLog entry describing what your PR does.
• Write a good description on what the PR does.

Description

This is the first step towards some improvements to the NamedTuple brain. One of the first things I thought we should fix is the fact that we currently lose the fact that something is a NamedTuple. We don't add it to the object's bases and they are undistinguishable from other instances.

I'm not sure if adding a new Instance is actually necessary. I thought it might make sense as NamedTuple behaves differently than tuple and can have default values for example. I have created the T(opic): NamedTuple label to identify some issues related to NamedTuple. I thought a new Instance type could be the basis for solving some of these.

Changes have been tested against current latest commit of pytest.

Type of Changes

	Type
✓	✨ New feature

Related Issue

[Pierre-Sassoulas]

@cdce8p do you think you'll have time to review this ? I see you self requested a review two months ago. I'm trying to release astroid 2.10 :)

[areveny]

I'm beginning to understand Astroid brain a lot better after looking at this!

[cdce8p]

Had finally time to look at it. Tbh I'm myself not exactly the "brain" expert. However from what I've gathered so far, I'm not sure this is the best solution to the problem.

I do understand your desire to have NamedTuple / namedtuple at least somewhere which isn't the case today, but it seems strange to add a new instance type just for that. Wouldn't it be much more intuitive to insert a new base which itself has tuple as base? That's at least roughly similar to the pattern we used in the typing brain.

Regarding the remaining changes, I would probably need more time but it looks like there are 3-4 others bundled here as well. That makes it more challenging to review.

Going forward, I would suggest we (1) move the PR back to 2.11, (2) unbundle additional changes, and (3) explore if inserting a dummy base class would work.

[DanielNoord]

I do understand your desire to have NamedTuple / namedtuple at least somewhere which isn't the case today, but it seems strange to add a new instance type just for that. Wouldn't it be much more intuitive to insert a new base which itself has tuple as base? That's at least roughly similar to the pattern we used in the typing brain.

Hmm, I guess this depends on what bases and Instances are supposed to be. Personally I think it makes more sense to create a NamedTuple Instance that can be re-used rather than defining them through f-strings. In my mind it goes like Instance is an instances of a general ClassDef. A sub-class of Instance is thus an instance of a specific type of ClassDef.
But I can see why that might not always be logical.

Regarding the remaining changes, I would probably need more time but it looks like there are 3-4 others bundled here as well. That makes it more challenging to review.

Going forward, I would suggest we (1) move the PR back to 2.11, (2) unbundle additional changes, and (3) explore if inserting a dummy base class would work.

I'll try and separate some of the work here. Some of the typing should be fairly straightforward to extract and then review for example. I'll get back to you!

[Pierre-Sassoulas]

@DanielNoord can we close this ?

[DanielNoord]

I'll triage this somewhere next week to see if there is anything still relevant in here

[DanielNoord]

Closing!