Bringing back numpy 1.x support?

[futurewasfree]

Just thinking out loud here and looking at the diff for numpy 2.x: afa736b
wondering if it's possible to bring back support for numpy 1.x and support both of them with a help of numpy.__version__ check?

I work on quite a big codebase with multiple projects and the reason why I'm asking is the following:
I have found myself a bit stuck between 3 constraints:

1. One of the project still uses numpy 1.x due to old 3rd party dependency and we can not replace it just now
2. This won't be a big problem if not recent renaming in 8.x. So now I have to pin down version globally to be < 8. So that that old project could resolve it to pydantic_numpy 5.x
3. This also means I can not upgrade numpy globally to 2.2. Remember latest typing fix we've did with tuple and dimensions.

Maybe there are other solutions to break this knot, but I want to hear your opinion on numpy 1.x. Maybe it's deprecated a bit too early?

[caniko]

Can't you just use the older versions?

[futurewasfree]

But the older versions won't work with numpy 2.x?

[futurewasfree]

I would like to use the latest numpy (if possible) and latest pydantic-numpy but the recent change to type names prevent me from doing it.
Depending on which version get resolved I would get different type names in pydantic-numpy. That's not a super stable set up if I do not pin-down max version.

[caniko]

You want to support both bumpy V1 and V2 while pinning the version of pydantic-numpy. Is this correct?

[futurewasfree]

Yes, this is correct. With or without pinning.

[caniko]

Without pinning should work already

[futurewasfree]

Let's say I have a shared library named base and I have two more libs: library_a and library_b.
base depends on pydantic_numpy and extends types. library_a has that old 3rd party lib that forces it to resolve dependencies to numpy 1.x. library_b in opposite uses the latest and greatest numpy 2.x.
Problem arises in version 8.0 where types were renamed and gives clash between pydantic_numpy 5.x (last one supporting numpy 1.x) and 8.x. In base, but also in library_a.