Tuple slicing is not supported

[o11c]

mypy -c 'v = (1, 2, 3); major, minor = v[:2]'

<string>:1: error: Tuple index must an integer literal

Oddly, this was not a failure before I rebased my branch.

[JukkaL]

Yeah, mypy should support this, obviously. Not sure why it got broken, though (or why it used to work).

I think that this needs to be fixed in visit_index_expr_helper (mypy/checkexpr.py).

[JukkaL]

Basic support for this is now in (by @jhance).

I'm leaving this open for now as a few additional things would be nice to have:

• Negative literal int in slice (e.g., t[:-1]).
• Non-literal in slice for uniform tuples. Basically a tuple like Tuple[int, int, int] would be treated similarly to Tuple[int, ...] for slicing. (See also Allow non-literals when indexing uniform tuples #899.)

[jhance]

Basically a tuple like Tuple[int, int, int] would be treated similarly to Tuple[int, ...] for slicing

I firmly believe this is evil. Having n=0 and slicing on n: would give me a different type than slicing on 0:.

Then again I hate the idea of variable-length homogenous tuples so I think my opinions on this are probably irrelevant.

[o11c]

There are plenty of ways to get the same object with a different type, there is nothing evil about that.

[jhance]

Its not about the fact that I get the same object with a different type, its the fact that I'm allowed to:

1. Do things engaging with variable-length tuples simply because it happens to be homogenous, despite non-variable-length homogenous tuples not being special in the type system.
2. Its too easy to accidently end up with a homogenous tuple when you don't intend to. If you want this why are you not using sequence?

[o11c]

But even if you don't intend to get a homogenous tuple, you'll just get an error a little bit later.