I am taking liberty and merge this change without code review because:

• it fixes nasty bug which causes CI fail in other PRs
• it affects only subscripted types
• change is simple and limited

If there are any comments/improvements can be implemented in another PR...

I don't think any of the test failures are particularly relevant here (they all affect every other branch right now and I think aren't issues that we can fix). I don't fully understand ones on the Python development branch; I've reported the pypu one to pypy.

Yeah I ser. Unfortunately it blocks my automerge ...

One extra check I'd add here:

d = { slice(1, 5): "some_value" }
cython.typeof(d[1:5])

In this case the return type should be a single element (or just object).

I've never seen this used in real code so I think the main thing is that it doesn't infer something incorrect like a dict return type. But it's fine if it just says "object"

One extra check I'd add here:

d = { slice(1, 5): "some_value" }
cython.typeof(d[1:5])

In this case the return type should be a single element (or just object).

I've never seen this used in real code so I think the main thing is that it doesn't infer something incorrect like a dict return type. But it's fine if it just says "object"

I think already cython in master contains error your code produces dict object. Python on the other hand:

>>> d = { slice(1, 5): "some_value" }
>>> d[1:5]
'some_value'

I don't fully understand ones on the Python development branch

See python/cpython#149583 (comment)

d = { slice(1, 5): "some_value" }
cython.typeof(d[1:5])

We could add the following to the start of the analyse_as_pyobject method:

if is_slice and base_type.is_pyanydict_type:
    is_slice = False

For a dict, a slice lookup is never special.

d = { slice(1, 5): "some_value" }
cython.typeof(d[1:5])

We could add the following to the start of the analyse_as_pyobject method:

if is_slice and base_type.is_pyanydict_type:
    is_slice = False

For a dict, a slice lookup is never special.

The fix is totally unrelated to this PR so I fixed it in separate PR: #7702