Slicing twice doesn't work

[konstin]

Iterators are expected to allow slicing twice, i.e.

>>> list(range(0,100)[10:15])
[10, 11, 12, 13, 14]
>>> list(range(0,100)[10:][:5])
[10, 11, 12, 13, 14]

This means essentially that [a:a+b] equals [a:][:b]

With a Search object slicing twice doesn't work:

>>> Search()[10:15].to_dict()
{'from': 10, 'query': {'match_all': {}}, 'size': 5}
>>> Search()[10:][:5].to_dict()
{'from': 0, 'query': {'match_all': {}}, 'size': 5}

(Sorry for submitting an incomplete issue first, I accidentally hit the wrong key)

[lolletsoc]

Hey @konstin,

I've created a branch in my fork with minor changes to support this. Take a look at the tests and let me know - specifically this one.

@honzakral, what do you think?

[konstin]

Hi @lolletsoc, the changes does solve the issue with slicing twice.

The behavior for slicing multiple times still doesn't match the expected pythonic behavior though. Even though I doubt anyone uses this in practice, it might be worth fixing it,

>>> list(range(0, 100)[10:][:5][5:])
[]
>>> Search()[10:][:5][5:].to_dict()
{'query': {'match_all': {}}, 'from': 15, 'size': 10}

[lolletsoc]

Yeah, exactly why I pointed you at that test. If we try to mold this around the pythonic way, we're going to end up with {'query': {'match_all': {}}, 'from': 0, 'size': 0}. Will wait on @honzakral's reponse - maybe he has some insights.

[honzakral]

Thanks for raising this issue @konstin, I understand the confusion. Right now any slicing will reset the from/size combination which may be unexpected. What people might be looking for is additive slicing - further limiting the scope, what python's lists do and even what django's querysets do (0).

That sounds very reasonable to me and I think we should do it, we just need to be careful about documenting it properly since it is a breaking change.

Thanks again for bringing it up!

0 - https://github.com/django/django/blob/master/django/db/models/sql/query.py#L1603-L1624