I often find myself defining function args with list[SomeClass] type and think “do I really care that it’s a list? No, tuple or Generator is fine, too”. I then tend to use Iterable[SomeClass] or Collection[SomeClass]. But when it comes to str, I really don’t like that solution, because if you have this function:
def foo(bar: Collection[str]) -> None:
pass
Then calling foo("hello") is fine, too, because “hello” is a collection of strings with length 1, which would not be fine if I just used list[str] in the first place.
What would you do in a situation like this?
I’m not sure why you wouldn’t just use packing to pass in a list of some objects that you need iterate over? Isn’t it normally bad form to pass lists as arguments? I feel like I’ve read this somewhere but can’t cite it
Kids these days and their type hinting. Back in my day, all objects were ducks, and we liked it!
🦆
A function that infinitely yields strings can be an Iterable without being a collection. You can create collections you can’t irritate over by implementing only certain functions.
They both require different dunder methods.
I know that
IterableandCollectionaren’t the same. My point is, that if you useIterable[str]orCollection[str]as a more flexible alternative tolist[str]you no longer have any type-hinting support protecting against passing in a plain string and you could end up with a subtle bug by unexpectedly looping over['f', 'o', 'o']instead of['foo'].
I’m rusty on my type hints because I’ve been living in lua land lately, but from ye olde PEP 20
Explicit is better than implicit.
I’d combine them so the hint was something like
Union[Collection[str], str]But what if you actually don’t want
strto be valid?If you’re writing code that generic, why wouldn’t you want
strto be passed in? For example,Counter('hello')is perfectly valid and useful. OTOH,average_length('hello')would always be1and not be useful. OTOOH, maybe there’s a valid reason for someone to do that. If I’ve got a list of items of various types and want to find the highest average length, I’d want to domax(map(average_length, items))and not have that blow up just because there’s a string in there that I know will have an average length of1.So this all depends on the specifics of the function you’re writing at the time. If you’re really sure that someone shouldn’t be passing in a
str, I’d probably raise aValueErroror a warning, but only if you’re really sure. For the most part, I’d just use appropriate type hints and embrace the phrase “we’re all consenting adults here”.Maybe something like passing in a list of patterns which should match some data, or a list of files/urls to download would be examples of where I would like to be generic, but taking in a string would be bad.
But the real solution be to convert it to
foo(*args: str). But maybe if you take 2Container[str]as input so you can’t use. But no real world example comes to mind.
Look at the official docs. There is a table part way down stating which methods are available for each. I pick the one closest to how I use it. So if I’m not mutating I’ll use Sequence over List to inform the caller I’m treating as immutable and to safe guard myself from mutating it in my implementation via static type analysis.
strmatches most of these contracts, though, requiring additional checks if astrwas passed or one of these collections containing strings.
I’d leave a docstring:
def foo(bor: Iterable[str]) -> None: """foos bars by doing x and y to each bar"""Type hinting isn’t intended to prevent all classes of errors, it’s intended to provide documentation to the caller.
Iterable[str]provides that documentation, and a docstring gives additional context if needed. If you want strict typing assurances, Python probably isn’t the tool you’re looking for.This + an assert seems like the way to go. I think that
strshould never have fulfilled these contracts in the first place and should have a.charsproperty that returns a list of one-character-strings. But this change would break existing code, so it is not going to happen.IDK, I think strings being simple lists is less surprising than having a unique type. Most other languages model them that way, and it’s nice to be able to use regular list actions to interact with them.
It’s really not something I’m likely to run into in practice. The only practical way I see messing this up is with untrusted inputs, but I sanitize those anyway.
