Relax contraints on list pattern to test emptiness in collections

[simonthum]

It is possible to pattern-match on an empty list/array/string/span/whatnot:

if (x is null or []) { /*code*/ }

It is clear why this should remain unworkable for enumerables. But there is no technical reason this can't work for collections (ICollection, IReadOnlyCollection), which are frequently returned from calls. There is even guidance to do so.

Currently, I get

error CS0021: Cannot apply indexing with [] to an expression of type 'IReadOnlyCollection<int>'

which forces me to fall back to x is null or {Count: 0} (or Length). Not too bad, but it harms regularity.

Thus, I suggest lifting this restriction in the particular case of an emptiness pattern.

https://sharplab.io/#v2:CYLg1APgAgTAjAWAFBQAwAIpwCwG5nJQDMmM6AwugN7Lp2YlTboCyAFAJTW328D0fdAEl0AUwAeAB1EBjAC7o5Ae3QA7JQoDuSgE4BrED150BmOEQA8AS1VyAfOnHoAvOgDaAXXxJj9I8dMRCWl5RRVtfUMfX3RA8iUAGwTZOSslVWtbBydXT28Y2MEhACVRAENgAHlVBIBPeKSUtIybe0cXdy9/Xm76UwB1XT0AZyiC1s923K7o30DSiuq6gBkrYblMtpzO/JjA1fXN7I683pNBLAxtgCJr3d8rADN0Nic1tQBXJPRdTq45AAWOiUmjUolBAEEdABzD4AW1EtgAouIZKJJKl0px7ugAL7IXFAA=

[jaredpar]

Closing as this is a request for a language change vs. being a compiler bug. Please open this as a discussion on dotnet/csharplang.

[simonthum]

Yeah, I was unsure about it, but the language change would be very small (I guess), and would allow a certain idiom to be streamlined much better.

Note that according to sharplab, the indexer is not used in the lowered pattern, as expected. So the error does not make much sense.

[333fred]

Duplicate of #6574

[simonthum]

This is not a duplicate, because I am suggesting a much smaller change that is intentionally unrelated to IEnumerable but just to ICollection. It also does not affect collection expressions, just pattern matching.

Currently, methods that adhere to the guidelines are likely to yield ICollection which, unlike the more specific IList, cannot be matched for emptiness the way other collections can. That is even true when the underlying collection implements IList.

The implied language change is rather small, and would preserve the performance characteristics of is []. It does not actually relate to collection expressions per se. That is why I think it quite different from #7744.

[jnm2]

@333fred Were you thinking of #6574, list patterns (not collection expressions)?

That proposal proposes "allow list patterns on non-countable input types." This subsumes what @simonthum is requesting, but goes much further. The request here is to allow the empty list pattern when the input type is countable but is merely not indexable.

I would much rather see us go the whole way and allow things like expr.Where(v => ...) is [var single], than to do something special for the empty list pattern.

[333fred]

@333fred Were you thinking of #6574, list patterns (not collection expressions)?

Yes, my mistake. I've updated my comment.

[simonthum]

I see that proposal #9005 would cover this request.

However, the request here is quite doable and improves the outcome for code that follows published guidelines.

Indeed I was surprised this case did not work OOTB. I often have code where the caller knows whether an empty collection just returned from somewhere is an error or needs to be handled differently.

Having seen many careless "helpful helpers" to address this, I'd prefer a proper language idiom.

Would you take patches?

[jnm2]

Changes would need a spec and need to be championed by a language design team member and taken through the LDM process. If approved, then yes, community patches are sometimes welcome when feasible. I do think the larger proposal would end up being desired if we took the time to take this to LDM.

[smarts]

I'm ambivalent about #9005, but if that one gets rejected then I think this one should be revived. Ideally though, it feels like this could be implemented with a lot less pushback (and possibly quicker?) than the enumerable pattern matching.

[jnm2]

I don't imagine much pushback, since the last LDM on this topic was "we still think we should finish this."