Provide LenTextSize for Arc<String>

[CAD97]

Unfortunately, this impl leads to infinite recursion for, say &(), or any reference type that doesn't impl LenTextSize. Cow::Owned(String::new()) falls into this trap via the type inference variable for the B in Cow<'_, B>.

I've reported the wonderful error which isn't even being caught by the "recursive requirements" folder.

I have some alternate constructions to try, but I'm putting this one in the pool before I forget that &Arc<String> should be accepted by TextSize::of.

[matklad]

Do we test ::of("") (without &) anywhere?

[CAD97]

In the doc-tests for TextSize::of, yes, but I'm going to add it here as well.

[matklad]

Hm, could you also check that this impl allows downstream crates to impl the trait? rowan's SyntaxNodeText comes into mind.

[matklad]

I guess, we can actually put this into an integration, test because it is a downstream crate.

[CAD97]

For my own sanity, here's a (nonexhaustive) list of types we definitely want to be (able to be) accepted:

• &str
• char
• &String
• &Arc<str>
• &Arc<String>
• &Rc<str>
• &Rc<String>
• &smol_str::SmolStr
• &sorbus::green::Node
• &rowan::GreenNode
• &rowan::SyntaxText
• &rowan::api::SyntaxNode

And types that don't have to be accepted, but would be nice to:

• &{above type}

I think we almost get what we want from LenTextSize::len_text_size(&self), LenTextSize for str, LenTextSize for char, and LenTextSize for D: Deref where D::Target: LenTextSize, but that fails to "upstream crates may add a new impl of trait std::ops::Deref for type char in future versions", thanks, coherence

[CAD97]

Based on further experimentation, I think we have to just accept that this is a limitation of the Rust language that deref coersion is not occurring here, and tell users to deref their smart pointers to str. Or if they want to get deref coersion, use TextSize::of::<&str>. I've asked a question on urlo about whether the lack of deref coersion here is fundamental or just an implementation limitation.

[matklad]

What about removing blanket impls and just adding a tone of impls manually?

[CAD97]

Yeah, that seems to be the best bet. I'm doing such now.