x/tools/gopls: improved completion of iterators

[adonovan]

Unlike iterators in many languages, which always have a distinct type such as Iterator<T>, an iterator in Go 1.23 is any function that accepts a yield func(T) bool. (The name "yield" is conventional but not required.) Often of course it has the type iter.Seq[T], but this isn't necessary, and when a method of a collection is itself an iterator (as opposed to returning one), as in this example:

func (c *Collection[T]) Elements(yield func(elem T) bool) { ... }

for range c.Elements { ... }

it may not be immediately obvious that one is looking at an iterator. Completion may be able to help here. Given a prefix for range c., we should offer completions such as for range c.Elements { ... }. Perhaps, in statement context, given the prefix c., it should offer not just c.Elements, but for for elem := range c.Elements { ... } (using the name of the yield parameter). There are likely other opportunities too.

[atdiar]

It's a bit off-topic for that issue and should be discussed in the main discussion but I'm wondering if making the bool return type a defined type has been addressed?
Seems that it could help discovery.
Admittedly, it will require some kind of closures to transform old forms of iterators using the plain bool type, if there is any.
Maybe it has been addressed in which case, just ignore.

[adonovan]

I'm wondering if making the bool return type a defined type has been addressed?

I think you're asking a question about the language change; this issue is about gopls completion support.

You might want to read the lengthy discussions around #61405 and #61897.

[atdiar]

@adonovan 👍🏿 thanks. I hadn't seen #61897

[KreanXie]

lg

[gopherbot]

Change https://go.dev/cl/592915 mentions this issue: gopls/completion: prefer rangeable funcs in range statements

[muirdm]

We can also improve the .range! postfix snippet to work on rangeable funcs (in statement context).