After CL183137, there is a problem with typenames completion where we want to do a type conversion.

It's caused by this check:

func (c *completer) matchingTypeName(cand *candidate) bool {
+	if !c.wantTypeName() {
+		return false
+	}

(Although removing it does solve this particular issue, we get other issues out of it, this is why someone more experienced with gopls needs to take a look.)

It bites when we try to autocomplete inside expression context and expect to use a type conversion:

package main

import (
	"fmt"

)

type coordX float64
type coordY float64

type Point struct {
	X coordX
	Y coordY
}

func main() {
	fmt.Println(Point{
		// <> completion
		// a want to do coordX(value) here
	})	
}

In my understanding, a type name is a valid candidate even when not inside type expression context, since it can be an attempt to write a type-converting expression.

Actual behavior

coordX is scored with 1.0, as all other "non matching" candidates.

Expected behavior

coordX is at least a little bit higher than 1.0.

Note that a575db70c06744141b7a52bb6c4aff8d860588c6 (a commit before mentioned CL works as expected):

If we need to define "when a type name is appropriate outside of wantTypeName", I would say it's "an expression context that expects a type that can be acquired by doing a conversion to the given type name". So, if we autocomplete Foo-typed expression, Foo is a matching candidate (score>1).