analyze_types: better support for type aliases

[pohly]

The recently merged semantic analysis has one drawback when dealing with type aliases: they always get replaced with the underlying type. Example:

package somepkg

import "example.com/anotherpkg"

type SomeType = anotherpkg.AnotherType

A pattern somepkg.SomeType.ForbiddenMethod does not match because SomeType is an alias. What works is anotherpkg.AnotherType.ForbiddenMethod.

This is a) unexpected and b) can cause a pattern that works with a version of a package where the type is not an alias to stop working when the packages switches to a type alias.

It would be better if forbidigo matched against both the underlying type (anotherpkg.AnotherType) and the alias (somepkg.SomeType).

[pohly]

The problem is that in many cases, we get the underlying type from static code analysis:

forbidigo/forbidigo/forbidigo.go

Lines 273 to 285 in 1396000

	// If we are lucky, the entire selector expression has a known
	// type. We don't care about the value.
	selectorText := v.textFor(node)
	if typeAndValue, ok := v.runConfig.TypesInfo.Types[selector]; ok {
	m, p, ok := pkgFromType(typeAndValue.Type)
	if !ok {
	v.runConfig.DebugLog("%s: selector %q with supported type %T", location, selectorText, typeAndValue.Type)
	}
	matchText = m + "." + field
	pkgText = p
	v.runConfig.DebugLog("%s: selector %q with supported type %q: %q -> %q, package %q", location, selectorText, typeAndValue.Type.String(), srcText, matchText, pkgText)
	return []string{matchText}, pkgText
	}

I'm not sure whether it is possible to detect that the type was reached through an alias.

[ashanbrown]

Even with the introduction of Alias nodes, it appears the typing analysis code still sets the underlying type in TypeAndValue at https://github.com/golang/go/blob/608acff8479640b00c85371d91280b64f5ec9594/src/go/types/api.go#L233. Since aliases are interchangeable, I'm not sure if it is even meangful to know which alias is in use here (maybe more than one is possible?). I think you'd have to look up all possible aliases that are in scope at this point and then try to match any of them. Maybe possible; definitely tricky.