BlockNode.Location uniqueness is documented and relied upon, but not enforced or true

[omarsy]

Summary

setNodeLocations claims twice that it "ensures uniqueness of
BlockNode.Locations" (gnovm/pkg/gnolang/preprocess.go:6324-6325 @
33763e8269), SetBlockNode keys cacheNodes by Location with a bare
// XXX duplicate? (store.go:934-947), and the checker meant to enforce it,
checkNodeLinesLocations (preprocess.go:6350), is an empty // TODO: XXX
stub.

The invariant does not hold. else if breaks it.

Reproducer

package main

func main() {
	x := 0
	if x == 0 {
		println("a")
	} else if x == 1 {
		println("b")
	} else {
		println("c")
	}
}

Transcribing the file node and bucketing block nodes by GetLocation():

DUPLICATE LOCATION test/main.go:7:9-11:3 -> [*gnolang.IfCaseStmt *gnolang.IfStmt]
block nodes: 7, duplicate locations: 1

Cause

go2gno.go:589 sets the synthesized else-branch IfCaseStmt's span from
gon.Else, which for an else if is the inner IfStmt:

elseStmt := IfCaseStmt{Body: ess}
if gon.Else != nil {
	setSpan(fs, gon.Else, &elseStmt)
}

Both block nodes then receive identical Location{PkgPath, File, Span}.

Impact

As far as I can tell, harmless today: neither colliding block can host a
declaration, so nothing currently disambiguates on their Locations in a way that
misbehaves. I have not audited every Location consumer, and SetBlockNode
silently overwrites on collision.

Filing because the invariant is load-bearing and growing more so — #6060 keys a
local type's TypeID to its declaring block's Location. A latent duplicate is a
landmine for future work, and the stub checker means a new one would go
unnoticed.