Type inferencing error (only in Eclipse, not in groovyc)

[mauromol]

Consider this:

package test70

import groovy.transform.CompileStatic

@CompileStatic
class Test70 {

	void foo() {
		Object v1 = 'a'
		Object v2 = 'b'
		[v1, v2].each {
			v ->
			if(v instanceof Map) {
				v.entrySet().each { e ->
					def s = e.value
					if(s instanceof String)
					e.value = s.toUpperCase()
				}
			}
		}
	}
}

An error is given on both occurrences of e.value, because e is evaluated as Object instead of Map.Entry.
This used to work until some version of the Greclipse plugin (now I'm on 4.3.0.v202107132103-e2009).
If I compile this with Gradle/Groovy, the JAR is built successfully.

If I remove @CompileStatic, I get underlining for v.entrySet() and again on both occurrences of e.value.

If I try to give e a type in the closure:

				v.entrySet().each { Entry e ->
					def s = e.value
					if(s instanceof String)
					e.value = s.toUpperCase()
				}

I see this:

• with @CompileStatic an additional error is given on Entry, saying it's expecting Object.
• without @CompileStatic, instead, the underlining for e.value is fixed, so the situation improves.

[eric-milles]

It is this change where the raw type Map is causing the return type of v.entrySet() to be changed to Set.

groovy-eclipse/base/org.codehaus.groovy30/src/org/codehaus/groovy/transform/stc/StaticTypeCheckingVisitor.java

Line 5740 in 74e24d8

returnType = returnType.getPlainNodeReference(); // GROOVY-10049: do not return "Stream<E>" for raw type "List#stream()"

As a workaround, you can add a cast: ((Map<?,?>)v).entrySet() and you may also need to add a cast to v1 to stop the flow type of String.

This will be an issue for groovyc as well wherever GROOVY-10049 is addressed.

[eric-milles]

Here are forms that should work the same; v.entrySet().each { e -> can be expressed as v.each { e -> or for (e in v) {:

@groovy.transform.TypeChecked
void test() {
  Object v1 = 'a'
  Object v2 = 'b'
  [v1, v2].each { v ->
    if (v instanceof Map) {
      v.each { e ->
        def s = e.value
        if (s instanceof String)
          e.value = s.toUpperCase()
      }
    }
  }
}

STC should infer Entry for e through DGM static <K, V> Map<K, V> each(Map<K, V> self, @ClosureParams(MapEntryOrKeyValue.class) Closure closure) for the first.

for (e in v) { and for (Map.Entry e in v) { both fail with similar errors but for (e in (Map)v) { does work. I'll have to investigate this later.

[eric-milles]

ready to test

[mauromol]

Version 4.3.0.v202107240018-e2009 fixes the problem for me, thank you!

[mauromol]

Still, if I remove @CompileStatic, I get two underlinings that shouldn't be there IMHO:

Should I open a new issue for this?

[eric-milles]

In this specific case, it can be determined that the only possible type for "v" is String. So an "instanceof Map" check presents an inconvertible type and so Map is ignored. You can add a typecast to the list: [(Object)v1, v2] or to the declaration Object v1 = (Object)'a' if you wish.

[mauromol]

Still, it's unexpected that static type inference and dynamic type inference behave in different ways in this example...