ECJ fails to compile 3 nested generic method calls whilst javac succeeds

[coehlrich]

ECJ fails to compile this whilst javac succeeds along with ECJ 2025-09:

	public static void main(List<?> l) {
		get(get(get(l)));
	}
	
	public static <T> List<?> get(List<T> l) {
		return l;
	}

If I remove one of the get calls or change any of the type parameters of List for the get method then it compiles.

This seems to be due to the bounds ending up as:

Type Bounds:
	TypeBound  T#0 = capture#1-of ?
	Dependency (?)#2 = T#1
	TypeBound  (?)#2 = java.lang.Object
	TypeBound  T#1 = java.lang.Object
	Dependency (?)#4 = T#3
Capture Bounds:
	List<(?)#4> = capt(List<?>)
	List<(?)#2> = capt(List<?>)

and JLS §18.3.2 stating αi = R implies the bound false.

@stephan-herrmann

[coehlrich]

The bug was introduced in #4520

The java.lang.Object comes from adding bounds from an inference context that has already been finished here:

eclipse.jdt.core/org.eclipse.jdt.core.compiler.batch/src/org/eclipse/jdt/internal/compiler/lookup/InferenceContext18.java

Line 713 in 545483d

this.currentBounds.addBounds(innerContext.currentBounds, this.environment);

Updating that to explicitly use b3 causes GenericsRegressionTest.testBug499725 to fail due to the b3 field not always being updated when b3 is determined multiple times even with different InvocationSites

Here is a list of times b3 gets updated for the test excluding Collections.emptyList and Arrays.asList: https://gist.github.com/coehlrich/f7371620c6a500d3434c2ba3174141a7
Attempted to update b3 is for the line that only updates b3 if it is currently null
Updating b3 is for the line that always updates b3

I have been working on a fix on https://github.com/coehlrich/eclipse.jdt.core/tree/fix-nested-inference but after I fixed testBug499725 testBug496579 has started failing

[stephan-herrmann]

The java.lang.Object comes from adding bounds from an inference context that has already been finished here:

eclipse.jdt.core/org.eclipse.jdt.core.compiler.batch/src/org/eclipse/jdt/internal/compiler/lookup/InferenceContext18.java

Line 713 in 545483d

This line indeed looks dubious, and https://bugs.eclipse.org/511252 where this was added gives no real justification for it.

[stephan-herrmann]

Resolved via #4735