Check for missing types in more positions

[stephan-herrmann]

During #2543 we're discussing whether code around collectMissingTypes() is complete vis-a-vis today's Java.

More specifically @srikanth-sankaran asks:

For example in org.eclipse.jdt.internal.compiler.ClassFile.addAttributes() we are checking if super types and super interfaces are missing. But we don't check for and collect any missing types from permitted types - Should we ?

While changes in #2543 affected only MethodBinding.collectMissingTypes() details around TypeBinding.collectMissingTypes() may indeed be incomplete.

[stephan-herrmann]

• See also Tolerate missing types in more positions? #2658 - are these actually different issues?

[stephan-herrmann]

I happened to observe an obscure case where the SyntheticMethodBinding for a canonical record constructor contains missing types, but itself is not tagged with TagBits.HasMissingTypes.

@srikanth-sankaran, fyi this affects the constructor for SMB which was significantly modified in #3928. I believe it should propagate this flag. There may be more such cases.

The symptom of missing to do so may be to trigger this assertion in ProblemReporter.missingTypeInConstructor:

assert false : "The constructor " + constructor + " is wrongly tagged as containing missing types"; //$NON-NLS-1$ //$NON-NLS-2$

[srikanth-sankaran]

@stephan-herrmann - it looks like ProblemReporter.missingTypeInConstructor will be entered only if the method/constructor binding is tagged as having missing types but a drill down to gather them produces naught ??

I am stumped: where is this bit set for a method/constructor binding ? I expected to see a block such as the one below in org.eclipse.jdt.internal.compiler.lookup.ParameterizedMethodBinding.ParameterizedMethodBinding(ParameterizedTypeBinding, MethodBinding) but for an ordinary MethodBinding and my search brings up nothing!

for (int i = 0, max = this.parameters.length; i < max; i++) {
				if ((this.parameters[i].tagBits & TagBits.HasMissingType) != 0) {
					this.tagBits |=  TagBits.HasMissingType;

[srikanth-sankaran]

Does this PR #4085 address what you observed ?

[stephan-herrmann]

@stephan-herrmann - it looks like ProblemReporter.missingTypeInConstructor will be entered only if the method/constructor binding is tagged as having missing types ...

To give the full picture, there's a second path into missingTypeInConstructor(): when ProblemReporter.invalidConstructor() is called with a ProblemMethodBinding with reason MissingTypeInSignature. That reason can be set in findConstructor0, the normal case, plus two trickier situations:

• when parameterCompatibilityLevel yields COMPATIBLE_IGNORING_MISSING_TYPE - still based on a parameter of the constructor itself.
• when type inference encountered a missing type somewhere during the algorithm, see PGMB.computeCompatibleMethod18(). - Here the computed substitute may be free from missing types, but inference could be wrong due to a missing type along the way. That missing type we store first in the InferenceContext18 then in the ProblemMethod.

Anyway, the case that I observed started in getConstructor0() invoked with a SyntheticMethodBinding. Here we enter the path in parameterCompatibilityLevel that leads to the ProblemMethodBinding(MissingTypeInSignature).

It blew up when that transported us into missingTypeInConstructor() calling collectMissingTypes().

... but a drill down to gather them produces naught ??

There is actually another questionable code location: inside collectMissingTypes() we (i.e., Philipe) assume that if the MethodBinding isn't tagged HasMissingType than drilling into details is unnecessary. So in the buggy case, we do have missing type parameters, but we don't even look at them, because the method is not tagged. Should we remove that "if" in collectMissingTypes() to be even safer?

I am stumped: where is this bit set for a method/constructor binding ? I expected to see a block such as the one below in org.eclipse.jdt.internal.compiler.lookup.ParameterizedMethodBinding.ParameterizedMethodBinding(ParameterizedTypeBinding, MethodBinding) but for an ordinary MethodBinding and my search brings up nothing!

for (int i = 0, max = this.parameters.length; i < max; i++) {
				if ((this.parameters[i].tagBits & TagBits.HasMissingType) != 0) {
					this.tagBits |=  TagBits.HasMissingType;

Ordinary MethodBindings get tagged below STB.resolveTypesFor().

[stephan-herrmann]

Does this PR #4085 address what you observed ?

It is even more complete than what I had in mind :)

[srikanth-sankaran]

Thanks for the explanation. I am now stumped how I my search didn't bring up STB.resolveTypesFor(). Strange, I have released the change. Thanks