Completion for unimported types doesn't work

[snjeza]

Fixes #1502

[stephan-herrmann]

Found this only after commenting in #1502 - still a few answers to my questions there would help the reviewer :)

[akurtakov]

Are @snjeza 's answers on the issue good?

[jukzi]

@snjeza please resolve conflicts and update the commit message to reflect the change and and contain a reference to the bug

[snjeza]

@jukzi I have updated the PR.

[jukzi]

LGTM, and it usefull. i tested it. It could be improved, please:

package aaa;

import java.util.ArrayList;

public class X {
	void foo() {
		new ArrayList<>().add(null);
		new HashMap().cl|
		}
}

results in not compile-able completion:
new HashMap<K, V>().clear();

it would be better if it completes to diamond operator new HashMap<>().clear(); as K and V are not defined.

[snjeza]

results in not compile-able completion:
new HashMap<K, V>().clear();
it would be better if it completes to diamond operator new HashMap<>().clear(); as K and V are not defined.

This should be resolved in eclipse.jdt.ui - https://github.com/eclipse-jdt/eclipse.jdt.ui/blob/a4c1b5bc5dbb294e8c23b93296effeb1d5c2cae0/org.eclipse.jdt.ui/ui/org/eclipse/jdt/internal/ui/text/java/JavaMethodCompletionProposal.java#L474

[rgrunber]

results in not compile-able completion:
new HashMap<K, V>().clear();
it would be better if it completes to diamond operator new HashMap<>().clear(); as K and V are not defined.

This should be resolved in eclipse.jdt.ui - https://github.com/eclipse-jdt/eclipse.jdt.ui/blob/a4c1b5bc5dbb294e8c23b93296effeb1d5c2cae0/org.eclipse.jdt.ui/ui/org/eclipse/jdt/internal/ui/text/java/JavaMethodCompletionProposal.java#L474

Looks to still be an issue for me with this PR on latest master of jdt.core and latest jdt.ui.

[stephan-herrmann]

results in not compile-able completion:
new HashMap<K, V>().clear();
it would be better if it completes to diamond operator new HashMap<>().clear(); as K and V are not defined.

This should be resolved in eclipse.jdt.ui - https://github.com/eclipse-jdt/eclipse.jdt.ui/blob/a4c1b5bc5dbb294e8c23b93296effeb1d5c2cae0/org.eclipse.jdt.ui/ui/org/eclipse/jdt/internal/ui/text/java/JavaMethodCompletionProposal.java#L474

Looks to still be an issue for me with this PR on latest master of jdt.core and latest jdt.ui.

I guess JavaMethodCompletionProposal is good, but perhaps somewhere in the implementation of canUseDiamon() this situation is handled wrongly. In particular InternalExtendedCompletionContext.canUseDiamond(String[], char[][]) from https://bugs.eclipse.org/578969 has a suspicious comment:

// If no LHS or return type expected, then we can safely use diamond

No LHS is actually the situation mentioned by @jukzi , and for this very reason a diamond is illegal.

[datho7561]

Roland sent me to this PR, since I took a look at another PR related to generics. Taking a look at this, it seems like there are two possible ways forward for the completion Stephan mentioned:

• substitute the erasure of the type variables for the type variables (eg. HashMap<Object, Object> in this case)
• use the raw type (i.e. HashMap with no diamond operator and no type arguments passed)

However, from my understanding, these fixes to completion seem unrelated to the changes in this PR. Is there anything blocking this PR from being merged?