completion returns no results with pattern matching

[snjeza]

What it does

Fixes #2106, https://bugs.eclipse.org/bugs/show_bug.cgi?id=568934#c8 and https://bugs.eclipse.org/bugs/show_bug.cgi?id=568934#c6

How to test

try CA at |

import java.util.ArrayList;
import java.util.List;

public class SimpleJavaClass {
  public static void main(String[] args) {
    Object unknown = new ArrayList<>();

    if (unknown instanceof List things) {
      things.is| // try CA here
      Object nothing = null;
      things.is|
    }
  }
}

Without the patch:

With the patch:

I will add a test.

[snjeza]

@srikanth-sankaran @iloveeclipse I have updated the PR.

the fix proposed in #2107 is suspect. You should never have to fiddle with ExtraCompilerModifiers.AccOutOfFlowScope bits.

I left this part out. The PR doesn't touch the bits field.

See also https://bugs.eclipse.org/bugs/show_bug.cgi?id=568934

It also fixes https://bugs.eclipse.org/bugs/show_bug.cgi?id=568934#c8 and https://bugs.eclipse.org/bugs/show_bug.cgi?id=568934#c6

If you agree, I can add a test.

[srikanth-sankaran]

No, this fix is not correct. It is kludgy and may work but I can't predict what other problems it may cause down the road.

Basically you are pairing a LocalVariableBinding where a TypeBinding would be the appropriate one. That is recipe for CCE. Even if it doesn't trigger CCE in testing, do we want to resort to such hacky fixes is the question.

The real problem as I outlined on the bug earlier is due to wrong parse tree recovery. Can you check why the scanner doesn't declare EOF upon seeing the completion at things.is| and why it returns the token Object at all ?

As you can see the two statements get jammed into one

things.is| Object

so it appears as though you are completing on a nested type, while you are actually completing on a QNR.

[snjeza]

and the parse tree recovered in the non-working case is:

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass {
  String things;
  public SimpleJavaClass() {
  }
  public static void main(String[] args) {
    Object unknown;
    {
      if ((unknown instanceof List things))
          <CompleteOnType:things.is> Object;
    }
  }
}

I have tried it and I get:

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass {
  public SimpleJavaClass() {
  }
  public static void main(String[] args) {
    Object unknown;
    {
      <CompleteOnType:things.is> Object;
    }
  }
}

There isn't if ((unknown instanceof List things))

Completion is being attempted on a Name in both instances. Why do we think it is being attempted on a Type ? I can help with reviewing an alternate fix. Thanks
The real problem as I outlined on the bug earlier is due to wrong parse tree recovery. Can you check why the scanner doesn't declare EOF upon seeing the completion at things.is| and why it returns the token Object at all ?
The parser returns CompletionOnQualifiedNameReference in line 10 (it works) and CompletionOnQualifiedTypeReference in line 8 (it doesn't work)
Right, that is the bug that needs fixing.

I have checked the following code:

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass2 {
  public static void main(String[] args) {
    Object unknown = new ArrayList<>();
    List list = new ArrayList<>();
    if (unknown instanceof List things) {
      list.is|
      Object nothing = null;
      list.is
    }
    
  }
}

the parse tree recovered is

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass2 {
  public SimpleJavaClass2() {
  }
  public static void main(String[] args) {
    Object unknown;
    List list;
    {
      <CompleteOnType:list.is> Object;
    }
  }
}

The parser also returns CompletionOnQualifiedTypeReference. It works because the parser finds the list variable.

Based on it and the fix for https://bugs.eclipse.org/bugs/show_bug.cgi?id=287939 I have created a new PR - 970acaf

Basically you are pairing a LocalVariableBinding where a TypeBinding would be the appropriate one. That is recipe for CCE. Even if it doesn't trigger CCE in testing, do we want to resort to such hacky fixes is the question.

This part is removed.

The recovered parse tree with the PR is:

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass {
  public SimpleJavaClass() {
  }
  public static void main(String[] args) {
    Object unknown;
    {
      List things;
      if ((unknown instanceof List things))
          <CompleteOnType:things.is> Object;
    }
  }
}

@srikanth-sankaran Could you, please, review.

[snjeza]

The completion with the PR

[snjeza]

I have updated the PR. It works similar to CA in

import java.util.ArrayList;
import java.util.List;
public class SimpleJavaClass2 {
  public static void main(String[] args) {
    Object unknown = new ArrayList<>();
    List list = new ArrayList<>();
    if (unknown instanceof List things) {
      list.is|
      Object nothing = null;
      list.is
    }
    
  }
}

I have also added some tests.