Completion test0123 v2 #1240
Open
Completion test0123 v2 #1240
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Change dom conversion logic to recover incomplete type names properly - Restrict the suggested classes based on if they are a class or interface - Do not suggest final classes - Do not suggest the current class Signed-off-by: David Thompson <davthomp@redhat.com>
Makes changes to what fields in completion results we set and what we set them to, with the goal of reducing test failures Signed-off-by: David Thompson <davthomp@redhat.com>
IProblem.UnsafeReturnTypeOverride seems the correct one.
These are compiler.note.removal.filename, compiler.note.deprecated.plural.additional and the fact that they don't have position info in them makes them not suitable to create Problems for them.
- recover `new ` in AST converter - some fixes in bindings to get some test cases working - scan backwards for first non-whitespace character when performing AST search - use type hierachy to build a list of potential constructors - This method is expensive but accurate - suggest creating anonymous classes when completing constructors for interfaces - always provide all constructors of current type to be bug compatible with existing CompletionEngine - use diamond operators Limitations: - relevance numbers are not quite right yet - generic types don't quite work as expected Signed-off-by: David Thompson <davthomp@redhat.com>
Make sure that calling substring is done with acceptable values, values bigger than string length or negative ones are plain wrong. ``` !ENTRY org.eclipse.jdt.core 4 0 2024-10-31 11:45:52.859 !MESSAGE Failed to convert Javadoc !STACK 0 java.lang.StringIndexOutOfBoundsException: Index 2912 out of bounds for length 1736 ```
eg.
```java
import java.util.List;
/**
* Like a {@link List} if it wasn't a list.
*/
public class NotAList {
}
```
The `java.util.List` import is needed to distiguish the Javadoc mention
of `List` from other potential `List` classes.
Signed-off-by: David Thompson <davthomp@redhat.com>
- set prefix for `this.t|` completions correctly
- improve extends and implements completion with keywords
eg. avoid generic completion results and instead recommend keywords here:
```java
public class MyClass | {
// ...
}
```
Signed-off-by: David Thompson <davthomp@redhat.com>
- Fix completion issue with `myVariable.|\nMyObj asdf = null;` - Fix for `myTarget.|` completion Signed-off-by: David Thompson <davthomp@redhat.com>
Reduce the parsed AST before resolving to get a performance boost.
The following cases are fixed: * Interface methods * Abstract methods * Methods marked with Override annotation
- `myMethod().|` - completion after method declaration in `@interface` Signed-off-by: David Thompson <davthomp@redhat.com>
Recover the prefix in this case:
```java
public class MyClass ext| {
}
```
So that we know to only suggest `extends` and not `implements`.
Also, do not suggest `implements` for interfaces.
Signed-off-by: David Thompson <davthomp@redhat.com>
- fix duplicates in annotation members - fix range for annotation members - fix type of annotation member completions - opening completion in a normal annotation or single value annotation before `(`, eg `@MyAnnota|tion(value = 1)`, now tries to complete the annotation name rather than the annotation members Fixes eclipse-jdt#930 Signed-off-by: David Thompson <davthomp@redhat.com>
- Javadoc will need to be revisited, but for now I turned off the default completion Signed-off-by: David Thompson <davthomp@redhat.com>
Signed-off-by: David Thompson <davthomp@redhat.com>
- Do not reuse the context when compiling to multiple output directories - Implement Alex's idea of a separate file of javac-specific error codes Signed-off-by: David Thompson <davthomp@redhat.com>
work-in-progress: - prefix filtering - see if I can get the binding key bugs resolved Fixes eclipse-jdt#938 Signed-off-by: David Thompson <davthomp@redhat.com>
In order to accomplish this, I have changed the logic for adjusting the offset used for the node search. This causes some regressions: - A few cases where the test expects the AST not to be recovered but javac does recover it (eg. org.eclipse.jdt.core.tests.model.CompletionTests_1_5.test0197) - @| in Javadoc is not handled. Before it got no completion due to not finding the correct node, but now it gets default completion, which causes some regressions. I'll fix these when dealing with @| properly - The test cases sometimes expect the current class to be suggested in new | completion even when the type doesn't match, but sometimes they don't. I'll need to investigate further Signed-off-by: David Thompson <davthomp@redhat.com>
When resolving, we do only need the symbols/bindings of depedencies. Minimize the amount of option to skip eg parsing Javadoc for transitive files that are only relevant as part of binding resolution.
- Both block and inline tags - Take into account which ones are applicable to the declaration being Javadocumented Fixes eclipse-jdt#948 Signed-off-by: David Thompson <davthomp@redhat.com>
as it's reused on the whole lifecyle of JDT.
* Also do not share the Platform filemanager with build (a bit more isolation) * a few other optimization to save CPU on irrelevant operations
Because resolving from .class deps is much faster than resolving from .java deps
It's most likely useless to get linting when none of those are requested.
Should fix 6 cases without regressions, because one of the projects used to test completion has two classes called `SuperClass` in the default package. Signed-off-by: David Thompson <davthomp@redhat.com>
eg.
```java
public class MyClass extends MyOtherClass {
public MyClass() {
sup|
}
}
class MyOtherClass {
public MyOtherClass(int i) {}
protected MyOtherClass(Object i) {}
MyOtherClass(boolean b) {}
private MyOtherClass(String s) {}
}
```
Requires eclipse-jdt#1206 in order to work properly if you have substring matching
enabled in your IDE.
Fixes eclipse-jdt#1205
Signed-off-by: David Thompson <davthomp@redhat.com>
- Do prefix filtering first; string comparison should be fast compared to more involved operations - Fix bad implementation of `protected` member classes; `getParent()` wasn't the right method to get the parent class of a member class - Impossible classes filtering was wrong, it was checking methods instead Signed-off-by: David Thompson <davthomp@redhat.com>
Fixes: `ASTConverterTest.test0113` `ASTConverterTest.test0114` `ASTConverterTest.test0115` as well as tests for other Java versions Note: we only need to accomodate Java >= 8 in the AST conversion process, since that's what upstream does now. Signed-off-by: David Thompson <davthomp@redhat.com>
Should fix 17 test cases Signed-off-by: David Thompson <davthomp@redhat.com>
eg. `MyType\u005b] myTypeArray = null;` Fixes 4 tests: `ASTConverterTest.test0266` and the versions for other Java compliances Signed-off-by: David Thompson <davthomp@redhat.com>
`ThisExpression` is it's own node, which we will need to handle differently. Should fix 2 test cases. Signed-off-by: David Thompson <davthomp@redhat.com>
AST no longer supports it thus no need for Javac converter to be complicated for nothing.
It's now taken into account for members of a qualified type being completed. Signed-off-by: David Thompson <davthomp@redhat.com>
Failback to type of declaration from Search if not resolved
Not needed as neither Javac nor ECJ nor JDT makes use of anything pre Java 8 nowadays. VariableDeclaration.extraArrayDeclarataion is not supported in Java 8+ but was used without guards which shouldn't have happened.
Fixes eclipse-jdt#1224 Signed-off-by: David Thompson <davthomp@redhat.com>
- Fix the number part of IVariableBinding's key
- This means that the bindings for different variables with the same
name in the same method can be distinguished eg. `rob` and `i` can
be distinguished:
```java
public class Main {
public void myMethod() {
for (int i = 0; i < 10; i++) {
int rob = 12;
}
for (String i = "asdf"; "hjkl".equals(i); i = "hjkl") {
String rob = "rob";
}
}
}
```
- As a result, completion works deterministically in these cases
(eg. before you just had to hope that the correct binding was cached
in order to get the right results)
Signed-off-by: David Thompson <davthomp@redhat.com>
Signed-off-by: Rob Stryker <rob@oxbeef.net>
Extra strategy in MatchLocator to use DOM, and tweaks in various Locators to properly handle DOM. Tests in JavaSearchTests help a lot (currently ~130/270 passing) It basically clones and adapt some methods and adapt them to DOM, A new flag is introduced to decide the strategy. Done-ish (often with remaining issues) * FieldLocator * LocalVariableLocator * MethodLocator * SuperTypeReferenceLocator * TypeDeclarationLocator * TypeParameterLocator * TypeReferenceLocator TODO: * AndLocator * ModuleLocator * PackageDeclarationLocator * PackageReferenceLocator * OrLocator * VariableLocator Make some match locator work with DOM Fix NPE, convert 4 errors to fails Signed-off-by: Rob Stryker <stryker@redhat.com> Handle match files in the order they arrived Signed-off-by: Rob Stryker <stryker@redhat.com> Fix testBug251827b and 5 others - import declarations java element must be found Signed-off-by: Rob Stryker <stryker@redhat.com> Move the DOM-based search into the javac bundle Signed-off-by: Rob Stryker <rob@oxbeef.net> Cleanup Signed-off-by: Rob Stryker <stryker@redhat.com> Move the DOM-based search into the javac bundle Signed-off-by: Rob Stryker <stryker@redhat.com>
- precalculate extends or implements info
- When there is no prefix, suggest types declared in the same class,
eg. suggest `MyInterface` in the following snippet
```java
public class MyClass implements | {
}
interface MyInterface {}
```
- Better recovery of bindings for classes with incomplete extends or
implements sections
- previously the binding key was always `*`, which caused many that
relied on the binding key to break
- Use ExtendsOrImplements info filtering in more places
Signed-off-by: David Thompson <davthomp@redhat.com>
Make it capable of resolving *any* type visible to this build.
resolveWellKnownType method specifies a minimum set of supported types, but the list is not meant to be exclusive and the definition of "well-known" type is not so constraining. So we allow test to pass even if a binding resolve can resolve java.lang.Runnable as a well-known type.
+ do not add import for fully qualified types
Signed-off-by: Rob Stryker <stryker@redhat.com>
Signed-off-by: Rob Stryker <stryker@redhat.com>
robstryker
force-pushed
the
dom-with-javac
branch
from
1bb87f3 to
197281f
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Can't figure out how to re-open #1109
So made a new PR instead