RFR: 8294020: improve errors for record declarations

[Rémi Forax]

On Thu, 3 Nov 2022 02:39:07 GMT, Vicente Romero <vromero at openjdk.org> wrote:

> Although the reporter originally complained about the error message for records with no header, I think the issue is deeper. We intentionally didn't follow the same path for parsing record declarations as we do for, for example, classes. This is mainly because `class` is a keyword but `record` is a contextual keyword. So when we find `record` we are not sure if it is an identifier or a record declaration. Although I think that given a context where the compiler expects a type declaration, we can be more aggressive than before and if we find `record` + `identifier` consider it a record declaration. The current implementation of `JavacParser::isRecordStart` is trying to be too clever but it is actually leaving several cases uncovered. So the proposed simpler version should be more stable and make record related errors more similar to those for other class declarations. Test `RecordDeclarationSyntaxTest.java` has been added just to have a golden file that stores the error position.
> 
> TIA

There is also the dual problem, students are writing *classes* with parenthesis after the name and the error message is not obvious too

FooNotARecord.java:1: error: '{' expected
public class FooNotARecord(String s) {
                          ^
1 error

I believe in both cases, the error message should mention that a record requires parenthesis (the word record should be in the error message).

By example here is the error message reported by IntelliJ when a class is declared with parenthesis
"record header declared for non record".

-------------

PR: https://git.openjdk.org/jdk/pull/10963