fix: add error for dot syntax in record selection

[jaschdoc]

Updates the weeder with a helpful error message when using dot syntax with record selection.

Related to #7916
Related to #7897

[jaschdoc]

Hmm it seems the # symbol clashes with ## for java types when used in an expression. However, we do not resolve all uses of java types against imports. For instance, a catch-rule works fine with ##java.lang... since a record selection cannot occur in that position, however, 1ii instanceof ##java.math.BigInt is a syntax error since it is parsed as record selection.
The way I see it, we have two options:

1. We do not merge this weeder change just yet and allow parsing record.label, i.e., the dot syntax for records and simultaneously update the resolver to resolve imports everywhere a java type is used. Hopefully, feat: resolve java types against namespace imports in try-catch #7902 completely fixes this. We then merge this change.
2. We merge this change and feat: resolve java types against namespace imports in try-catch #7902 in one big PR... I think that becomes too big, though.

In summary, the first option silently allows the old record syntax and then we fix things in the background and finally disallow it, or we do everything in one pr.

[magnus-madsen]

In summary, the first option silently allows the old record syntax and then we fix things in the background and finally disallow it, or we do everything in one pr.

Lets do it incrementally.

[jaschdoc]

@magnus-madsen compiler works now. We "just" have breaking changes in the community build.

[jaschdoc]

Somehow, the IllegalRecordSelectSyntax error makes it all the way to Lowering which crashes the compiler. Where should this be handled?

EDIT: When I try this in VSCode everything works, i.e., compilation of a program with record.label fails but it does not crash the compiler. So apparently it only crashes in some tests... weird.

[magnus-madsen]

Somehow, the IllegalRecordSelectSyntax error makes it all the way to Lowering which crashes the compiler. Where should this be handled?

EDIT: When I try this in VSCode everything works, i.e., compilation of a program with record.label fails but it does not crash the compiler. So apparently it only crashes in some tests... weird.

Uh. Hmm. It should not if you return a soft failure.

[magnus-madsen]

Where do you see it crashing? I don't see it.

[magnus-madsen]

Could you split this PR into two: one with all the fixes to the syntax and one with the actual error. (because the latter PR is kind of blocked until community builds are updated).

[magnus-madsen]

What's the "status" on this?

[jaschdoc]

This is done. Only fails because of the community build where libraries still use the dot notation for records