Make error-handling more testable
e.g. right now we test that skipped/missing tokens are properly included in the tree, but we have no tests for the post-parse getDiagnostics walk.
Thoughts?
I am currently playing around with this.
What I have so far is similar to the mechanism by which the generated AST is tested:
For each test case X.php there is an accompanying file X.php.diag that contains the JSON-serialized output of the DiagnosticsProvider.
Are localized error messages planned? IMHO it's not that important to ensure that error messages stay the exact same between test runs. Maybe a "system-readable" l10n setting could provide stable messages between test runs.
What I have so far is similar to the mechanism by which the generated AST is tested:
That sounds like the right approach
Are localized error messages planned?
Definitely not in the near future at least.