[display] diagnostics miss errors from Haxe's post-processing

[Gama11]

class Main {
	static function main() {
		var s:String;
		s += "test";
	}
}

This fails to compile with Local variable s used without being initialized, but there's no diagnostics error for it.

Seems like this is not a regression / has always been this way.

[Simn]

Any error from our post-processing is missed because we don't run it for diagnostics.

We can explicitly run the var init check and null-safety in diagnostics mode. They are early filters, so it should be fine.

[Simn]

By the way, do we have diagnostics tests now?

[Gama11]

Yep, although just a single one that checks that there are no diagnostics. So no hasDiagnostics() helper function or anything like that yet.

b20cb70#diff-d56e6b80704188e7ea0e5d667a79876b

[Simn]

Not easy to fix because we don't have access to a good error function in post-processing. This is going to require some refactoring.

[Gama11]

I assume "Type name should start with an uppercase letter" is one of these too?

[Simn]

No, that's a parser error. But it might be a hard error still.

[Simn]

I kinda want to solve this for 4.1, but it's not so easy. I think we can do something like try check_local_vars_init with Error _ (* turn into diagnostics *) in the diagnostics preparation step. That way we at least get one error per method.

However, the more general problem here is there is various post-processing that could cause errors and we don't want to run all of that for diagnostics. Unfortunately, this then means that diagnostics will never 100% match compilation errors.

[Gama11]

However, the more general problem here is there is various post-processing that could cause errors and we don't want to run all of that for diagnostics.

We don't? What's an example of something that would be too expensive to run?

[Simn]

The analyzer for sure, but this is more about the sum of the parts. Our post-processing is not designed to just detect errors, it's a huge AST transformation framework which just happens to spit out some errors in the process.

We should go through the individual steps and check which ones might report user-facing errors (not the assert false kind of stuff that comes from internal problems). Then we can maybe look into isolating these out and run them as part of diagnostics.

[Simn]

This has been addressed by #11220. I'll go through the filters and see that I add some tests.

[Simn]

Adding sane tests for this is kind of blocked by #9134.

[RblSb]

Filters like null safety not yet here, this or #7933 should be reopened?

[kLabz]

Filters like null safety not yet here, this or #7933 should be reopened?

Is that still a thing ? Do you have a way to repro ?

[RblSb]

I think it is, see this example, but for diagnostics check
vshaxe/vshaxe#509

[kLabz]

Indeed, thanks!

[kLabz]

Interestingly, it reaches the error but the reporting isn't working properly.