De-duplicate error spam.

[lawnjelly]

Prevent large amounts of duplicate error spam reaching the MessageQueue.

Helps address #84713 .

Notes

• Simple solution for error spam within the same frame for discussion.
• Compiled out when compiled with TESTS_ENABLED, because tests rely on duplicate errors being individually reported.
• Could alternatively runtime check whether the tests are being run.
• Toaster might deal with duplicates, but is preferable in some ways to prevent this before spam reaches the MessageQueue, so not to fill it up.
• This won't deal with duplication of 2 or more error messages.
• Could possibly pass the number of duplicates in the error message if desired (with some cunning), perhaps to the toaster so it could correctly report number of duplicates without straining MessageQueue.
• Uses mutex just in case of multithread calls to _err_print_error(). Could possibly just lock around the whole function.

[Calinou]

This makes sense, but I suggest documenting this in the ERR_ and WARN_ macros that if you want several errors from the same function in the same frame to be printed, they must differentiate themselves in some way (e.g. by inserting some context about function parameters, etc). This is good practice in general.

Also consider that the editor Output panel can collapse lines already, although this is disabled by default: #84499

[lawnjelly]

This makes sense, but I suggest documenting this in the ERR_ and WARN_ macros that if you want several errors from the same function in the same frame to be printed, they must differentiate themselves in some way (e.g. by inserting some context).

Good idea I'll do this. 👍
(Although might be subject to change if we think of variations of how to do the de-duplication.)

Also consider that the editor Output panel can collapse lines already, although this is disabled by default: #84499

Yes, the problem here is though that that doesn't prevent the spam entering the MessageQueue, which can fill it up and lock up the editor.

UPDATE:
Ah see asan is failing. Might be to do with the static lifetime of CharString and COW etc. I'll have a look in the morning.

[lawnjelly]

Intriguingly, the tests failing seems not to be to do with the static lifetimes or lack of mutex, they are failing because l->errfunc() is not called in the case of a duplicate.

Maybe the tests are counting the number of errors or something... Still investigating.

Yes:

	SUBCASE("set_type_variation") {
		for (const StringName &name : names) {
			Ref<Theme> theme = memnew(Theme);

			ErrorDetector ed;
			theme->set_type_variation(valid_type_name, name);
			CHECK(ed.has_error);
		}

Fails here, as there is no error, as the test is presumably duplicating the previous error. Maybe the solution is simple - to turn off deduplication during tests. 🤷‍♂️