[wip] Improve parser error handling

[Simn]

This PR changes how we communicate with our parser. I acknowledge that it looks scary, but I think it's a necessary step and I would like to make that step now rather than later.

First of all, when we're not in display mode, the parser still raises Parser.Error early as it did before. However, when we're in display mode (including diagnostics), it returns an element of this ADT:

type 'a parse_result =
	(* Parsed display file. There can be errors. *)
	| ParseDisplayFile of 'a * parse_error list
	(* Parsed non-display-file without errors. *)
	| ParseSuccess of 'a
	(* Parsed non-display file with errors *)
	| ParseError of 'a * parse_error * parse_error list

It might still raise Parser.Error in some (rare?) cases. The endgame here is to eliminate all these cases so it always parses something. This is going to require some more refactoring though and is not something we have to do for 4.0.

The idea is to look at all occurrences of ParseDisplayFile and friends (there are only 14 at the moment) and adjust what we want to do. For instance, the compilation server naturally reacts like this:

| ParseError(_,_,_) -> "not cached, has parse error",true
| ParseDisplayFile _ -> "not cached, is display file",true
| ParseSuccess data ->"cached",false

This addresses #7793. Once we cleaned up error handling completely, we could also cache the display file if it has no errors (useful for hover-mode where the AST is often intact).

As another example --macro parsing does this:

| ParseSuccess data -> data
| ParseError(_,(msg,p),_) -> (Parser.error msg p)
| ParseDisplayFile _ -> assert false (* cannot happen *)

There are some PARSERTODOs I have to look at still, but I would like to know if people agree that this is the right way forward.

As a more or less unintended side-effect, we collect syntax errors on the diagnostics file which should allow us to address #5306.

[ncannasse]

Should we not save syntax_errors at beginning of parsing and restore them here as well ?

[Simn]

Indeed, good catch

[ncannasse]

that seems unrelated to the PR ;)

[Simn]

Something in the grammar is broken. It currently emits this error on every file:

source/Main.hx: 64-4611686018427387903: Expected }

[Simn]

There's a travis OS X display test failure:

Command: haxe [build.hxml]
utest/ui/text/PrintReport.hx:52: 
assertations: 1030
successes: 1029
errors: 1
failures: 0
warnings: 0
execution time: 128.564
results: SOME TESTS FAILURES (success: false)
cases.IImport
  testUsing1: ERROR ..E
    HaxeInvocationException
Called from local function #4 (src/cases/IImport.hx line 41 column 2)
Called from cases.IImport.testUsing1 (src/cases/IImport.hx line 44 column 20)
Called from DisplayTestContext.fields (src/DisplayTestContext.hx line 41 column 24)
Called from DisplayTestContext.callHaxe (src/DisplayTestContext.hx line 113 column 4)

I cannot reproduce this locally though:

Josefienes-MacBook-Pro:display Simon$ haxe --version
4.0.0-rc.1+67550f526
Josefienes-MacBook-Pro:display Simon$ haxe build.hxml 
utest/ui/text/PrintReport.hx:52: 
assertations: 1033
successes: 1033
errors: 0
failures: 0
warnings: 0
execution time: 47.97

results: ALL TESTS OK (success: true)

Any idea what could cause this?

[Simn]

Oh it's this shit again...

cases.Signature
  testBadAst: ERROR .....E
    HaxeInvocationException(Error: Invalid format: (����
, testBadAst, [-cp,src,-D,display-stdin,-lib,utest,--display,src/cases/Signature.hx@63@signature], package cases;
	class Some {
		function main() {
			test("foo"
		}
		static function test(a:String, b:Int) { }
	}
	])
Called from local function #2 (src/cases/Signature.hx line 31 column 2)
Called from cases.Signature.testBadAst (src/cases/Signature.hx line 33 column 37)
Called from DisplayTestContext.signature (src/DisplayTestContext.hx line 77 column 26)
Called from DisplayTestContext.callHaxe (src/DisplayTestContext.hx line 117 column 4)