parser: Update error-recovery mode implemented in gocc

[shivansh]

The following steps are performed when an invalid input symbol is encountered -

• In case the input reaches an invalid symbol, we save its error attribute and keep discarding all the immediately following error symbols until a valid symbol is found.

• If the action corresponding to the found valid symbol is reduce, it is performed. In case it is a shift, it is deferred until the saved error attribute is pushed on the stack.

• Once the saved error attribute is pushed on stack, parsing can resume normally.

• It should be noted that the current model cannot report multiple errors for invalid symbols occurring one after the other. If such a case arises, the error for only the first invalid symbol is reported. This is also the case with the currently implemented error recovery model in gocc.

Demo for error-recovery example

=== RUN   TestFail
input: a ; b
output: [
	a
Error in S0: INVALID(0,;), Pos(offset=2, line=1, column=3), expected one of: $ id error 
	b
]
--- PASS: TestFail (0.00s)
PASS
ok  	github.com/goccmack/gocc/example/errorrecovery	0.002s

input: a ; ; b
output: [
	a
Error in S0: INVALID(0,;), Pos(offset=2, line=1, column=3), expected one of: $ id error 
	b
]
--- PASS: TestFail (0.00s)
PASS
ok  	github.com/goccmack/gocc/example/errorrecovery	0.002s

input: a ; b ; c
output: [
	a
Error in S0: INVALID(0,;), Pos(offset=2, line=1, column=3), expected one of: $ id error 
	b
Error in S0: INVALID(0,;), Pos(offset=6, line=1, column=6), expected one of: $ id error 
	c
]
--- PASS: TestFail (0.00s)
PASS
ok  	github.com/goccmack/gocc/example/errorrecovery	0.002s

It should be noted that there might be scope of further refactoring the code in this PR. I'll defer doing it until the proposed heuristic has been verified.

[awalterschulze]

Am I understanding correctly that this is an attempt to fix https://github.com/goccmack/gocc/issues/76

I wouldn't call this a heuristic, but rather just a possibly better way to do error recovery.

I am sick though, so I probably missed something.

@goccmack this is probably something worth checking out. What do you think?

[shivansh]

Am I understanding correctly that this is an attempt to fix #76

Yes.

I wouldn't call this a heuristic, but rather just a possibly better way to do error recovery.

oic, updating the title and PR message.

I am sick though, so I probably missed something.

Get well soon :)

[goccmack]

@shivansh put his finger on one of the aspects of gocc that didn't receive sufficient attention originally. If I remember correctly I tried to implement error recovery according to the dragon book "Error Recovery in LR Parsing" but under time pressure. I think @shivansh is on the right track.