Stack overflow on EOF*

[KvanTTT]

Consider the following grammar:

lexer grammar StackoverflowOnClosure;
ClosureEof: EOF*;

It fails with Stackoverflow on empty input:

at org.antlr.v4.runtime.misc.Array2DHashSet.getOrAdd(Array2DHashSet.java:59)
	at org.antlr.v4.runtime.atn.ATNConfigSet.add(ATNConfigSet.java:146)
	at org.antlr.v4.runtime.atn.ATNConfigSet.add(ATNConfigSet.java:122)
	at org.antlr.v4.runtime.atn.LexerATNSimulator.closure(LexerATNSimulator.java:446)
	at org.antlr.v4.runtime.atn.LexerATNSimulator.closure(LexerATNSimulator.java:455)
	at org.antlr.v4.runtime.atn.LexerATNSimulator.closure(LexerATNSimulator.java:455)

And it's freezing on any not empty input.

The samples is extracted from #1943 (comment)

[KvanTTT]

It seems like the bug is similar to the bug with EPSILON_CLOSURE that I've already fixed. But here we have EOF_CLOSRE instead of EPSILON_CLOSURE. Also, closure on EOF is useless.

So, I fixed it in a similar way.

[parrt]

Hmm..gotta be careful here. EOF is an actual condition/"char" that we sometimes wanna match. maybe not in * loop but certainly in a set

[KvanTTT]

It works fine with set. There is no error for such case:

EofInSet: 'y' ('z' | EOF);

But the following code where EOF closure actually exists (a sample from runtime tests):

prog : stat EOF;
stat : 'x' ('y' | EOF)*?;

Should be rewritten in the following way:

prog : stat EOF;
stat : 'x' 'y'* EOF?;

Anyway, stack overflow on runtime is much worse than restricting EOF closures on grammar level.

[parrt]

Should this be merged before or after your #3349 ?

[KvanTTT]

The PR fixes this bug :)