Parse tree rendering fails in the presence of rules matching an empty string

[MDoerner]

I have the issue mentioned in the title in the current version in the VS Marketplace (2.0.4).

MCVE:

grammar:

grammar TreeVisualizationFail;

LETTER : [a-zA-Z];
DIGIT : [0-9];

 
startRule : digits letters EOF;

letters : LETTER+;

digits : DIGIT*;

input:

foo

Note that for the given input the mandatory rule digits in startRule matches no tokens, i.e. an empty string.

After making digits optional in startRule and forcing digits to match at least one digit, the parse tree visualization gets generated.

Originally, I encountered this problem with a grammar used to parse VBA.

[mike-lischke]

That's kinda duplicate of #44. The cause is the same. The underlying antlr4ts library tries to optimize the ATN from this grammar and fails.

However, matching an empty input string is a problem anyway and must be avoided. ANTLR4 will issue a warning for that.

[Vogel612]

FWIW I would not consider a rule that may match empty string and a warning (note: not an error) generated by the ANTLR runtime to be sufficient reason to expect no parse tree at all.

IIUC the matching behaviour is also subtly different. When examining the output generated by startRule in the minimal example, accessing the generated digits will never be null (interesting for languages such as C# or java), but instead will always generate a DigitsContext node.

While that context does not necessarily have any children, it makes for a different API

[MDoerner]

Just as a small note: Antlr only generates a warning for a rule matching an empty string if this rule is made optional in another rule using it. (In that case, the outcome is non-deterministic.)

[mike-lischke]

I see. Didn't think of that possibility (why would one want a node for no match?), but it seems people have their own use cases. Thanks.