context-free sequence containing a nullable fails on empty

[jurgenvinju]

Describe the bug

Noticed and isolated by @rodinaarssen

rascal>syntax X = ("(" () ")");
ok
rascal>[X] "()"
|TODO:///|: ParseError(|prompt:///|(0,0,<1,0>,<1,0>))
rascal>

The issue goes away if we remove the (...) sequence wrapper. It stays for any other nullable nonterminal replaced for the () empty non-terminal.

Wrote a failing "Sequence4" test for this: https://github.com/usethesource/rascal/pull/2214

[jurgenvinju]

If we add spaces the issue goes away. So it seems this has to do with sequence members being empty:

rascal>syntax X = ("(" () ")");
ok
rascal>layout L = [\ ]*;
ok
rascal>[X] "(  )"
X: (X) `(  )`
rascal>[X] "( )"
X: (X) `( )`
rascal>[X] "()"
|TODO:///|: ParseError(|prompt:///|(0,0,<1,0>,<1,0>))

[jurgenvinju]

In this case we have 5 sequence members:

1. "("
2. layout
3. ()
4. layout
5. ")"

With only one space given, the parse is ambiguous as it should be:

rascal>/amb(_) := [X] "( )"
bool: true

With two spaces it's the same. Only when no spaces are given, the parse fails erroneously.

[jurgenvinju]

No follow restrictions or other filters are in play here. I have removed the generated syntax for meta holes from the Sequence4 test, for simplicity's sake.

[arnoldlankamp]

This one is pretty easy to explain. Empty sequences aren't currently supported.

See: https://github.com/usethesource/rascal/blob/d4cb8063195343cdb3de0039067e0f886e9cc9c9/src/org/rascalmpl/parser/gtd/stack/SequenceStackNode.java#L82

Adding support may just be as easy as changing the return value of the canBeEmpty method to true in case the children array has zero length 😉 . The resulting tree should then contain an epsilon, like with optionals and star lists.

The same thing goes for alternatives, in case you'd want to update it's semantics as well: https://github.com/usethesource/rascal/blob/d4cb8063195343cdb3de0039067e0f886e9cc9c9/src/org/rascalmpl/parser/gtd/stack/AlternativeStackNode.java#L84