Rule names can collide with Tatsu internals and raise SemanticError

[EGrunzke]

Summary

The generated ModelBuilderSemantics class raises a tatsu.exceptions.SemanticError: Could not call constructor for Sequence: 'closure' object is not callable

Context

Some cases work:

• tatsu.compile(grammar).parse("(test)")
• BugParser().parse("(test)")

But BugParser().parse("(test)", semantics=BugModelBuilderSemantics()) fails

The grammar

# bug.peg
@@grammar :: Bug

start = sequence $ ;

sequence::Sequence = children:{term [',']}+ ;

term = parens | word ;

parens = '(' ~ @:sequence ')' ;

word::Word = val:/\w+/ ;

Reproduction

Reproduction

1. Generate the parser and model:
   python -m tatsu bug.peg -o bug_parser.py -G bug_model.py

2. Run

parser = BugParser()
semantics = BugModelBuilderSemantics()
parser.parse("(test)", semantics=semantics)

GitHub repo with full code

Full reproduction code is available:
https://github.com/EGrunzke/tatsu-sematic-error-closure-not-callable

[EGrunzkeCvg]

I've kept looking at this and I think the problem is a naming collision. My Sequence rule has a name children. This collides with the internal Node object that Sequence extends.

https://github.com/neogeny/TatSu/blob/master/tatsu/objectmodel.py#L145

The solution is to either consider children a reserved word (and append an underscore) or to change the variable name in Node to be something safer, like _children, though that could still collide.

As a workaround, switching from
sequence::Sequence = children:{term [',']}+ ; to
sequence::Sequence = kids:{term [',']}+ ;
works great!

[apalala]

Ah! Collisions!

I need to think about how to get rid of those in the next version of TatSu. Most of the collisions come from trying to make AST and Node provide access to fields through both __getitem__() and __getattr__(). Most grammars should be able to get away with AST nodes that don't have a base class (@dataclass perhaps?).

[apalala]

Hello @EGrunzke ! Do we still need to do something about this issue?

[EGrunzke]

I am still able to reproduce using the grammar from the original post. I used Tatsu 5.6.1

This isn't much of a blocker, as a simple rename of the rule solves the problem. But it does still exist!