Change ParseOutput to return a NonTerminal instead of a Node
#1187
Conversation
|
beta-ziliani
force-pushed
the
beta/parse_result
branch
from
06692b6 to
95137a0
Compare
| pub(crate) errors: Vec<ParseError>, | ||
| } | ||
|
|
||
| impl ParseOutput { | ||
| pub fn tree(&self) -> Node { | ||
| self.parse_tree.clone() | ||
| pub fn tree(&self) -> Rc<NonterminalNode> { |
There was a problem hiding this comment.
TOL: I wonder why cloning by default? WDYT of returning &Rc<> and leave cloning up to the user? This is similar to other APIs like Node::as_nonterminal().
I realize that Rc::clone() is verbose, but we decided to enable a lint rule that enforces that for clarity (we can revisit this decision separately).
| ParseOutput { | ||
| parse_tree: tree, | ||
| parse_tree: Rc::new(NonterminalNode::new(no_match.kind.unwrap(), trivia_nodes)), |
There was a problem hiding this comment.
if we are always unwrapping no_match.kind, should it still be an Option<>?
There was a problem hiding this comment.
NoMatch is being used in the default() function with None
There was a problem hiding this comment.
Here is an example of how it's used today:
There was a problem hiding this comment.
IIUC, in the case you mentioned, the kind is already known (nonterminal_name). Right? Is it possible to initialize it with that expected kind, to prevent another code path from accidentally leaving a None here?
Solves #1184
To go on top of #1172
This has a big impact in the failing examples, which now return the expected
NonTerminal.This was acknowledgedly coded by playing the whack-a-mole with the type system. I'm not so happy with the bunch of
Rc::clone()that leaked out; I will consider an alternative once we settle this is what we need and there are no more pressing issues.Currently doing a full CI run locally to find what more to fix. In the meantime, I'll mark this as draft.