Rust: extract source files of dependencies

[aibaars]

No description provided.

[redsun82]

I cannot shake the feeling this is a bit a roundabout way of doing it: what I mean, is reconstructing the context around the AstNode by visiting the parent, when the emission was requested by emitting the parent already.

I wonder if we shouldn't do this by generating a new exclusion point where the emission of the body takes place. What about having this in the template for ast node emission:

    pub(crate) fn emit_{{snake_case_name}}(&mut self, node: &ast::{{ast_name}}) -> Option<Label<generated::{{name}}>> {
        pre_emit!({{name}}, self, node);
        {{#has_attrs}}
        if self.should_be_excluded(node) { return None; }
        {{/has_attrs}}
        {{#fields}}
        let {{name}} =
            {{#is_body}}
            if self.should_skip_bodies() { None } else {
            {{/is_body}}
            {{#predicate}}
            node.{{method}}().is_some()
            {{/predicate}}
            {{#string}}
            node.try_get_text()
            {{/string}}
            {{#list}}
            node.{{method}}().filter_map(|x| self.emit_{{snake_case_ty}}(&x)).collect()
            {{/list}}
            {{#optional}}
            node.{{method}}().and_then(|x| self.emit_{{snake_case_ty}}(&x))
            {{/optional}}
            {{#is_body}}
            }
            {{/is_body}}
            ;
        {{/fields}}
        ...

(and adding is_body: field.name == "body", to the FieldType::Optional(ty) => ExtractorNodeFieldInfo creation). This will cover all possible bodies, and avoid their emission at the sources without working up the AST. For example:

        let body = if self.should_skip_bodies() {
            None
        } else {
            node.body().and_then(|x| self.emit_expr(&x))
        };

[redsun82]

ah, the above does not also skip pat, but something could be done there at generation level as well

[redsun82]

here's what I mean: #19559 shifts the logic from climbing up the AST in the base extractor to generating the property skipping at field emission level, with the logic in the ast-generator instead.

[redsun82]

LGTM, but we do need a DCA run on this before merging. #19559 can be considered as separate follow-up.

[redsun82]

Happy to merge this if we're satisfied by DCA