Partially fix bug in type resolution of paths #3277
Conversation
| nr_ctx.lookup (ast_node_id).map ([&ref_node_id] (NodeId resolved) { | ||
| ref_node_id = resolved; | ||
| }); | ||
|
|
||
| // HACK: on failure, redo lookup using path node id | ||
| // This is incorrect, but will help some tests pass for now | ||
| // TODO: remove this | ||
| if (ref_node_id == UNKNOWN_NODEID) | ||
| nr_ctx.lookup (expr.get_mappings ().get_nodeid ()) | ||
| .map ( | ||
| [&ref_node_id] (NodeId resolved) { ref_node_id = resolved; }); |
There was a problem hiding this comment.
I think I would rather spend some time fixing the root cause for this. This seems a bit dodgy. what do you think is missing from nr2.0 in order to enable this? simply resolving each segment when we resolve a path?
There was a problem hiding this comment.
do your lookup on the ast_node_it which is the segment not the path name resolution cant resolve a full path on its own.
| .map ([&ref_node_id, &path] (NodeId resolved) { | ||
| ref_node_id = resolved; | ||
| }); | ||
| nr_ctx.lookup (ast_node_id).map ([&ref_node_id] (NodeId resolved) { |
There was a problem hiding this comment.
you need to be looking at the segment node_id.
The name resolution for Paths is a really huge pain in the ass. this is resolving the ROOT of a path so for example you could have:
mod_a::Type::impl_item
the root path should just resolve this:
mod_a::Foo then stop then fall back to the resolve segments to then find the impl_item
So i think you should just be using the ast_node_id like i do in the old name resolver here.
gcc/rust/ChangeLog: * typecheck/rust-hir-type-check-path.cc (TypeCheckExpr::resolve_root_path): Use segment node ids instead of the path node id to look up segment resolutions in the 2.0 resolver, like is done with the 1.0 resolver, when possible. * typecheck/rust-hir-type-check-type.cc (TypeCheckType::resolve_root_path): Likewise, also remove unnecessary capture of path in lambda. Signed-off-by: Owen Avery <[email protected]>
When name resolution 2.0 is enabled these sections of the typechecker should be performing resolution lookups using segment node ids, rather than path node ids, as is done when name resolution 1.0 is enabled. Since name resolution 2.0 doesn't currently insert resolutions for individual segments, lookups based on path node id are left as a fallback.