Code Diagnostics

While most end attributes of the semantic graph aim to infer specific semantic facts about particular syntax tree nodes, code diagnostics (semantic errors and warnings) are intended to be collected from the entire syntax tree.

To tackle this issue and improve the incremental nature of code diagnostics, you can gather local diagnostic messages within scopes by iterating through scope nodes and their attributes, potentially containing diagnostic issues. These issues can then be collected into the hash set of the scope's diagnostics attribute.

Subsequently, in the root node's global diagnostics attribute, you can iterate through all local diagnostic attributes of scopes and aggregate their values into a single set, wrapping it into a Shared structure for efficient cloning. Furthermore, you can enhance the final diagnostics set with syntax errors from the normal compilation unit by directly reading them from the document¹.

The resulting global diagnostics attribute would indirectly depend on the majority of the semantic graph. Despite potential optimizations by the validator due to granularity, querying this attribute could still be computationally intensive in edge cases. To mitigate this, the language server could periodically examine this attribute with a low-priority analysis task.

Moreover, when utilizing the Attr::snapshot function to retrieve a copy of the current diagnostics sets, you can leverage the version number of the attribute value to determine whether this set needs to be republished to the client.