In crates/assembly-syntax/src/sema/passes/toposort.rs, the toposort implementation uses nested loops that result in O(n²) or worse complexity for large call graphs. Specifically:

• The visitor pattern traverses each node
• For each node, it may scan the entire graph to find dependencies
• The cycle detection adds another layer of scanning

For projects with many procedures and complex call graphs, this can significantly slow down compilation.

Fix: replace with a standard Kahn's algorithm or DFS-based toposort that maintains O(V + E) complexity. Consider using a more efficient data structure (e.g., IndexSet or FxHashMap) for tracking visited nodes.