Jac LSP Re-Architecture

[kugesan1105]

Why need re-architecture??

[kugesan1105]

Dependency Analysis

---

[kugesan1105]

LSP Performance Bottlenecks

Problem

LSP response time scales ~quadratically with file size (1K lines = fast, 3K lines = noticeably slow).

Root Causes

1. get_token_start() is O(n) called O(n) times → O(n²)

• find_index() in utils.impl.jac:352 builds absolute positions by walking from index 0 for every token.
• 3K-line file (~6K tokens) = ~36M iterations per hover/position lookup.

2. Symbol resolution in gen_sem_tokens() scans entire hub → O(names × modules × scope_depth)

• sem_manager.impl.jac:524: unresolved NameAtom nodes brute-force search all loaded modules.
• Should be resolved during compilation, not token generation.

3. SemTokManager rebuilt from scratch on every compile

• worker.impl.jac:271: full AST traversal + symbol resolution on every save.
• Impl files trigger 3 rebuilds (impl → parent → impl again).

Fixes (Priority Order)

1. Pre-compute absolute token positions — cache list[tuple[int,int,int]] on SemTokManager, use binary search in find_index(). Turns O(n²) → O(log n).
2. Remove hub-scanning symbol resolution from SemTokManager — rely on compiler passes.
3. Build SemTokManager once after all compilation (not 3× for impl files).
4. Store errors per-file in dict[str, list[Alert]] instead of filtering a flat list.

[kugesan1105]

UP-NEXT

1. improve performance: semtoken generation
2. Re-org and refactor the request handler between process
3. Ensure all jac lsp features are working: outline symbols, completion
4. Refactor test to be as it was in previous version