fix(extract): resolve TS/JS barrel re-exports and extensionless imports

[abhishekkolge-design]

Problem

Two gaps in the JS/TS AST extractor caused barrel-exported symbols to appear as isolated nodes with no inbound edges in the graph, even when they were actively imported by other files.

Gap 1 — Re-export statements were invisible

export * from './X' and export { A } from './X' produce export_statement nodes in the tree-sitter AST. These were never in import_types, so the walk() function never processed them. In any TypeScript project using barrel index.ts files, every component re-exported through a barrel appeared as a disconnected island.

Gap 2 — Extensionless imports didn't resolve to real file IDs

import { X } from './components' (no extension) resolved to a target node ID for the bare path (e.g. …_components) which never matched any actual file node (e.g. …_components_index_ts). Same issue for import { Y } from './UnverifiedEmailBanner' — the target ID lacked the _tsx suffix.

Fix

1. LanguageConfig — new reexport_types field

reexport_types: frozenset = frozenset()  # e.g. {"export_statement"} for JS/TS barrel re-exports

2. walk() — re-export handling

After the import_types block, check for re-export nodes. Only triggers when the node has a from child keyword — this distinguishes export * from './X' from regular export const/function/class declarations, which must still walk their children for node extraction.

if config.reexport_types and t in config.reexport_types:
    has_from = any(c.type == "from" for c in node.children)
    if has_from and config.import_handler:
        config.import_handler(node, source, file_nid, stem, edges, str_path)
        return
    # Regular exports fall through to walk children normally

3. _import_js() — extensionless + index file resolution

if not resolved.suffix:
    for ext in (".ts", ".tsx", ".js", ".jsx"):
        candidate = resolved.with_suffix(ext)
        if candidate.exists():
            resolved = candidate
            break
    else:
        for ext in (".ts", ".tsx", ".js", ".jsx"):
            candidate = resolved / ("index" + ext)
            if candidate.exists():
                resolved = candidate
                break

4. Enabled for _JS_CONFIG and _TS_CONFIG

reexport_types=frozenset({"export_statement"}),

Verification

Tested on a large TypeScript monorepo (~2,800 files, Next.js + React). The chain:

NavigationSidebar.tsx  →  components/index.ts  →  UnverifiedEmailBanner.tsx

Previously: all three nodes appeared isolated (0 cross-file edges between them).
After fix: NavigationSidebar.tsx --imports_from→ index.ts --imports_from→ UnverifiedEmailBanner.tsx ✓

The graph went from treating every barrel-exported component as a disconnected island to correctly tracing the full import chain through index.ts files.

🤖 Generated with Claude Code

[vhsantos26]

Empirical signal that this fix matters in practice: running graphify on a TS codebase with path aliases gives imports_ratio: 0.0%. On a codebase without aliases but with .mjs relative imports, it's 11.6%. Both cases point to the same class of problem this PR addresses — silent dropping of import edges when target IDs don't match file node IDs.

The extensionless + barrel resolution logic here should recover a significant portion of the missing signal, especially in monorepos with barrel index.ts files.

PR is large — if it would help review, consider splitting Gap 1 (re-export handling) and Gap 2 (extensionless resolution) into separate commits or PRs? Happy to help benchmark either path.

[abhishekkolge-design]

@vhsantos26
Thanks for the review, the imports_ratio datapoints were really helpful for validation.

the PR isn’t just the fixes, it’s a refactor plus the fixes bundled together.

The bulk of the diff comes from the refactor. Specifically, it consolidates the per language extractors (extract_python, extract_js_ts, extract_go, etc.) into a shared LanguageConfig abstraction with a walk() dispatcher.

The actual fix logic is quite small roughly 25 lines total:

Gap 1: Adds a reexport_types field to LanguageConfig and handles it in walk()
Gap 2: Adds extensionless resolution + index.{ts,tsx,js,jsx} handling in the new _import_js helper

Both fixes rely on the abstractions introduced in the refactor, which is why they ended up in the same PR.

If you’d prefer a cleaner review flow, the proper split would be:

1. Refactor (introduce LanguageConfig + walk())
2. Gap 1 fix
3. Gap 2 fix

I’ve already prepared this as stacked PRs locally and verified that the final result is byte-identical to the current PR. Happy to push that version, or keep this as-is and expand the PR description to better explain the refactor scope, whatever you prefer.