Svelte: every .svelte file reports a syntax error at line 1; functions inside <script> are never extracted (extract_svelte feeds the whole file to the JS grammar)

[chrisltq97]

Summary

extract_svelte calls _extract_generic(path, _JS_CONFIG) on the raw .svelte file. The HTML template is not JavaScript, so tree-sitter hits a top-level ERROR at line 1 and never reaches function_declaration nodes inside <script>. Only imports survive, via the regex rescue pass. On a SvelteKit app with 394 components, every single one was flagged "syntax errors and may be partially extracted".

This is the same failure mode as #2788 / PR #3902 for Astro. Vue already avoids it via _vue_mask_non_script.

Reproduce (graphify 0.9.71, macOS, Python 3.13)

mkdir t && cd t
printf '<script>\n  import { foo } from "./x";\n  let n = 1;\n  function bump() { n++; }\n</script>\n<button onclick={bump}>{n}</button>\n' > A.svelte
graphify extract . --code-only --no-cluster

Output:

warning: 1 file(s) had syntax errors and may be partially extracted: A.svelte (first error at line 1, 1 symbol(s) extracted)
wrote graphify-out/graph.json — 2 nodes, 1 edges

bump() is missing. Same result with <script lang="ts"> and with Svelte 5 runes ($state, $props).

Fix that works locally

Mirror extract_vue: mask non-script regions with _vue_mask_non_script (regex is generic <script …>…</script>), pick _JS_CONFIG when lang is js/jsx else _TS_CONFIG, run _normalize_ts_import_types, and pass source_override= to _extract_generic. The existing regex import-rescue stays as-is.

src = path.read_text(encoding="utf-8", errors="replace")
masked, lang = _vue_mask_non_script(src)
cfg = _JS_CONFIG if lang in ("js", "jsx") else _TS_CONFIG
mb = masked.encode("utf-8")
if cfg is _TS_CONFIG:
    mb = _normalize_ts_import_types(mb) or mb
result = _extract_generic(path, cfg, source_override=mb)

Results on a real SvelteKit repo (658 code files, 394 .svelte):

	files with parse errors	nodes	edges
before	394	1829	3816
after	0	3377	5921

Minimal repro above: 2 nodes / 1 edge -> 3 nodes / 3 edges (bump() present). Line numbers stay correct because the mask preserves \n.

Happy to open a PR if useful.

[github-actions]

Thanks for opening this issue, @chrisltq97. A maintainer will take a look soon.

If you would like to discuss it in real time, come say hi on our Discord server. For longer-form questions and ideas there is also GitHub Discussions.

To help us triage, please make sure the report includes what you expected, what actually happened, and the steps (and a small sample) to reproduce it.