`CREATE INDEX` is silently dropped from SQL graphs, the same way `CREATE POLICY` is in #3401 — 218 declarations, 0 nodes, on a real schema

[lrafasouza]

Summary

CREATE INDEX and CREATE UNIQUE INDEX produce no nodes. In the same file, CREATE TABLE,
CREATE VIEW and CREATE FUNCTION all land correctly, so the file is not skipped and the grammar
is loading — indexes specifically are lost.

This is the same shape as #3401 (CREATE POLICY), and I could not find an existing issue for the
index case, so filing separately. If you consider them one bug, feel free to close this into #3401 —
but note that PR #3456 addresses policies only, so merging it would leave this half open.

Reproducer

One file, 19 lines of plain PostgreSQL. No project, no config.

-- schema.sql
CREATE TABLE public.profiles (
  id uuid PRIMARY KEY,
  owner uuid NOT NULL
);

CREATE INDEX profiles_owner_idx ON public.profiles (owner);

CREATE UNIQUE INDEX profiles_id_uniq ON public.profiles (id);

CREATE VIEW public.active_profiles AS SELECT * FROM public.profiles;

CREATE FUNCTION public.is_owner(u uuid) RETURNS boolean AS $$
  SELECT u = auth.uid();
$$ LANGUAGE sql;

ALTER TABLE public.profiles ENABLE ROW LEVEL SECURITY;

CREATE POLICY profiles_select_own ON public.profiles
  FOR SELECT USING (public.is_owner(owner));

CREATE POLICY profiles_update_own ON public.profiles
  FOR UPDATE USING (public.is_owner(owner));

$ graphify extract . --code-only
[graphify extract] AST extraction on 1 code files...
[graphify extract] wrote graphify-out/graph.json: 4 nodes, 4 edges, 2 communities

Expected vs actual

Every node in graph.json:

'schema.sql'
'public.is_owner()'
'public.active_profiles'
'public.profiles'

declaration	node
CREATE TABLE public.profiles	present
CREATE VIEW public.active_profiles	present
CREATE FUNCTION public.is_owner	present
CREATE INDEX profiles_owner_idx	absent
CREATE UNIQUE INDEX profiles_id_uniq	absent
CREATE POLICY profiles_select_own	absent (#3401)
CREATE POLICY profiles_update_own	absent (#3401)

There is no warning and no error. graphify extract exits 0.

Why this matters beyond the toy case

Measured on a production Next.js + Supabase codebase, 166 tracked .sql files, comments and
single-quoted strings stripped before counting, then matching each declaration name against node
labels in the same source file:

category	declared	in graph
CREATE TABLE	139	136
CREATE VIEW	42	38
CREATE FUNCTION	188	188
CREATE TRIGGER	122	113
CREATE INDEX	218	0
CREATE POLICY	440	0

The part that makes this more than a coverage gap: per-file coverage reads as a clean pass. All
166 .sql files produced at least one node, so every file-level coverage check says SQL is fully
mapped, while two entire entity classes are absent. Nothing in the output signals the hole.

And the graph does not go quiet when asked about the missing entities — it answers from whatever
else it has. graphify query "which RLS policies protect the profiles table" on that repo returned
394 nodes drawn from test files, a Drizzle schema symbol, UI components and prose documentation,
citing zero .sql sources. A plausible substitute answer is harder to catch than an empty one.

Environment

• graphify 0.9.57 (uv tool install 'graphifyy[sql,leiden,mcp,watch,svg]')
• Python 3.12.14, macOS (arm64)
• tree_sitter_sql imports cleanly in the tool env — verified before filing, since without that
  extra extract_sql returns zero nodes for a different reason

Possible cause

graphify/extractors/sql.py handles create_table and create_view node types explicitly, and the
ERROR-recovery regex _ROUTINE_RECOVERY_RX covers only FUNCTION/PROCEDURE. Searching that file
for index turns up no rule, the same way searching it for policy turns up none. If the grammar
emits an ERROR node for CREATE INDEX, the recovery path has nothing to catch it.

[safishamsi]

Fixed in v0.9.58 by #3470 (thanks @L4XB) — SQL CREATE [UNIQUE] INDEX emits index nodes. Release: https://github.com/Graphify-Labs/graphify/releases/tag/v0.9.58