feat(sql): make tree-sitter-sql a core dependency so default installs parse .sql (#1745)

README.md

@@ -262,7 +262,6 @@ Codex users also need `multi_agent = true` under `[features]` in `~/.codex/confi
 | `anthropic` | Anthropic Claude API (`--backend claude`, uses `ANTHROPIC_API_KEY`) | `uv tool install "graphifyy[anthropic]"` |
 | `bedrock` | AWS Bedrock (uses IAM, no API key) | `uv tool install "graphifyy[bedrock]"` |
 | `azure` | Azure OpenAI Service (`--backend azure`, uses `AZURE_OPENAI_API_KEY` + `AZURE_OPENAI_ENDPOINT`) | `uv tool install "graphifyy[openai]"` |
-| `sql` | SQL schema extraction | `uv tool install "graphifyy[sql]"` |
 | `postgres` | Live PostgreSQL introspection (`--postgres DSN`) | `uv tool install "graphifyy[postgres]"` |
 | `dm` | BYOND DreamMaker `.dm`/`.dme` AST extraction (may need a C compiler + `python3-dev` if no wheel matches your platform) | `uv tool install "graphifyy[dm]"` |
 | `terraform` | Terraform / HCL `.tf`/`.tfvars`/`.hcl` AST extraction | `uv tool install "graphifyy[terraform]"` |
@@ -272,6 +271,8 @@ Codex users also need `multi_agent = true` under `[features]` in `~/.codex/confi
 | `chinese` | Chinese query segmentation (jieba) | `uv tool install "graphifyy[chinese]"` |
 | `all` | Everything above | `uv tool install "graphifyy[all]"` |
 
+> **SQL is no longer an extra.** `.sql` schema extraction is included in every default install (`tree-sitter-sql` is a core dependency). If you previously installed the `sql` extra or ran `pipx inject graphifyy tree-sitter-sql`, the normal upgrade (`uv tool upgrade graphifyy` or `pipx upgrade graphifyy`) is all you need — the `sql` extra remains as a harmless alias for older install scripts.
+
 </details>
 
 ---

graphify/extract.py

@@ -5073,7 +5073,9 @@ def add_existing_edge(edge: dict) -> None:
 # rather than falling back like Pascal does. Used by the #1745 warning in
 # extract() to tell the user which extra restores the language.
 _EXTRA_FOR_EXTENSION = {
-    ".sql": "sql",
+    # .sql is deliberately absent: tree-sitter-sql is a core dependency, so a
+    # missing grammar means a broken install, not a missing extra
+    # — the extractor's own error names the direct pip install that repairs it.
     ".tf": "terraform",
     ".tfvars": "terraform",
     ".hcl": "terraform",

graphify/extractors/sql.py

@@ -41,7 +41,9 @@ def extract_sql(path: Path, content: str | bytes | None = None) -> dict:
         # and surface the real exception in the latter case.
         if importlib.util.find_spec("tree_sitter_sql") is None:
             return {"nodes": [], "edges": [],
-                    "error": "tree_sitter_sql not installed. Run: pip install tree-sitter-sql"}
+                    "error": ("tree_sitter_sql not installed. It is a core dependency, "
+                              "so this install is incomplete — reinstall "
+                              "graphifyy or run: pip install tree-sitter-sql")}
         return {"nodes": [], "edges": [],
                 "error": f"tree_sitter_sql is installed but failed to load: {e}"}
 

pyproject.toml

@@ -41,6 +41,13 @@ dependencies = [
     "tree-sitter-fortran>=0.6,<0.8",
     "tree-sitter-bash>=0.23,<0.27",
     "tree-sitter-json>=0.23,<0.26",
+    # Core dependency (was the [sql] extra): .sql files are a mainstream
+    # corpus language and the extra-gated grammar made a default install
+    # silently skip them (#1745). Ships prebuilt abi3 wheels for every
+    # supported platform (win/macOS/Linux glibc+musl, x86_64+arm64) from
+    # 0.3.9, so no C toolchain is needed — the reason tree-sitter-dm stays
+    # optional does not apply here.
+    "tree-sitter-sql>=0.3.9,<0.4",
 ]
 
 [project.urls]
@@ -74,7 +81,10 @@ anthropic = ["anthropic"]
 gemini = ["openai", "tiktoken"]
 openai = ["openai", "tiktoken"]
 chinese = ["jieba"]
-sql = ["tree-sitter-sql"]
+# tree-sitter-sql is now a core dependency. The extra is kept as
+# an alias so existing `graphifyy[sql]` install commands and scripts keep
+# resolving; it adds nothing beyond the core install.
+sql = ["tree-sitter-sql>=0.3.9,<0.4"]
 # extract_pascal() uses tree-sitter-pascal for AST-quality extraction (more
 # accurate calls/inherits edges) and falls back to a regex extractor when it is
 # absent (#781), so this stays optional. Unlike tree-sitter-dm below, it ships

tests/test_extract.py

@@ -3610,12 +3610,21 @@ def test_extract_no_warning_when_all_code_has_extractors(tmp_path, capsys):
     assert "no AST extractor" not in err
 
 
-def test_extract_warns_when_sql_extra_missing(tmp_path, capsys, monkeypatch):
+def test_extract_warns_when_sql_grammar_missing(tmp_path, capsys, monkeypatch):
     # #1745: .sql HAS a dispatch entry, so the #1689 warning can't fire, and
     # extract_sql returns an "error" result when tree-sitter-sql is absent, so
-    # the #1666 warning skips it too. The files must not vanish silently:
-    # extract() surfaces them with the [sql] extra named.
+    # the #1666 warning skips it too. The files must not vanish silently.
+    # The grammar is a core dependency, so absence means a broken
+    # install: the warning must name the direct repair, not a [sql] extra.
+    # The dev environment now ships the grammar, so simulate genuine absence
+    # by failing the import AND blanking find_spec for this one module.
+    import importlib.util as _ilu
     monkeypatch.setitem(sys.modules, "tree_sitter_sql", None)  # import -> ImportError
+    _real_find_spec = _ilu.find_spec
+    monkeypatch.setattr(
+        _ilu, "find_spec",
+        lambda name, *a, **k: None if name == "tree_sitter_sql" else _real_find_spec(name, *a, **k),
+    )
     s1 = tmp_path / "schema.sql"; s1.write_text("CREATE TABLE users (id INT);\n")
     s2 = tmp_path / "views.sql"; s2.write_text("CREATE VIEW v AS SELECT * FROM users;\n")
     py = tmp_path / "main.py"; py.write_text("def main():\n    return 1\n")
@@ -3625,7 +3634,8 @@ def test_extract_warns_when_sql_extra_missing(tmp_path, capsys, monkeypatch):
 
     assert "2 .sql file(s)" in err
     assert "tree_sitter_sql not installed" in err
-    assert 'graphifyy[sql]' in err
+    assert "core dependency" in err, "message must say the install is broken, not point at an extra"
+    assert "graphifyy[sql]" not in err, ".sql must not be hinted as an optional extra any more"
     assert "#1745" in err
     # the Python file still extracts normally
     labels = [n.get("label") for n in result["nodes"]]
@@ -3639,15 +3649,15 @@ def test_extract_warns_when_sql_extra_missing(tmp_path, capsys, monkeypatch):
 
 def test_extract_failed_sources_empty_when_sql_installed(tmp_path):
     """#2543: successful extracts do not appear in failed_sources."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     s = tmp_path / "schema.sql"; s.write_text("CREATE TABLE users (id INT);\n")
     py = tmp_path / "main.py"; py.write_text("def main():\n    return 1\n")
     result = extract([s, py], cache_root=tmp_path)
     assert result.get("failed_sources") == []
 
 
 def test_extract_no_missing_dep_warning_when_sql_installed(tmp_path, capsys):
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     s = tmp_path / "schema.sql"; s.write_text("CREATE TABLE users (id INT);\n")
     extract([s], cache_root=tmp_path)
     err = capsys.readouterr().err
@@ -3661,7 +3671,7 @@ def test_extract_sql_reports_load_failure_not_missing(tmp_path, monkeypatch):
     # `pip install` — but surface the real load exception instead.
     import builtins
     from graphify.extractors.sql import extract_sql
-    pytest.importorskip("tree_sitter_sql")  # find_spec must see it as installed
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip  # find_spec must see it as installed
 
     _orig_import = builtins.__import__
 
@@ -3682,7 +3692,7 @@ def test_extract_warns_sql_grammar_failed_to_load(tmp_path, capsys, monkeypatch)
     # grammar with the real cause and WITHOUT the misleading "install the extra"
     # hint, so the files are neither silently dropped nor sent to a no-op fix.
     import builtins
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
 
     _orig_import = builtins.__import__
 

tests/test_multilang.py

@@ -450,39 +450,39 @@ def test_cache_miss_after_file_change(tmp_path):
 
 # ── SQL ───────────────────────────────────────────────────────────────────────
 
-def _extract_sql_or_skip(fixture: str = "sample.sql"):
-    pytest.importorskip("tree_sitter_sql")
+def _extract_sql_required(fixture: str = "sample.sql"):
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     return extract_sql(FIXTURES / fixture)
 
 
 def test_sql_finds_tables():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     labels = [n["label"] for n in r["nodes"]]
     assert any("users" in l for l in labels)
     assert any("organizations" in l for l in labels)
 
 def test_sql_finds_view():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     labels = [n["label"] for n in r["nodes"]]
     assert any("active_users" in l for l in labels)
 
 def test_sql_finds_function():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     labels = [n["label"] for n in r["nodes"]]
     assert any("get_user" in l for l in labels)
 
 def test_sql_emits_foreign_key_edge():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     relations = {e["relation"] for e in r["edges"]}
     assert "references" in relations
 
 def test_sql_emits_reads_from_edge():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     relations = {e["relation"] for e in r["edges"]}
     assert "reads_from" in relations
 
 def test_sql_no_dangling_edges():
-    r = _extract_sql_or_skip()
+    r = _extract_sql_required()
     node_ids = {n["id"] for n in r["nodes"]}
     for e in r["edges"]:
         assert e["source"] in node_ids, f"dangling source: {e['source']}"
@@ -495,7 +495,7 @@ def test_sql_cte_is_not_read_as_a_table():
     with a same-named node from another language. The real table in the same
     FROM/JOIN must still resolve.
     """
-    r = _extract_sql_or_skip("sample_cte.sql")
+    r = _extract_sql_required("sample_cte.sql")
     labels = [n["label"] for n in r["nodes"]]
     assert "levels" not in labels, "CTE name leaked into the graph as a table node"
 
@@ -509,7 +509,7 @@ def test_sql_cte_is_not_read_as_a_table():
 
 def test_sql_column_list_cte_is_not_read_as_a_table(tmp_path):
     """#2577: `WITH levels(a, b) AS (...)` — the name precedes a column list."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     p = tmp_path / "schema.sql"
     p.write_text(
         "CREATE TABLE users (id INT, role TEXT);\n"
@@ -527,7 +527,7 @@ def test_sql_cte_shadows_same_named_table_within_its_statement(tmp_path):
     """#2577: inside the declaring statement the CTE shadows a real same-named
     table (SQL scoping), so v1's FROM binds to the CTE and emits nothing; v2 has
     no CTE in scope and reads the real table. Exactly one deterministic edge."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     p = tmp_path / "schema.sql"
     p.write_text(
         "CREATE TABLE levels (role TEXT);\n"
@@ -546,7 +546,7 @@ def test_sql_subquery_cte_does_not_suppress_outer_real_table(tmp_path):
     """#2577 refinement: a WITH inside a subquery is scoped to that subquery
     only. A statement-wide pre-collect would also swallow the OUTER reference
     to the real `t2`, dropping a true edge — per-subtree scoping keeps it."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     p = tmp_path / "schema.sql"
     p.write_text(
         "CREATE TABLE t2 (id INT);\n"
@@ -567,7 +567,7 @@ def test_sql_cte_never_binds_to_cross_language_symbol(tmp_path):
     so _rewire_unique_stub_nodes bound it to a same-named symbol from ANOTHER
     language (schema_v_roles -> ui_levels). With the CTE excluded, no reads_from
     edge may target a TypeScript node."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     sql = tmp_path / "schema.sql"
     sql.write_text(
         "CREATE TABLE users (id INT, role TEXT);\n"
@@ -594,7 +594,7 @@ def test_sql_cross_file_fk_resolves_and_never_leaks_scan_path(tmp_path):
     minted a node-less id under the referencing file's own stem, which with
     absolute inputs leaked the machine path AND could never match the m1
     definition, so prisma-style cross-migration FKs dangled."""
-    pytest.importorskip("tree_sitter_sql")
+    import tree_sitter_sql  # noqa: F401 — core dependency; absence must FAIL, not skip
     from graphify.ids import make_id
 
     m1 = tmp_path / "prisma" / "migrations" / "m1"
@@ -647,7 +647,7 @@ def test_sql_cross_file_fk_resolves_and_never_leaks_scan_path(tmp_path):
 
 def test_sql_alter_table_fk_edge():
     """ALTER TABLE ... FOREIGN KEY ... REFERENCES produces a references edge."""
-    r = _extract_sql_or_skip("sample_alter_fk.sql")
+    r = _extract_sql_required("sample_alter_fk.sql")
     fk_edges = [e for e in r["edges"] if e["relation"] == "references"]
     assert len(fk_edges) >= 1
     node_ids = {n["id"] for n in r["nodes"]}
@@ -657,14 +657,14 @@ def test_sql_alter_table_fk_edge():
 
 def test_sql_schema_qualified_names():
     """Schema-qualified table names (Schema.Table) are preserved."""
-    r = _extract_sql_or_skip("sample_schema_qualified.sql")
+    r = _extract_sql_required("sample_schema_qualified.sql")
     labels = [n["label"] for n in r["nodes"]]
     assert any("Sales.Customer" in l for l in labels)
     assert any("Sales.SalesOrder" in l for l in labels)
 
 def test_sql_schema_qualified_alter_fk():
     """ALTER TABLE with schema-qualified names produces correct edges."""
-    r = _extract_sql_or_skip("sample_schema_qualified.sql")
+    r = _extract_sql_required("sample_schema_qualified.sql")
     fk_edges = [e for e in r["edges"] if e["relation"] == "references"]
     assert len(fk_edges) >= 1
     node_ids = {n["id"] for n in r["nodes"]}
@@ -675,7 +675,7 @@ def test_sql_schema_qualified_alter_fk():
 def test_sql_plpgsql_functions_survive_parse_errors():
     """PL/pgSQL bodies make tree-sitter-sql emit ERROR nodes; the functions
     must still be extracted (#1910), without cascading into later statements."""
-    r = _extract_sql_or_skip("sample_plpgsql.sql")
+    r = _extract_sql_required("sample_plpgsql.sql")
     labels = [n["label"] for n in r["nodes"]]
     # Both PL/pgSQL functions extracted, schema-qualified name kept whole
     assert "exposed.important_function()" in labels
@@ -694,7 +694,7 @@ def test_sql_plpgsql_functions_survive_parse_errors():
 
 def test_sql_plpgsql_clean_function_not_double_emitted():
     """A cleanly-parsed LANGUAGE sql function in the same file is emitted once."""
-    r = _extract_sql_or_skip("sample_plpgsql.sql")
+    r = _extract_sql_required("sample_plpgsql.sql")
     labels = [n["label"] for n in r["nodes"]]
     assert labels.count("plain_sql_fn()") == 1
     # And nothing else is duplicated either
@@ -712,7 +712,7 @@ def test_sql_quoted_plpgsql_routines_are_recovered():
     with an *unquoted* name recovered fine, which is why the drop looked like it
     depended only on the body statement.
     """
-    r = _extract_sql_or_skip("sample_plpgsql_quoted.sql")
+    r = _extract_sql_required("sample_plpgsql_quoted.sql")
     labels = [n["label"] for n in r["nodes"]]
     for name in (
         "raise_exception_fn",
@@ -727,7 +727,7 @@ def test_sql_quoted_plpgsql_routines_are_recovered():
 
 def test_sql_quoted_plpgsql_file_stays_clean():
     """The #2180 recovery must not add junk, duplicates, or drop the tables."""
-    r = _extract_sql_or_skip("sample_plpgsql_quoted.sql")
+    r = _extract_sql_required("sample_plpgsql_quoted.sql")
     labels = [n["label"] for n in r["nodes"]]
     # Tables before and after the unparseable routines still extract.
     assert any("accounts" in l for l in labels)