Merge pull request 'dev → main: drop cross-project hook uses edges (#6483), plain-word calls need no space (#6477)' (#218) from dev into main

alembic/versions/0126_drop_cross_project_hook_uses.py

@@ -0,0 +1,34 @@
+"""drop write-path uses edges that cross projects
+
+Revision ID: 0126
+Revises: 0125
+Create Date: 2026-10-26
+
+The write path records a `uses` edge with basis `hook` when a session pulled a
+snippet and then wrote code naming its symbol. That is by-name evidence, like
+the proposer's `reference` edges 0125 removed, and a body never calls a
+function defined in another project's repository. The write path now applies
+the same by-name test, so it writes no such edge; this removes the ones it
+wrote before. Judged edges keep their own basis and are left alone.
+
+Downgrade restores nothing: the removed edges were never facts.
+"""
+from alembic import op
+
+revision = "0126"
+down_revision = "0125"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.execute(
+        "DELETE FROM code_shape_uses u USING code_shapes s, notes n "
+        "WHERE u.shape_id = s.id AND u.snippet_id = n.id "
+        "AND u.basis = 'hook' "
+        "AND n.project_id IS NOT NULL AND n.project_id <> s.project_id"
+    )
+
+
+def downgrade() -> None:
+    pass

src/grimoire/services/shape_ledger.py

@@ -1798,7 +1798,7 @@ def _semantic_priority(row) -> tuple:
 # were examined under, so a tightened rule re-examines everything once. A
 # change that does not alter what is proposed needs no bump. A stale "miss"
 # basis on an old row is inert: a miss is not evidence (see _SEMANTIC_LIMIT).
-_PROPOSER_VERSION = 7
+_PROPOSER_VERSION = 8
 # Signature resemblance floor, name blanked (difflib ratio) — and a length
 # floor, because `def NAME():` resembles `def NAME(x):` at 0.95 while saying
 # nothing; a family shape has parameters to resemble.
@@ -1939,7 +1939,9 @@ def names_canon(
       unless the canon is recorded in this very file;
     - never a generic verb, with or without a leading underscore;
     - a bare lowercase word (`file`, `seen`, `seal`) only where it is called
-      as `word(`, because as a plain word it is everyone's variable.
+      as `word(`, because as a plain word it is everyone's variable. No space
+      is allowed before the parenthesis: formatted code never has one, and
+      prose does ("a file (not a directory)").
     Same kind and language family are checked by the callers' loops."""
     sym = _norm_symbol(canon.symbol or "")
     if not sym:
@@ -1954,7 +1956,7 @@ def names_canon(
         if not any(lp and _path_touches(lp, path) for lp, _ls in canon.locations):
             return False
     if kind == "sym" and re.fullmatch(r"[a-z]+", sym):
-        return re.search(rf"(?<![\w$]){re.escape(sym)}\s*\(", body or "") is not None
+        return re.search(rf"(?<![\w$]){re.escape(sym)}\(", body or "") is not None
     return references_symbol(body, sym, kind)
 
 

tests/test_shape_ledger.py

@@ -601,7 +601,9 @@ def test_a_plain_word_counts_only_where_it_is_called():
     from grimoire.services.shape_ledger import Canon, _norm_text, reference_canons
     seen = Canon(8, "sym", "seen", (("src/tree.py", "seen"),), "def seen(f):", _norm_text("x"), 2, "python")
     assert reference_canons("sym", "src/a.py", "walk", "if path in seen:\n    return", [seen]) == []
-    assert reference_canons("sym", "src/a.py", "walk", "if seen (f):\n    return", [seen]) == [8]
+    assert reference_canons("sym", "src/a.py", "walk", "if seen(f):\n    return", [seen]) == [8]
+    # Prose puts a space before the parenthesis; a call does not.
+    assert reference_canons("sym", "src/a.py", "walk", "// a seen (or unseen) path\nreturn p", [seen]) == []
     # A leading underscore does not get a generic verb past the stoplist.
     run = Canon(9, "sym", "_run", (("src/x.py", "_run"),), "def _run(cmd):", _norm_text("x"), 2, "python")
     assert reference_canons("sym", "src/a.py", "go", "_run(['ls'])", [run]) == []