diffcontext 0.5.3__tar.gz → 0.5.4__tar.gz

{diffcontext-0.5.3 → diffcontext-0.5.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diffcontext
-Version: 0.5.3
+Version: 0.5.4
 Summary: Static-analysis-powered repository context compiler for LLMs
 Author-email: Trakshan Mishra <trakshanmishra477@gmail.com>
 License: MIT

{diffcontext-0.5.3 → diffcontext-0.5.4}/diffcontext/__init__.py

@@ -32,7 +32,7 @@ Usage as a library:
     print(ctx.reduction_pct)    # how much code was filtered out
 """
 
-__version__ = "0.5.3"
+__version__ = "0.5.4"
 
 # Public, semver-covered API. Everything not listed here (graph_builder,
 # resolver, symbols, scanner, parser internals) is importable but carries no

{diffcontext-0.5.3 → diffcontext-0.5.4}/diffcontext/cli/__init__.py

@@ -392,6 +392,11 @@ def _cmd_compile(args):
         from ..history import CoChangeIndex
         history = CoChangeIndex(args.repo)
 
+    # Untuned default; not derived from any sweep. The query_weight controls
+    # how much the problem-description signal moves candidates relative to
+    # the graph + BM25 + same-file blend. 0.3 is a guess that "matters but
+    # doesn't dominate" — it has NOT been benchmarked or optimized. Do not
+    # cite this as a measured value; tune it with evidence when data exists.
     query_weight = 0.3 if args.query_text else 0.0
     impact = analyze_impact(
         idx, changed, max_depth=args.depth, hybrid=not args.graph_only,

{diffcontext-0.5.3 → diffcontext-0.5.4}/diffcontext/mcp_server.py

@@ -131,8 +131,10 @@ def main():
         if not changed and not task_description:
             return "No changed symbols found. Pass changed_symbols, git_ref, or task_description."
 
-        # query_weight > 0 only when task_description is provided — the
-        # signal is opt-in and has no effect when absent (backwards compat).
+        # Untuned default; not derived from any sweep. The query_weight controls
+        # how much the problem-description signal moves candidates relative to
+        # the graph + BM25 + same-file blend. 0.3 is a guess that "matters but
+        # doesn't dominate" — it has NOT been benchmarked or optimized.
         query_weight = 0.3 if task_description else 0.0
         impact = analyze_impact(
             idx, changed, query_text=task_description, query_weight=query_weight,

{diffcontext-0.5.3 → diffcontext-0.5.4}/diffcontext.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diffcontext
-Version: 0.5.3
+Version: 0.5.4
 Summary: Static-analysis-powered repository context compiler for LLMs
 Author-email: Trakshan Mishra <trakshanmishra477@gmail.com>
 License: MIT

{diffcontext-0.5.3 → diffcontext-0.5.4}/tests/test_core.py

@@ -308,55 +308,80 @@ class TestPipeline:
 class TestQueryText:
     """The query_text signal: BM25 scores the bug report / issue text
     against every symbol's source code, adding a relevance-to-the-problem
-    signal the graph alone can't provide."""
+    signal the graph alone can't provide.
 
-    def test_query_text_adds_candidates(self):
+    Two requirements the user asked for:
+    1. query_text=None reproduces current ranking byte-identically
+       (regression guard — no behavior change when the feature is off).
+    2. A non-None value measurably reorders results (the signal actually
+       changes the ranking, not just the scores).
+    """
+
+    def test_none_reproduces_baseline_byte_identically(self):
+        """The regression guard: query_text=None must not change ANY score,
+        not even by a floating-point epsilon. This is what keeps the feature
+        truly opt-in — existing users see zero behavior change."""
         _require_fixture(MEDIUM, "medium_repo")
         idx = index_repository(MEDIUM)
-        sym_id = "./service.py:onboard_user"
-        assert sym_id in idx.graph
-
-        # Without query_text: baseline scores
-        impact_base = analyze_impact(idx, [sym_id])
-        base_sids = set(impact_base.scores)
+        seed = "./service.py:onboard_user"
 
-        # With query_text: should surface additional candidates whose code
-        # matches the described problem but may have no call-graph connection.
-        # Use a query that matches symbols in the medium repo.
-        impact_q = analyze_impact(
-            idx, [sym_id], query_text="create user order onboard", query_weight=0.3,
+        baseline = analyze_impact(idx, [seed])
+        with_query_kwarg = analyze_impact(
+            idx, [seed], query_text=None, query_weight=0.0,
         )
-        query_sids = set(impact_q.scores)
-
-        # Query-text candidates can add symbols the graph didn't reach.
-        # At minimum, the query should not REMOVE any baseline candidates.
-        assert base_sids.issubset(query_sids), (
-            "query_text must not remove baseline candidates, only add them"
+        # Every score must be identical — not just the set of candidates,
+        # but the actual float values. query_text=None is a true no-op.
+        assert baseline.scores == with_query_kwarg.scores, (
+            "query_text=None must reproduce baseline scores exactly; "
+            "the feature must be a true no-op when disabled."
         )
 
-    def test_query_weight_zero_is_noop(self):
+    def test_query_text_reorders_results(self):
+        """A non-None query_text must measurably change the ranking — not
+        just add candidates, but move at least one existing symbol to a
+        different rank position. This proves the signal is doing work."""
         _require_fixture(MEDIUM, "medium_repo")
         idx = index_repository(MEDIUM)
-        sym_id = "./service.py:onboard_user"
+        seed = "./service.py:onboard_user"
 
-        # query_weight=0.0 = no effect (backwards compatible)
-        impact_off = analyze_impact(idx, [sym_id], query_text="anything", query_weight=0.0)
-        impact_none = analyze_impact(idx, [sym_id])
-        # Same scores (query_text with weight 0 is a no-op)
-        assert impact_off.scores == impact_none.scores, (
-            "query_weight=0.0 must be a no-op"
+        baseline = analyze_impact(idx, [seed])
+        base_ranked = sorted(baseline.scores.items(), key=lambda x: -x[1])
+        base_order = [sid for sid, _ in base_ranked]
+
+        # "email validation check" should boost is_valid_email (whose code
+        # contains "email" and "valid") above User.__init__ (which doesn't).
+        impacted = analyze_impact(
+            idx, [seed], query_text="email validation check", query_weight=0.3,
+        )
+        q_ranked = sorted(impacted.scores.items(), key=lambda x: -x[1])
+        q_order = [sid for sid, _ in q_ranked]
+
+        assert base_order != q_order, (
+            "query_text must reorder at least one symbol; if the ranking "
+            "is identical the signal has no effect."
+        )
+        # Specifically: is_valid_email should rank higher with the query
+        # "email validation check" than without it.
+        base_email_rank = base_order.index("./validators.py:is_valid_email")
+        q_email_rank = q_order.index("./validators.py:is_valid_email")
+        assert q_email_rank < base_email_rank, (
+            f"is_valid_email should rank HIGHER with query_text='email "
+            f"validation check' (baseline rank {base_email_rank} -> query "
+            f"rank {q_email_rank}), not lower or unchanged."
         )
 
-    def test_no_query_text_is_noop(self):
+    def test_query_weight_zero_is_noop(self):
+        """query_weight=0.0 must be a no-op even with non-None query_text."""
         _require_fixture(MEDIUM, "medium_repo")
         idx = index_repository(MEDIUM)
-        sym_id = "./service.py:onboard_user"
+        seed = "./service.py:onboard_user"
 
-        # No query_text = no effect even if query_weight > 0
-        impact_off = analyze_impact(idx, [sym_id], query_weight=0.3)
-        impact_none = analyze_impact(idx, [sym_id])
-        assert impact_off.scores == impact_none.scores, (
-            "no query_text with query_weight > 0 must be a no-op"
+        baseline = analyze_impact(idx, [seed])
+        with_weight_zero = analyze_impact(
+            idx, [seed], query_text="email validation", query_weight=0.0,
+        )
+        assert baseline.scores == with_weight_zero.scores, (
+            "query_weight=0.0 must be a no-op even with non-None query_text."
         )