fix: excuse dedup-explained node drops from the #479 shrink guard

graphify/build.py

@@ -1113,6 +1113,13 @@ def build_from_json(extraction: dict, *, directed: bool = False, root: str | Pat
         G.remove_node(ghost_id)
         node_set.discard(ghost_id)
 
+    # #1847: record how many nodes this canonicalization pass legitimately
+    # collapsed (AST/manifest duplicates sharing (source_file, label)), so a
+    # downstream shrink guard (export.to_json) can tell a merge-explained node
+    # count drop from an unexplained data-loss shrink instead of refusing both
+    # alike.
+    G.graph["_ghost_dedup_count"] = G.graph.get("_ghost_dedup_count", 0) + len(_ghost_remap)
+
     # Normalized ID map: lets edges survive when the LLM generates IDs with
     # slightly different casing or punctuation than the AST extractor.
     # e.g. "Session_ValidateToken" maps to "session_validatetoken".

graphify/export.py

@@ -312,18 +312,31 @@ def to_json(G: nx.Graph, communities: dict[int, list[str]], output_path: str, *,
                     return False
             new_n = G.number_of_nodes()
             if new_n < existing_n:
-                import sys as _sys
-                print(
-                    f"[graphify] WARNING: new graph has {new_n} nodes but existing "
-                    f"graph.json has {existing_n} (net -{existing_n - new_n}). "
-                    f"Refusing to overwrite. Possible causes: missing chunk files from "
-                    f"a previous session, or fuzzy dedup collapsed same-named symbols "
-                    f"across files during an --update on an already-current graph. "
-                    f"Run a full rebuild (/graphify .) to be safe, or pass force=True "
-                    f"only if you have verified the reduction is legitimate.",
-                    file=_sys.stderr,
-                )
-                return False
+                # #1847: build_from_json legitimately collapses duplicate
+                # nodes that resolve to the same (source_file, label) — e.g. a
+                # manifest-derived package node and its AST-canonical twin
+                # (crate:foo vs pkg_foo). That merge is recorded on the graph
+                # as `_ghost_dedup_count`. When the observed drop is fully
+                # explained by that count, it is a known-safe canonicalization,
+                # not the unverified data-loss shrink #479 guards against —
+                # proceed instead of refusing.
+                dedup = int(getattr(G, "graph", {}).get("_ghost_dedup_count", 0) or 0)
+                unexplained = (existing_n - new_n) - dedup
+                if unexplained > 0:
+                    import sys as _sys
+                    print(
+                        f"[graphify] WARNING: new graph has {new_n} nodes but existing "
+                        f"graph.json has {existing_n} (net -{existing_n - new_n}, "
+                        f"{dedup} explained by duplicate-node canonicalization, "
+                        f"{unexplained} unexplained). "
+                        f"Refusing to overwrite. Possible causes: missing chunk files from "
+                        f"a previous session, or fuzzy dedup collapsed same-named symbols "
+                        f"across files during an --update on an already-current graph. "
+                        f"Run a full rebuild (/graphify .) to be safe, or pass force=True "
+                        f"only if you have verified the reduction is legitimate.",
+                        file=_sys.stderr,
+                    )
+                    return False
 
     node_community = _node_community_map(communities)
     _labels: dict[int, str] = {int(k): v for k, v in (community_labels or {}).items()}

tests/test_build.py

@@ -488,6 +488,24 @@ def test_ghost_merge_unique_located_node_still_merges():
     assert G.has_edge("caller", "ast_render")
 
 
+def test_ghost_merge_records_dedup_count():
+    """#1847: each ghost collapsed into its AST canonical twin is a legitimate
+    node-count reduction, not data loss. build_from_json must record how many
+    were merged so a downstream shrink guard can tell the two apart instead of
+    refusing to overwrite a graph.json that only shrank via this dedup."""
+    ext = {
+        "nodes": [
+            {"id": "ast_render", "label": "render", "file_type": "code",
+             "source_file": "src/app/index.ts", "source_location": "L10", "_origin": "ast"},
+            {"id": "ghost_render", "label": "render", "file_type": "code",
+             "source_file": "src/app/index.ts"},
+        ],
+        "edges": [], "input_tokens": 0, "output_tokens": 0,
+    }
+    G = build_from_json(ext)
+    assert G.graph.get("_ghost_dedup_count") == 1
+
+
 def test_ghost_merge_uses_source_file_not_basename():
     """#2068: the ghost-merge key is the full source_file, not the bare basename.
     A ghost from src/a/index.ts merges into THAT file's AST node (a_render), never

tests/test_export.py

@@ -960,6 +960,30 @@ def test_to_json_refuses_shrink(tmp_path):
     assert to_json(_mkG(2), {}, str(p), force=True) is True  # force overrides
 
 
+def test_to_json_allows_dedup_explained_shrink(tmp_path):
+    """#1847: a cluster-only re-run that reloads graph.json through
+    build_from_json legitimately collapses a duplicate node (e.g. a
+    manifest package node merged into its AST-canonical twin, same
+    (source_file, label)). build_from_json records that collapse count on
+    `G.graph["_ghost_dedup_count"]`; the #479 shrink guard must let a drop
+    fully explained by it through instead of refusing."""
+    p = tmp_path / "graph.json"
+    json.dump({"nodes": [{"id": f"n{i}"} for i in range(5)]}, p.open("w"))
+    G = _mkG(4)
+    G.graph["_ghost_dedup_count"] = 1  # exactly explains the 5 -> 4 drop
+    assert to_json(G, {}, str(p), force=False) is True
+
+
+def test_to_json_still_refuses_shrink_beyond_dedup_count(tmp_path):
+    """The dedup allowance only excuses the EXPLAINED portion of a drop — a
+    further, unexplained loss on top of it must still refuse."""
+    p = tmp_path / "graph.json"
+    json.dump({"nodes": [{"id": f"n{i}"} for i in range(5)]}, p.open("w"))
+    G = _mkG(2)  # drop of 3, but only 1 is dedup-explained
+    G.graph["_ghost_dedup_count"] = 1
+    assert to_json(G, {}, str(p), force=False) is False
+
+
 def test_to_json_fails_safe_on_corrupt_existing(tmp_path):
     """A non-empty but unparseable existing graph.json (corrupt or mid-write)
     must NOT be silently overwritten — we can't verify the new graph isn't a