codingest 0.2.4__tar.gz → 0.2.6__tar.gz

{codingest-0.2.4 → codingest-0.2.6}/Cargo.lock

@@ -308,16 +308,18 @@ dependencies = [
 
 [[package]]
 name = "codingest"
-version = "0.2.4"
+version = "0.2.6"
 dependencies = [
  "aho-corasick",
  "base64 0.22.1",
  "kglite",
+ "proc-macro2",
  "rayon",
  "regex",
  "serde",
  "serde_json",
  "sha2",
+ "syn 2.0.119",
  "tempfile",
  "toml",
  "tree-sitter",
@@ -342,7 +344,7 @@ dependencies = [
 
 [[package]]
 name = "codingest-cli"
-version = "0.2.4"
+version = "0.2.6"
 dependencies = [
  "anyhow",
  "clap",
@@ -354,7 +356,7 @@ dependencies = [
 
 [[package]]
 name = "codingest-mcp"
-version = "0.2.4"
+version = "0.2.6"
 dependencies = [
  "anyhow",
  "codingest",
@@ -363,7 +365,7 @@ dependencies = [
 
 [[package]]
 name = "codingest-py"
-version = "0.2.4"
+version = "0.2.6"
 dependencies = [
  "codingest",
  "codingest-cli",
@@ -1390,9 +1392,9 @@ dependencies = [
 
 [[package]]
 name = "kglite"
-version = "0.16.2"
+version = "0.16.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e8d8175004ae15c78c32ea60abfa7515dc384ceac8c9dbeb2272884bd24b2401"
+checksum = "e79966b681151c31fd9062a0e219cb69965b19715354201d408b6aed8a257116"
 dependencies = [
  "bzip2",
  "chrono",
@@ -1425,9 +1427,9 @@ dependencies = [
 
 [[package]]
 name = "kglite-mcp-server"
-version = "0.16.2"
+version = "0.16.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "79ff376e0f81bcb1312256c762d3cac9a7665f6340d72f8503b266b0e25eece2"
+checksum = "889deddaa25ae6358bccd00ebeb43cd2ed07ee35a6f3d3624aeb625c75260470"
 dependencies = [
  "anyhow",
  "bytes",
@@ -1529,9 +1531,9 @@ dependencies = [
 
 [[package]]
 name = "mcp-methods"
-version = "0.4.4"
+version = "0.4.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d6d41f37fa137ea40a0cd505729e41ae1156ed119dcf8f240210579889c5ba29"
+checksum = "d413193821b58fa600d8b110fc5b7141cf71fcf367af0dbda77dca86c37eb1c8"
 dependencies = [
  "anyhow",
  "clap",

{codingest-0.2.4 → codingest-0.2.6}/Cargo.toml

@@ -3,7 +3,7 @@ members = ["crates/codingest", "crates/codingest-cli", "crates/codingest-mcp", "
 resolver = "2"
 
 [workspace.package]
-version = "0.2.4"
+version = "0.2.6"
 edition = "2021"
 rust-version = "1.88"
 license = "MIT"
@@ -12,8 +12,65 @@ repository = "https://github.com/kkollsga/codingest"
 [workspace.dependencies]
 base64 = "0.22"
 sha2 = "0.11"
-# kglite engine + MCP server — imported as crates.io libraries. The 0.16.2
-# floor fixes the edge-property type registry: `add_connections` — the ONLY
+# kglite engine + MCP server — imported as crates.io libraries. The 0.16.6
+# floor picks up the advisory traversal caps (our reported LIMIT bug: capped
+# pass + exact uncapped retry), the BREAKING var-length trail-semantics fixes
+# (a closed trail re-emits its source; a segment's relationships are visible to
+# its clause's uniqueness check), and per-section .kgl CRC32s. Zero source
+# changes: the Rust API delta is purely additive for every symbol we name, and
+# add_nodes/add_connections behavior is unchanged, so the frozen goldens hold.
+# The .kgl format gains a `section_digests` metadata field — additive in BOTH
+# directions (serde default/skip), so unlike 0.16.2 a stale .kgl needs NO
+# rebuild; saves pay the CRC pass (upstream ~+11-15%). The semantics change IS
+# observable in query answers on cyclic graphs — pinned by
+# crates/codingest/tests/traversal_semantics.rs, which is red on 0.16.5 and
+# green on 0.16.6 by design.
+#
+# History — the 0.16.5 floor was a LOCKSTEP REFRESH, not a fix we needed:
+# nothing in 0.16.3-0.16.5
+# reaches a path codingest exercises, and the evidence is the gate rather than
+# the changelog. `golden_parity` is green and `kgl_bytes_are_stable_across_builds`
+# holds, so the frozen goldens are byte-identical across the bump; the fixture
+# corpus builds 405 nodes / 616 edges docs-on and 398 / 596 docs-off on both
+# 0.16.2 and 0.16.5. Perf is flat: three interleaved A/B pairs per docs mode on
+# corpus ad8b3084…, control `defs_per_file` moving +0.0%, every cell flat except
+# `varlen_callers_1_3` at -9.7% in BOTH modes (0.031 -> 0.028 ms) — a small
+# consistent win, plausibly the PropMap refcount clone below. A same-machine
+# before/after pair taken 20 minutes apart instead read +50% on three cells with
+# the control at +15%; that is the instrument, and the interleaved pairs are why
+# the flat reading is the one recorded (doctrine R11).
+#
+# What each release contains, and why none of it lands on us:
+#   * 0.16.5 is a BREAKING Rust API change — node, relationship and map
+#     properties are a `kglite::datatypes::PropMap`, not a
+#     `BTreeMap<String, Value>`. We are not exposed: codingest names neither
+#     `NodeValue::properties` nor `RelValue::properties`, and its one
+#     `Value::Map` site (docs/mod.rs) matches the wildcard `Value::Map(_)` to
+#     stringify a non-scalar, which binds nothing. The workspace compiles clean
+#     with no source change. No file format moved — postcard's map framing is
+#     identical — which is what lets the goldens hold. Also in 0.16.5:
+#     relationship constraints, CDC before-images and selectors, and a
+#     re-save determinism fix. codingest declares no constraints, consumes no
+#     CDC, and only ever writes FRESH saves, so the re-save fix (and its
+#     property-name reordering) cannot reach our bytes.
+#   * 0.16.4 adds the opt-in parallel Cypher runtime (off by default and NOT
+#     exposed by the MCP server), property-type constraints, and CDC v1. Its
+#     one format note — a `.kgl` carrying a property-type constraint does not
+#     load on 0.16.3 or earlier — is inert here because codingest declares
+#     none, so our files stay readable by an older engine. That is why the
+#     Python floor moves for LOCKSTEP, not because the format broke: the
+#     `test_kglite_cargo_and_python_floors_are_in_lockstep` gate requires
+#     Cargo.toml, pyproject.toml and ci.yml to state one floor, and a
+#     "the reader could stay lower" argument is exactly the drift it exists to
+#     stop.
+#   * 0.16.3 is MCP-side: `reload_graph`, `extensions.graph_watch`, and
+#     `extensions.tools_allow`. It also carries mcp-methods 0.4.5, which makes
+#     the GitHub tools require an explicit `builtins.github: true` — codingest-mcp
+#     registers no GitHub tools and ships no manifest declaring them, so its
+#     agent-visible tool surface is unchanged.
+#
+# Floor history below; the 0.16.2 entry is the last one that changed our output.
+# The 0.16.2 floor fixed the edge-property type registry: `add_connections` — the ONLY
 # edge path codingest uses (builder/load.rs, other_edges.rs, cross_lang.rs,
 # docs/mod.rs) — registered every property as "Unknown" instead of its observed
 # type. Measured on a two-function Python corpus: `CALL
@@ -34,7 +91,7 @@ sha2 = "0.11"
 # which kglite 0.15.14 and earlier refuse by version number. Because
 # `codingest.build()` writes the bytes with the Rust engine compiled in here and
 # reads them back through the separately-installed Python `kglite` wheel, the
-# two sides must be on the same side of that break — hence `>=0.16.2,<0.17` in
+# two sides must be on the same side of that break — hence `>=0.16.6,<0.17` in
 # pyproject.toml, and hence a reader on 0.15.x is not "old but workable", it is
 # a hard failure with `FileFormatError`.
 # 0.16.0 also removed `DirGraph::enable_columnar` from the public API (every
@@ -99,8 +156,8 @@ sha2 = "0.11"
 # and no disk storage. Neither 0.15.7 nor 0.15.8 changes the `kglite::api` we
 # call, graph output, property encoding, or `.kgl` serialization.
 # Keep both libraries aligned so the graph and server share one engine release.
-kglite = { version = "0.16.2", default-features = false }
-kglite-mcp-server = { version = "0.16.2", default-features = false }
+kglite = { version = "0.16.6", default-features = false }
+kglite-mcp-server = { version = "0.16.6", default-features = false }
 
 # Mirror KGLite's release codegen so the standalone builder has
 # performance parity with the in-tree component (see KGLite

{codingest-0.2.4 → codingest-0.2.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codingest
-Version: 0.2.4
+Version: 0.2.6
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Programming Language :: Python :: 3
@@ -12,7 +12,7 @@ Classifier: Programming Language :: Python :: 3.14
 Classifier: Programming Language :: Rust
 Classifier: Topic :: Software Development
 Classifier: Topic :: Software Development :: Libraries
-Requires-Dist: kglite>=0.16.2,<0.17
+Requires-Dist: kglite>=0.16.6,<0.17
 License-File: LICENSE
 Summary: Build fast, queryable code graphs for AI agents, MCP code review, and local or GitHub repository analysis.
 Keywords: ai-agents,mcp,code-review,knowledge-graph,code-analysis,github,tree-sitter,cypher
@@ -192,7 +192,7 @@ codingest skill install        # code-review Agent Skill
 ```
 
 The wheel bundles every grammar plus the `codingest` and `codingest-mcp`
-commands. KGLite ≥0.16.2 is installed automatically as the query/storage
+commands. KGLite ≥0.16.6 is installed automatically as the query/storage
 engine; its MCP server and the transitive `mcp-methods` framework power the
 builder-aware Codingest server. Nothing else needs to be installed.
 
@@ -205,7 +205,7 @@ Python prerequisites.
 ```toml
 [dependencies]
 codingest = "0.2"
-kglite = "0.16.2"
+kglite = "0.16.6"
 ```
 
 ```rust
@@ -277,7 +277,7 @@ diagnostic.
 ## Dependency policy
 
 `kglite` and `kglite-mcp-server` use matching crates.io requirements with a
-0.16.2 minimum and a shared lockfile. This keeps the builder, persistence
+0.16.6 minimum and a shared lockfile. This keeps the builder, persistence
 handoff, and embedded MCP server on one compatible engine patch line.
 
 ## Parity with the (now-removed) in-tree component

{codingest-0.2.4 → codingest-0.2.6}/README.md

@@ -165,7 +165,7 @@ codingest skill install        # code-review Agent Skill
 ```
 
 The wheel bundles every grammar plus the `codingest` and `codingest-mcp`
-commands. KGLite ≥0.16.2 is installed automatically as the query/storage
+commands. KGLite ≥0.16.6 is installed automatically as the query/storage
 engine; its MCP server and the transitive `mcp-methods` framework power the
 builder-aware Codingest server. Nothing else needs to be installed.
 
@@ -178,7 +178,7 @@ Python prerequisites.
 ```toml
 [dependencies]
 codingest = "0.2"
-kglite = "0.16.2"
+kglite = "0.16.6"
 ```
 
 ```rust
@@ -250,7 +250,7 @@ diagnostic.
 ## Dependency policy
 
 `kglite` and `kglite-mcp-server` use matching crates.io requirements with a
-0.16.2 minimum and a shared lockfile. This keeps the builder, persistence
+0.16.6 minimum and a shared lockfile. This keeps the builder, persistence
 handoff, and embedded MCP server on one compatible engine patch line.
 
 ## Parity with the (now-removed) in-tree component

{codingest-0.2.4 → codingest-0.2.6}/crates/codingest/Cargo.toml

@@ -10,7 +10,7 @@ keywords = ["knowledge-graph", "code-analysis", "tree-sitter", "cypher", "kglite
 categories = ["development-tools", "parser-implementations"]
 
 # Requires the kglite 0.15 engine API (`kglite::api::code_entities`); the
-# workspace pins the current engine floor (0.16.2).
+# workspace pins the current engine floor (0.16.6).
 
 [lib]
 name = "codingest"
@@ -75,3 +75,12 @@ aho-corasick = "1.1.4"
 
 [dev-dependencies]
 tempfile = "3.27.0"
+# `tests/grammar_kinds.rs` parses the parser sources as Rust instead of
+# regexing them, which is what makes `match node.kind() { "lit" => … }` arms
+# exactly extractable (a regex over arm bodies false-positives on every other
+# string literal in the body). `full` = item/statement/expression syntax,
+# `visit` = the borrowing visitor the extractor implements. `span-locations`
+# on proc-macro2 is what gives the failure message a real `file:line` for each
+# offending literal; without it every span reports line 1.
+syn = { version = "2.0.119", features = ["full", "visit"] }
+proc-macro2 = { version = "1.0.107", features = ["span-locations"] }

{codingest-0.2.4 → codingest-0.2.6}/crates/codingest/src/builder/load.rs

@@ -944,10 +944,16 @@ impl<'a> LoadPipeline<'a> {
             defaults.iter().find(|(_, exists)| *exists).map(|(n, _)| *n)
         };
 
-        if !type_out.implements.is_empty() {
-            // Route IMPLEMENTS per-row based on the resolved source/target types.
-            // Python's _add_typed_connections does the equivalent via name_to_qname.
-            let mut qname_to_type: HashMap<String, &'static str> = HashMap::new();
+        // `qualified_name` (and bare name) → the node type that symbol was
+        // actually inserted as. IMPLEMENTS and EXTENDS both route per-row
+        // through this map: `add_connections` is given ONE label per endpoint,
+        // so handing it a label the endpoint was not stored under makes it
+        // mint a `_provisional` stub of that wrong type beside the real node
+        // and hang the edge off the phantom. Julia's `struct Circle <: Shape`
+        // is the shape that proves it — a Struct child under a Class parent.
+        // Python's _add_typed_connections does the equivalent via name_to_qname.
+        let mut qname_to_type: HashMap<String, &'static str> = HashMap::new();
+        if !type_out.implements.is_empty() || !type_out.extends.is_empty() {
             for c in &result.classes {
                 let nt = super::class_node_type(&c.kind);
                 qname_to_type.insert(c.qualified_name.clone(), nt);
@@ -967,35 +973,27 @@ impl<'a> LoadPipeline<'a> {
                 qname_to_type.insert(i.name.clone(), nt);
             }
             // External stubs we just inserted: traits → Trait, classes → Class/Struct.
-            let ext_trait_type = if graph.has_node_type("Trait") {
-                Some("Trait")
-            } else if graph.has_node_type("Protocol") {
-                Some("Protocol")
-            } else if graph.has_node_type("Interface") {
-                Some("Interface")
-            } else {
-                None
-            };
+            let ext_trait_type = pick(&[
+                ("Trait", has_trait),
+                ("Protocol", has_protocol),
+                ("Interface", has_interface),
+            ]);
             if let Some(nt) = ext_trait_type {
                 for ext in &type_out.external_traits {
                     qname_to_type.insert(ext.qualified_name.clone(), nt);
                     qname_to_type.insert(ext.name.clone(), nt);
                 }
             }
-            let ext_class_type = if graph.has_node_type("Class") {
-                Some("Class")
-            } else if graph.has_node_type("Struct") {
-                Some("Struct")
-            } else {
-                None
-            };
+            let ext_class_type = pick(&[("Class", has_class), ("Struct", has_struct)]);
             if let Some(nt) = ext_class_type {
                 for ext in &type_out.external_classes {
                     qname_to_type.insert(ext.qualified_name.clone(), nt);
                     qname_to_type.insert(ext.name.clone(), nt);
                 }
             }
+        }
 
+        if !type_out.implements.is_empty() {
             let default_src =
                 pick(&[("Class", has_class), ("Struct", has_struct)]).unwrap_or("Class");
             let default_tgt = pick(&[
@@ -1048,15 +1046,49 @@ impl<'a> LoadPipeline<'a> {
             }
         }
         if !type_out.extends.is_empty() {
-            let src = pick(&[("Class", has_class), ("Struct", has_struct)]);
-            if let Some(src) = src {
+            // Route EXTENDS per-row exactly like IMPLEMENTS above. The two
+            // endpoints of an inheritance edge need not share a node type:
+            // julia's `struct Circle <: Shape` is Struct → Class, and passing
+            // one picked label for BOTH ends matched `child_name` against the
+            // wrong type and attached the edge to a `_provisional` Class stub
+            // standing beside the real Struct node.
+            let default_type =
+                pick(&[("Class", has_class), ("Struct", has_struct)]).unwrap_or("Class");
+
+            let mut by_pair: BTreeMap<
+                (&'static str, &'static str),
+                Vec<&super::type_edges::ExtendsEdge>,
+            > = BTreeMap::new();
+            for edge in &type_out.extends {
+                let src = qname_to_type
+                    .get(&edge.child_name)
+                    .copied()
+                    .unwrap_or(default_type);
+                let tgt = qname_to_type
+                    .get(&edge.parent_name)
+                    .copied()
+                    .unwrap_or(default_type);
+                by_pair.entry((src, tgt)).or_default().push(edge);
+            }
+
+            for ((src, tgt), edges) in by_pair {
+                if !graph.has_node_type(src) || !graph.has_node_type(tgt) {
+                    continue;
+                }
+                let owned: Vec<super::type_edges::ExtendsEdge> = edges
+                    .iter()
+                    .map(|e| super::type_edges::ExtendsEdge {
+                        child_name: e.child_name.clone(),
+                        parent_name: e.parent_name.clone(),
+                    })
+                    .collect();
                 maintain::add_connections(
                     graph,
-                    extends_edges_df(&type_out.extends),
+                    extends_edges_df(&owned),
                     "EXTENDS".into(),
                     src.into(),
                     "child_name".into(),
-                    src.into(),
+                    tgt.into(),
                     "parent_name".into(),
                     None,
                     None,