refactor(ast): split the parse and classification layer into big-code-analysis-ast

.claude/rules/formatting.md

@@ -66,7 +66,7 @@ Three things that version gets right and the hand-rolled probe below
 did not:
 
 - **It feeds rustfmt on stdin.** Given a *path*, rustfmt resolves and
-  recurses into `mod` declarations, so `src/getter.rs` and
+  recurses into `mod` declarations, so `big-code-analysis-ast/src/getter.rs` and
   `src/metrics/cognitive.rs` error out with "file not found for module"
   — which reads exactly like a bail if stderr is discarded. On stdin
   there is nothing to resolve, so those two probe like any other file.
@@ -81,10 +81,11 @@ did not:
   literals**. Do not act on a count from a probe that skips this step.
 - **It probes every arm, not the first.** The bail is match-scoped, so a
   file whose first `match` formats cleanly still hides a later one that
-  does not (`src/getter/c.rs`), and a module whose arms are all
+  does not (`big-code-analysis-ast/src/getter/c.rs`), and a module whose arms are all
   expression-bodied (`… => HalsteadType::Operator,` in
-  `src/getter/go.rs`) has no `=> {` to probe at all. Over `src/getter`
-  the first-arm-only version gave 11 false verdicts out of 18.
+  `big-code-analysis-ast/src/getter/go.rs`) has no `=> {` to probe at
+  all. Over `big-code-analysis-ast/src/getter` the first-arm-only
+  version gave 11 false verdicts out of 18.
 
 ## Two causes, one measurement
 
@@ -114,7 +115,7 @@ never go away.
 That is a deliberate change from how this section used to read. It
 carried a hand-maintained list of files, and that list was wrong twice:
 first by naming only `src/metrics/`, which hid the largest cluster
-(`src/getter/`, 18 of 25 modules) for two revisions of this file, and
+(`big-code-analysis-ast/src/getter/`, 18 of 25 modules) for two revisions of this file, and
 then by going stale the moment #1136 hoisted seven comments out of
 `src/metrics/cognitive/`. A stale list of bailing files reads exactly
 like a clean tree — the failure this whole rule exists to prevent — so
@@ -132,7 +133,7 @@ reported clean, and every one was found by reading the diff instead.
 
 ## How to apply
 
-- After any bulk, scripted, or regex edit under `src/getter/` or
+- After any bulk, scripted, or regex edit under `big-code-analysis-ast/src/getter/` or
   `src/metrics/`, read the resulting diff rather than trusting the fmt
   gate. Check indentation and line length by eye.
 - Line length is worth a direct check, since it is mechanical:

.claude/rules/grammar-dispatch.md

@@ -25,13 +25,13 @@ Confirm every numeric-suffix variant of every matched rule is either
 listed or excluded with a comment:
 
 ```bash
-rg 'Lang::([A-Za-z]+)\b' src/getter/ src/checker/ src/alterator.rs \
+rg 'Lang::([A-Za-z]+)\b' big-code-analysis-ast/src/getter/ big-code-analysis-ast/src/checker/ big-code-analysis-ast/src/alterator.rs \
    src/spaces.rs src/metrics/
 ```
 
 The bug class reaches every match on a grammar rule — `alterator.rs`,
 `spaces.rs`, and `src/metrics/*` are as susceptible as `getter.rs` and
-`checker.rs`. Centralised alias sets live in `src/macros/kind_sets.rs`;
+`checker.rs`. Centralised alias sets live in `big-code-analysis-ast/src/macros/kind_sets.rs`;
 prefer extending those over open-coding a list. When a rule has many
 aliases, prefer one `node.kind()` string comparison over enumerating
 seventeen variants — pay the small runtime cost for forward
@@ -42,7 +42,8 @@ manifests in lockstep (root `Cargo.toml` and `enums/Cargo.toml` — the
 excluded crate cannot inherit the workspace pin), then:
 
 ```bash
-cargo run --manifest-path ./enums/Cargo.toml -- -lrust -o ./src/languages
+cargo run --manifest-path ./enums/Cargo.toml -- \
+    -lrust -o ./big-code-analysis-ast/src/languages
 ```
 
 Stable *named*-node ids are not evidence the ids held. Inserting one
@@ -55,7 +56,7 @@ anonymous terminal renumbers the whole anonymous block after it, so
 A rule whose name begins with `_` (`_string`, `_multiline_string_literal`)
 is hidden: the variant exists in the enum and the parser never emits it.
 Check the `Lang::Variant => "name"` arm in
-`src/languages/language_<lang>.rs` before listing a "looks like an alias"
+`big-code-analysis-ast/src/languages/language_<lang>.rs` before listing a "looks like an alias"
 variant. Keep the defensive arm *and* pin its hidden status with a
 `!ast_has_kind_id(&parser, Lang::HiddenVariant as u16)` assertion naming
 the hidden rule — otherwise a future grammar that promotes the rule
@@ -92,8 +93,8 @@ new one. **Re-derive it rather than trusting the list below** — it moves
 whenever a language is added:
 
 ```bash
-rg -o 'impl_is_else_if_(\w+)!\(\s*(\w+)' -r '$1 $2' src/checker/ --no-filename | sort
-rg -l 'fn is_else_if' src/checker/          # hand-written impls
+rg -o 'impl_is_else_if_(\w+)!\(\s*(\w+)' -r '$1 $2' big-code-analysis-ast/src/checker/ --no-filename | sort
+rg -l 'fn is_else_if' big-code-analysis-ast/src/checker/          # hand-written impls
 ```
 
 | Strategy | Macro | Languages |
@@ -183,7 +184,8 @@ Minimum cross-walk when you edit one:
   into a **wrong count**, which a snapshot diff shows you. This one
   disagrees into an **absent key**: a node the walker promoted but the
   getter left `Unknown` is not a member scope
-  (`SpaceKind::is_member_scope`), so its space serializes no `npm` /
+  (`MemberScopeExt::is_member_scope`, `src/metrics/mod.rs`), so its
+  space serializes no `npm` /
   `npa` block at all — which looks exactly like a language that
   legitimately has no containers. Nothing diffs. §6's "gate all three
   on the same predicate" is this bullet's fix.
@@ -280,7 +282,7 @@ that descends — and test-via-revert that arm alone per
 ## When you fix one language, sweep the rest
 
 Every item above is a per-language failure that almost always exists in
-siblings. `src/languages/` modules are deliberate clones, so the fix for
+siblings. `big-code-analysis-ast/src/languages/` modules are deliberate clones, so the fix for
 one is the audit table for the other twenty. Build that table in the
 issue, land the sibling fixes in one commit so the symmetry is visible
 to a reviewer, and anchor any known-wrong-but-unfixed case with an

.claude/rules/testing.md

@@ -220,7 +220,7 @@ whenever many tests reach a line while all supplying the same value to
 the part that matters, and the tool cannot see the difference because
 which-inputs-varied is not what it measures.
 
-`CommaIndex::splits` (`src/cfg_predicate.rs`) measured 11 of 11 regions
+`CommaIndex::splits` (`big-code-analysis-ast/src/cfg_predicate.rs`) measured 11 of 11 regions
 covered and was entered 150,200 times in one run. Replacing its
 `region.start` lower bound with `0` panics on ordinary input — and
 before #1105 that perturbation failed **none** of the 3,969 tests then

.claude/rules/tool-output.md

@@ -16,7 +16,7 @@ it:
 - `sort | uniq -c` — the counts you are hunting are the largest, and
   they sort **last** unless you passed `-rn`.
 - `rg` over a tree — hits arrive in path order, so a sweep across
-  `src/languages/` or `src/getter/` shows the alphabetically early
+  `big-code-analysis-ast/src/languages/` or `big-code-analysis-ast/src/getter/` shows the alphabetically early
   languages and hides every one after them.
 - `cargo test` / `make pre-commit` — the failure summary is at the end,
   behind all the passing output.

.claude/skills/add-lang/SKILL.md

@@ -25,7 +25,7 @@ Parse `$ARGUMENTS` as: `<lang-name> <grammar-crate>=<version> [<file-ext>...]`
 
 - `<lang-name>` (required): PascalCase enum variant name, e.g. `Go`,
   `Ruby`, `Swift`. Must not collide with existing variants in
-  `src/langs.rs` or `enums/src/languages.rs`.
+  `big-code-analysis-ast/src/langs.rs` or `enums/src/languages.rs`.
 - `<grammar-crate>=<version>` (required): the tree-sitter crate name
   and pinned version, e.g. `tree-sitter-ruby=0.23.1`. The version MUST
   be pinned with `=X.Y.Z` (project convention — see `AGENTS.md`).
@@ -45,8 +45,10 @@ continuing.
 - **No public-API breaks** in unrelated crates. The new language
   variant is itself a public-API addition (acceptable; minor bump);
   do not change other variants or trait signatures.
-- **Pin the grammar version** with `=X.Y.Z` in both `Cargo.toml` and
-  `enums/Cargo.toml`. Never use a range without explicit user
+- **Pin the grammar version** with `=X.Y.Z` in the root `Cargo.toml`
+  (`[workspace.dependencies]`) and in `enums/Cargo.toml`; the
+  `big-code-analysis-ast` manifest inherits the pin with
+  `workspace = true`. Never use a range without explicit user
   approval.
 - **Cross-language parity.** All 12 metric trait impls
   (`Abc`, `Cognitive`, `Cyclomatic`, `Exit`, `Halstead`, `Loc`, `Mi`,
@@ -77,7 +79,7 @@ the user's behalf.
 ### 0b: Validate name and version pin
 
 - Confirm `<lang-name>` is not already a variant in
-  `src/langs.rs` (search for `Lang::<lang-name>`).
+  `big-code-analysis-ast/src/langs.rs` (search for `Lang::<lang-name>`).
 - Confirm `<grammar-crate>=<version>` parses cleanly and that
   `crates.io/crates/<grammar-crate>` actually publishes
   `<version>` — fetch the crate page or `cargo search` to verify.
@@ -99,7 +101,7 @@ symbol-level navigation/editing is the default for all `.rs` edits.
 ## Step 1: Wire up the `enums` codegen helper
 
 The `enums` crate is excluded from the default workspace and exists
-solely to regenerate `src/languages/language_<lang>.rs` from a tree-sitter
+solely to regenerate `big-code-analysis-ast/src/languages/language_<lang>.rs` from a tree-sitter
 grammar's node-kind table. Wire it up first so we can produce the enum
 file before touching the main crate.
 
@@ -182,7 +184,8 @@ empty kind name at one position. If it is missing, add it.
 From the repo root, mirroring `recreate-grammars.sh`:
 
 ```bash
-cargo run --manifest-path ./enums/Cargo.toml -- -l rust -o ./src/languages
+cargo run --manifest-path ./enums/Cargo.toml -- \
+    -l rust -o ./big-code-analysis-ast/src/languages
 cargo fmt --all
 ```
 
@@ -197,8 +200,8 @@ one. The `enums` binary iterates `Lang::into_enum_iter()` and writes
 one file per registered variant. After running, inspect the diff:
 
 ```bash
-git status -- src/languages/
-git diff src/languages/
+git status -- big-code-analysis-ast/src/languages/
+git diff big-code-analysis-ast/src/languages/
 ```
 
 The new `language_<lang>.rs` should appear as a new file. Existing
@@ -212,14 +215,15 @@ no longer matching what the workspace clippy gate expects); fix the
 template — not the emitted output — and re-run codegen. See lesson
 17 in `lessons_learned.md`.
 
-Confirm the new file exists at `src/languages/language_<lang>.rs` and
+Confirm the new file exists at `big-code-analysis-ast/src/languages/language_<lang>.rs` and
 that it begins with `// Code generated; DO NOT EDIT.`.
 
 If the project also depends on C-macro tables for the new language
 (only relevant for C/C++ family preprocessor work), also run:
 
 ```bash
-cargo run --manifest-path ./enums/Cargo.toml -- -l c_macros -o ./src/c_langs_macros
+cargo run --manifest-path ./enums/Cargo.toml -- \
+    -l c_macros -o ./big-code-analysis-ast/src/c_langs_macros
 ```
 
 Most languages do not need this step.
@@ -228,16 +232,44 @@ Most languages do not need this step.
 
 ## Step 2: Wire the grammar into the main crate
 
-### 2a: Add the grammar to root `Cargo.toml`
+### 2a: Wire the grammar into two manifests
 
-Same pinned-version line as 1a, inserted alphabetically among the
-other `tree-sitter-*` deps:
+Since #1376 the grammar crates are dependencies of
+`big-code-analysis-ast`, not of the root crate, so a language needs
+three edits across two manifests. Missing any of them leaves
+`LANG::<New>` returning `LanguageDisabled` in every build, or leaves an
+unused dependency that `cargo +nightly udeps` fails on.
 
-```toml
-tree-sitter-<lang> = "=<version>"
-```
+1. Root `Cargo.toml`, `[workspace.dependencies]` — the version pin,
+   inserted alphabetically among the other `tree-sitter-*` entries
+   (same pinned-version line as 1a):
+
+   ```toml
+   tree-sitter-<lang> = "=<version>"
+   ```
+
+2. `big-code-analysis-ast/Cargo.toml` — the optional dependency and the
+   feature that enables it, plus an entry in `all-languages`:
+
+   ```toml
+   [dependencies]
+   tree-sitter-<lang> = { workspace = true, optional = true }
+
+   [features]
+   all-languages = [..., "<lang>", ...]
+   <lang> = ["dep:tree-sitter-<lang>"]
+   ```
+
+3. Root `Cargo.toml`, `[features]` — the forwarding feature and the
+   matching `all-languages` entry, so the root's feature set stays a
+   superset of the sub-crate's:
+
+   ```toml
+   all-languages = [..., "<lang>", ...]
+   <lang> = ["big-code-analysis-ast/<lang>"]
+   ```
 
-### 2b: Export the generated module from `src/languages/mod.rs`
+### 2b: Export the generated module from `big-code-analysis-ast/src/languages/mod.rs`
 
 ```rust
 pub mod language_<lang>;
@@ -246,7 +278,7 @@ pub use language_<lang>::*;
 
 Insert alphabetically.
 
-### 2c: Add the language definition to `src/langs.rs`
+### 2c: Add the language definition to `big-code-analysis-ast/src/langs.rs`
 
 Append a `mk_langs!` tuple alphabetically:
 
@@ -306,7 +338,7 @@ test that parses a deliberately malformed fixture and asserts
 `blank ≥ 0` and `kind == Unit` at the file level. See lesson 9 in
 `lessons_learned.md` (issue #80, `dc09eb3`).
 
-### 3a: `Checker` impl in `src/checker.rs`
+### 3a: `Checker` impl in `big-code-analysis-ast/src/checker.rs`
 
 Append an `impl Checker for <LangName>Code` block. Required methods:
 
@@ -346,7 +378,7 @@ Append an `impl Checker for <LangName>Code` block. Required methods:
 - `is_primitive` — usually `false` unless the grammar emits a
   primitive-type kind (most don't).
 
-### 3b: `Getter` impl in `src/getter.rs`
+### 3b: `Getter` impl in `big-code-analysis-ast/src/getter.rs`
 
 Append an `impl Getter for <LangName>Code` block. Required methods:
 
@@ -374,7 +406,7 @@ clashed with `use Go::*` in pattern position; the fix was
 `use Go as G;`. Detect the collision proactively after Step 1e:
 
 ```bash
-rg "^\s*<LangName>\s*=" src/languages/language_<lang>.rs
+rg "^\s*<LangName>\s*=" big-code-analysis-ast/src/languages/language_<lang>.rs
 ```
 
 If the search returns a hit, alias the import at the top of the
@@ -392,7 +424,7 @@ assert the load-bearing invariants from lesson 4: run both
 assert that `len(dedupe(ops.operators)) == n1` and
 `len(dedupe(ops.operands)) == n2`.
 
-### 3c: `Alterator` impl in `src/alterator.rs`
+### 3c: `Alterator` impl in `big-code-analysis-ast/src/alterator.rs`
 
 If the language has string/raw-string/char-literal node kinds whose
 default text representation should be preserved verbatim (no whitespace
@@ -451,7 +483,7 @@ rule-name root for which both an unsuffixed variant and one or more
 numbered siblings exist in the generated enum:
 
 ```bash
-LANG_FILE="src/languages/language_<lang>.rs"
+LANG_FILE="big-code-analysis-ast/src/languages/language_<lang>.rs"
 comm -12 \
   <(rg -o '^\s+([A-Z][A-Za-z]*)\d+\s*=' -r '$1' "$LANG_FILE" | sort -u) \
   <(rg -o '^\s+([A-Z][A-Za-z]*)\s*=' -r '$1' "$LANG_FILE" | sort -u)
@@ -465,8 +497,8 @@ names whose every numbered variant is a potential aliasing risk.
 
 For each printed base, confirm that EVERY numbered variant in its
 group holds one of the following in EVERY file that does a `match`
-on the underlying rule (`src/checker.rs`, `src/getter.rs`,
-`src/alterator.rs`, `src/metrics/*.rs`, `src/spaces.rs`):
+on the underlying rule (`big-code-analysis-ast/src/checker.rs`, `big-code-analysis-ast/src/getter.rs`,
+`big-code-analysis-ast/src/alterator.rs`, `src/metrics/*.rs`, `src/spaces.rs`):
 
 1. The variant is explicitly listed in the relevant arm (typically
    alongside its unsuffixed sibling:
@@ -561,7 +593,7 @@ whose name suggests the construct:
 
 ```bash
 rg 'For[A-Z]|While[A-Z]|If[A-Z]|Switch[A-Z]|Conditional|Ternary|Try[A-Z]|Catch[A-Z]|Match[A-Z]|Case[A-Z]' \
-   src/languages/language_<lang>.rs
+   big-code-analysis-ast/src/languages/language_<lang>.rs
 ```
 
 Confirm each hit is either explicitly matched, or explicitly excluded
@@ -1059,16 +1091,17 @@ Before exiting, print a one-screen summary:
 Added <LangName> language support.
 
 Files changed:
-  Cargo.toml
+  Cargo.toml                                               ([workspace.dependencies] pin + forwarding feature)
+  big-code-analysis-ast/Cargo.toml                         (optional dep + feature)
   enums/Cargo.toml
   enums/src/languages.rs
   enums/src/macros.rs
-  src/langs.rs
-  src/languages/mod.rs
-  src/languages/language_<lang>.rs   (generated)
-  src/checker.rs
-  src/getter.rs
-  src/alterator.rs                   (if applicable)
+  big-code-analysis-ast/src/langs.rs
+  big-code-analysis-ast/src/languages/mod.rs
+  big-code-analysis-ast/src/languages/language_<lang>.rs   (generated)
+  big-code-analysis-ast/src/checker.rs
+  big-code-analysis-ast/src/getter.rs
+  big-code-analysis-ast/src/alterator.rs                   (if applicable)
   src/metrics/{abc,cognitive,cyclomatic,nexits,halstead,loc,mi,nargs,nom,npa,npm,wmc}.rs
   big-code-analysis-book/src/languages.md
   *.snap                             (new insta snapshots)
@@ -1083,7 +1116,7 @@ manually.
 **Heads up: mutation testing.** Quarterly mutation testing
 (`.github/workflows/mutation-test.yml`, see
 `docs/development/mutation_testing.md`) runs against
-`src/metrics/`, `src/checker.rs`, and `src/getter.rs`. Within one
+`src/metrics/`, `big-code-analysis-ast/src/checker.rs`, and `big-code-analysis-ast/src/getter.rs`. Within one
 cycle, expect auto-filed issues labelled `mutation-testing` against
 the new language's impls; treat them as standard fix-issue work.
 Escapes mean the per-language test set under-specified the new

.claude/skills/audit-crate/SKILL.md

@@ -19,7 +19,7 @@ titles). Never let an unresolved or empty `$ARGUMENTS` reach a template like
 `audit-state-$ARGUMENTS` — that would write to a malformed memory key.
 
 **Memory-key sanitization**: when `$ARGUMENTS` is a directory path (e.g.,
-`src/languages`), replace `/` with `-` before composing memory keys so the
+`big-code-analysis-ast/src/languages`), replace `/` with `-` before composing memory keys so the
 key is a single flat token (e.g., `audit-state-src-languages`, not
 `audit-state-src/languages`). This avoids backend interpretation of slashes
 as path separators. Apply the same rule to every memory-key reference
@@ -249,7 +249,7 @@ Group every file into one of these categories before auditing:
 
 | Group | Contents |
 |-------|----------|
-| A — Library core | `src/lib.rs` and the modules it exports (`src/languages/`, `src/metrics/`, `src/output/`, `src/spaces.rs`, `src/parser.rs`, `src/checker.rs`, `src/getter.rs`, `src/alterator.rs`, `src/node.rs`, `src/traits.rs`, etc.) |
+| A — Library core | `src/lib.rs` and the modules it exports (`big-code-analysis-ast/src/languages/`, `src/metrics/`, `src/output/`, `src/spaces.rs`, `big-code-analysis-ast/src/parser.rs`, `big-code-analysis-ast/src/checker.rs`, `big-code-analysis-ast/src/getter.rs`, `big-code-analysis-ast/src/alterator.rs`, `big-code-analysis-ast/src/node.rs`, `big-code-analysis-ast/src/traits.rs`, etc.) |
 | B — Binaries | `src/bin/` entries plus the workspace crates `big-code-analysis-cli` and `big-code-analysis-web` when those are the audit target |
 | C — Tests | `tests/` directory |
 | D — Supporting files | `README.md`, examples, `Cargo.toml`, `big-code-analysis-book/`, helper scripts, `.claude/rules/` if present |
@@ -353,7 +353,7 @@ will apply, so the vocabulary must match end-to-end. Mapping rules:
 
 ### Project-Specific (big-code-analysis)
 
-27. Per-language modules under `src/languages/` deliberately mirror each
+27. Per-language modules under `big-code-analysis-ast/src/languages/` deliberately mirror each
     other. Does any change introduce a discrepancy that one language exhibits
     and another does not (different metric formula, different node-type
     handling, different operator/operand classification) without justification?
@@ -554,7 +554,7 @@ last_model: <model-id>
 
 ## File Coverage
 src/lib.rs | full | 2026-04-25 | 3 findings | claude-opus-4-7
-src/languages/language_rust.rs | partial | 2026-04-25 | 1 finding | claude-opus-4-7
+big-code-analysis-ast/src/languages/language_rust.rs | partial | 2026-04-25 | 1 finding | claude-opus-4-7
 src/metrics/halstead.rs | none | - | - | -
 tests/parser.rs | full | 2026-04-25 | 0 findings | claude-sonnet-4-6
 ```