code-gauge 4.5.0 → 4.7.0

package/README.md

@@ -5,7 +5,7 @@
 [![Test](https://github.com/WillBooster/code-gauge/actions/workflows/test.yml/badge.svg)](https://github.com/WillBooster/code-gauge/actions/workflows/test.yml)
 [![Test rust](https://github.com/WillBooster/code-gauge/actions/workflows/test-rust.yml/badge.svg)](https://github.com/WillBooster/code-gauge/actions/workflows/test-rust.yml)
 [![semantic-release](https://img.shields.io/badge/%20%20%F0%9F%93%A6%F0%9F%9A%80-semantic--release-e10079.svg)](https://github.com/semantic-release/semantic-release)
-[![wbfy](https://img.shields.io/badge/wbfy-20.19.0-1e90ff.svg)](https://github.com/WillBooster/shared/tree/main/packages/wbfy)
+[![wbfy](https://img.shields.io/badge/wbfy-20.20.2-1e90ff.svg)](https://github.com/WillBooster/shared/tree/main/packages/wbfy)
 
 A command-line tool that ranks the files of a project by refactoring priority and gates changes
 against metric regressions, built for AI-agent workflows: an agent asked to "refactor this
@@ -157,10 +157,17 @@ The `duplication` section tunes how clones are detected:
 - `minSimilarityPercent` (default 70): blocks the exact pipeline misses are additionally compared by
   similarity (n-gram filtration, then token-level longest-common-subsequence verification, following
   NIL and NiCad), so a near-miss (Type-3) clone with scattered small edits is still reported when
-  both blocks are at least this similar and share more than half of their content-bearing tokens.
-  `100` disables near-miss detection. Applies to within-file detection and to cross-file matching
-  alike; across files, n-grams shared by more than 1000 blocks (syntax boilerplate) are left out of
-  the filtration index so boilerplate cannot make candidate counting quadratic in the block count.
+  both blocks are at least this similar and share more than half of their content-bearing tokens
+  (names and literal values, weighted by rarity so ubiquitous names count less, after ECScan). Two
+  refinements apply the same threshold: blocks whose top-level statements were reordered are also
+  compared in a canonical statement order, and a copy embedded in added code (on one side or both)
+  is matched on its cores, provided the two blocks are within 3 times each other's length (a
+  threshold below 34% widens this to whatever the threshold allows): the chain of n-grams unique to
+  both blocks (only those continuing a diagonal run) is split at gaps of more than 30 tokens, each
+  segment must pass the same threshold on its own, and the verified cores, not the whole blocks,
+  are reported. `100` disables near-miss detection. Applies to within-file detection and to cross-file matching alike; across files,
+  n-grams shared by more than 1000 blocks (syntax boilerplate) are left out of the filtration index
+  so boilerplate cannot make candidate counting quadratic in the block count.
 
 ## Metrics
 
@@ -184,7 +191,8 @@ The `duplication` section tunes how clones are detected:
 - Within-file duplication: copy-pasted blocks matched on normalized tokens (identifiers anonymized
   consistently, literals by kind, and literal-dense data tables excluded unless their values also
   match), with adjacent matches around a small edit merged into gapped (Type-3) clone groups and
-  near-miss (Type-3) clones matched by token-LCS similarity, plus duplicated line count and ratio
+  near-miss (Type-3) clones matched by token-LCS similarity (tolerating reordered statements and
+  copies embedded in added code), plus duplicated line count and ratio
 - Cross-file duplication (via `measureCrossFileDuplication`): copy-pasted blocks shared between
   files, matched with the same normalization (exact, gapped, and near-miss clones) and reported as
   groups with their file locations
@@ -249,3 +257,28 @@ console.log(metrics.maxCognitiveComplexity);
 `detectLanguage(filePath)` maps a file extension to a supported language (or `undefined`) with the
 same table as the CLI, so `measureCode` can be fed arbitrary source files. Unlike the CLI scan, it
 does not skip generated files (e.g. `.d.ts`, `.min.js`) or test files.
+
+### Cloudflare Workers
+
+Workers cannot load native addons, so the package's `workerd` export runs the same API on a
+WebAssembly build of the Rust engine (about 2.3 MB gzipped). Wrangler resolves that export
+automatically, and no `nodejs_compat` flag is needed:
+
+```ts
+import { measureCode } from 'code-gauge';
+
+export default {
+  async fetch(request: Request): Promise<Response> {
+    const code = await request.text();
+    return Response.json(measureCode(code, { language: 'typescript' }));
+  },
+};
+```
+
+Only the programmatic API is available; the CLI needs a file system and `git`. The Workers
+runtime's stack limits nesting to a few thousand syntax-tree levels (the native addon allows
+5,000); deeper sources fail with an error instead of being measured.
+
+In this repository, build the WebAssembly module with `bun run build-wasm`. It downloads a
+[wasi-sdk](https://github.com/WebAssembly/wasi-sdk) release for the grammars' C sources unless
+`WASI_SDK_PATH` points to an installation of the same release.