rgapi 0.2.0__tar.gz → 0.2.2__tar.gz

{rgapi-0.2.0 → rgapi-0.2.2}/.github/workflows/ci.yml

@@ -12,11 +12,12 @@ jobs:
     steps:
       - uses: actions/checkout@v7
       - uses: dtolnay/rust-toolchain@stable
-      - run: cargo test
       - uses: actions/setup-python@v7
         with:
           python-version: '3.12'
+      - run: cargo test
       - run: pip install -e '.[dev]'
+      - run: cargo develop
       - run: pytest -q
 
   build:
@@ -61,7 +62,7 @@ jobs:
       - uses: dtolnay/rust-toolchain@stable
       - id: crates-auth
         uses: rust-lang/crates-io-auth-action@v1
-      - run: cargo publish
+      - run: cargo publish -p rgapi
         env:
           CARGO_REGISTRY_TOKEN: ${{ steps.crates-auth.outputs.token }}
       - uses: actions/download-artifact@v8

{rgapi-0.2.0 → rgapi-0.2.2}/Cargo.lock

@@ -11,6 +11,21 @@ dependencies = [
  "memchr",
 ]
 
+[[package]]
+name = "bit-set"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "08807e080ed7f9d5433fa9b275196cfc35414f66a0c79d864dc51a0d825231a3"
+dependencies = [
+ "bit-vec",
+]
+
+[[package]]
+name = "bit-vec"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5e764a1d40d510daf35e07be9eb06e75770908c27d411ee6c92109c9840eaaf7"
+
 [[package]]
 name = "bstr"
 version = "1.13.1"
@@ -18,7 +33,6 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6bb31b46c14244e20ee9984b11bf5c992b91fb6939fea616e3512c8baecdbe5f"
 dependencies = [
  "memchr",
- "regex-automata",
  "serde_core",
 ]
 
@@ -91,6 +105,17 @@ dependencies = [
  "encoding_rs",
 ]
 
+[[package]]
+name = "fancy-regex"
+version = "0.19.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d301f5bf187b3c295fce6468d3875037a0bccc5f6b151c63cac2f85babf21912"
+dependencies = [
+ "bit-set",
+ "regex-automata",
+ "regex-syntax",
+]
+
 [[package]]
 name = "globset"
 version = "0.4.20"
@@ -113,19 +138,6 @@ dependencies = [
  "memchr",
 ]
 
-[[package]]
-name = "grep-regex"
-version = "0.1.14"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0ce0c256c3ad82bcc07b812c15a45ec1d398122e8e15124f96695234db7112ef"
-dependencies = [
- "bstr",
- "grep-matcher",
- "log",
- "regex-automata",
- "regex-syntax",
-]
-
 [[package]]
 name = "grep-searcher"
 version = "0.1.17"
@@ -171,9 +183,9 @@ checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682"
 
 [[package]]
 name = "libc"
-version = "0.2.189"
+version = "0.2.190"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3eaf3ede3fee6db1a4c2ee091bf8a8b4dccdc6d17f656fb07896ee72867612f2"
+checksum = "ce5d3ddc6d3fa000eb1536d85e147bfe31aacaba692ed6a876f95cb7c855be78"
 
 [[package]]
 name = "log"
@@ -225,9 +237,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3"
-version = "0.29.2"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4688ddedf473e32662b9b067670129a8afb8c18e351482c70d62ba4a88171e8b"
+checksum = "700d18fa267b73b9b521fd7e13580e2f446916f176cacee1ab63fcc8191f1655"
 dependencies = [
  "libc",
  "once_cell",
@@ -239,18 +251,18 @@ dependencies = [
 
 [[package]]
 name = "pyo3-build-config"
-version = "0.29.2"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f41027e41b4bd03f6e60f9f417fe24a6341a6bb744edd62b6f709f2a52ea30e9"
+checksum = "7b3fc0c4d08f6bb10e71fe39dfb9e2f59c6eb6854e22ec8092f50c69a4499adb"
 dependencies = [
  "target-lexicon",
 ]
 
 [[package]]
 name = "pyo3-ffi"
-version = "0.29.2"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e591a95526fead067432c3b3a33fc74770b87b1e04e73671090d9c2055a2b327"
+checksum = "dfc0b8e19df29aad7086cf977bb0c2a2f143e30567eb113e9cf72b62ca698330"
 dependencies = [
  "libc",
  "pyo3-build-config",
@@ -258,9 +270,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros"
-version = "0.29.2"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "73225868fc1cd84eef2c3c230ddb91273bf1de46aeb8a4248da76d32a0924a1c"
+checksum = "6100e8a4b5eba53afaa5ed078364851a0b2499c553a44c31026b929049b49dc6"
 dependencies = [
  "proc-macro2",
  "pyo3-macros-backend",
@@ -270,9 +282,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros-backend"
-version = "0.29.2"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "571575aa3749fa6216757dd47d2a3e7ef360f329a40f0666a9fbd14889024952"
+checksum = "6143877a16e82b5a727b7127ff4cd86858a24a28d745d72f43e6f227c7b1bdb3"
 dependencies = [
  "heck",
  "proc-macro2",
@@ -308,19 +320,29 @@ checksum = "d6f6ff9a378485b298a5286656da665ba74413d36db0979633275d2e708145d4"
 
 [[package]]
 name = "rgapi"
-version = "0.2.0"
+version = "0.2.2"
 dependencies = [
  "crc32fast",
+ "fancy-regex",
  "globset",
  "grep-matcher",
- "grep-regex",
  "grep-searcher",
  "ignore",
- "pyo3",
+ "regex-syntax",
  "serde",
  "serde_json",
 ]
 
+[[package]]
+name = "rgapi-py"
+version = "0.2.2"
+dependencies = [
+ "grep-matcher",
+ "pyo3",
+ "pyo3-build-config",
+ "rgapi",
+]
+
 [[package]]
 name = "rustversion"
 version = "1.0.23"

{rgapi-0.2.0 → rgapi-0.2.2}/Cargo.toml

@@ -1,37 +1,44 @@
+[workspace]
+members = ["py"]
+
+[workspace.package]
+version = "0.2.2"
+edition = "2024"
+
+[workspace.dependencies]
+grep-matcher = ">=0.1.8"
+
 [package]
 name = "rgapi"
-version = "0.2.0"
-edition = "2024"
+version.workspace = true
+edition.workspace = true
 rust-version = "1.91"
 license = "Apache-2.0"
 description = "Python API for ripgrep-style file walking and searching"
 repository = "https://github.com/AnswerDotAI/rgapi"
 homepage = "https://github.com/AnswerDotAI/rgapi"
 documentation = "https://github.com/AnswerDotAI/rgapi"
-readme = "README.md"
 
 [lib]
 name = "rgapi"
-crate-type = ["cdylib", "rlib"]
 
 [dependencies]
 crc32fast = "1"
+fancy-regex = "0.19"
 globset = ">=0.4.18"
-grep-matcher = ">=0.1.8"
-grep-regex = ">=0.1.14"
+grep-matcher.workspace = true
 grep-searcher = ">=0.1.16"
 ignore = ">=0.4.26"
-pyo3 = { version = ">=0.28", optional = true }
+regex-syntax = "0.8"
 serde = { version = "=1", features = ["derive"] }
 serde_json = "=1"
 
-[features]
-python = ["dep:pyo3"]
-extension-module = ["python", "pyo3/extension-module"]
-
 [lints.clippy]
 too_many_arguments = "allow"
 
+[profile.test]
+inherits = "release"
+
 [profile.release]
 lto = false
 codegen-units = 16
@@ -39,6 +46,9 @@ codegen-units = 16
 [profile.release.package.rgapi]
 incremental = true
 
+[profile.release.package.rgapi-py]
+incremental = true
+
 [profile.dist]
 inherits = "release"
 lto = true
@@ -48,3 +58,6 @@ strip = true
 
 [profile.dist.package.rgapi]
 incremental = false
+
+[profile.dist.package.rgapi-py]
+incremental = false

{rgapi-0.2.0 → rgapi-0.2.2}/DEV.md

@@ -5,10 +5,11 @@
 ## Layout
 
 ```text
-src/walk.rs       ignore/globset/grep-regex-backed path walking and filtering
-src/search.rs     grep-regex/grep-searcher-backed searching
+src/matcher.rs    shared fancy-regex engine, smart-case policy, and grep-matcher adapter
+src/walk.rs       ignore/globset-backed path walking and filtering
+src/search.rs     grep-searcher-backed searching
 src/block.rs      blank-line-delimited block grouping, matching, and block context
-src/python.rs     PyO3 classes and private core functions
+py/src/lib.rs     PyO3 classes and private core functions
 python/rgapi/     public Python wrappers over `rgapi._core`, plus the `rgapi-nbrg` CLI
 tests/            pytest coverage for the Python API
 ```
@@ -18,11 +19,11 @@ The public Python API lives in `python/rgapi/__init__.py`. The extension module
 ## Commands
 
 ```bash
-maturin develop
+cargo develop
 pytest -q
 ```
 
-Python tests (`tests/`) run against the built extension. `cargo test` covers the Rust-only API, and takes no feature flags: `extension-module` stops pyo3 linking libpython, which a test binary needs. For a fast local loop use `maturin develop && pytest -q`. Run `cargo fmt --check` and `cargo check --all-features` for Rust-only edits. Run `chkstyle` after Python edits once tests pass.
+`cargo develop` and bare `cargo test` use the same Cargo profile and share the core and binding libraries. Unit tests also build a separate `cfg(test)` executable. The published `rgapi` crate has no Python dependency; the unpublished `rgapi-py` crate in `py/` builds the extension. Run `cargo fmt` after Rust edits and `chkstyle` before committing Python edits.
 
 ## Release
 
@@ -30,14 +31,18 @@ The canonical version lives in `Cargo.toml`. `pyproject.toml` gets the Python pa
 
 Release flow is: release first, then bump - `ship-release` does both.
 
-1. Run `maturin develop && pytest -q`.
-2. Confirm the release version in `Cargo.toml` (`[package].version`).
+1. Run `cargo develop && pytest -q`.
+2. Confirm the release version in `Cargo.toml` (`[workspace.package].version`).
 3. Run `ship-release`. It tags `v<version>`, pushes branch and tag (CI builds and publishes), then bumps `Cargo.toml`, refreshes the editable install, and pushes the bump without a tag.
 
 The GitHub workflow builds wheels for Python 3.10-3.13 on Linux and macOS and publishes the Rust crate, GitHub release artifacts, and PyPI package when a `v*` tag is pushed.
 
 ## Design notes
 
+All user-supplied regexes use the shared `RegexMatcher` in `src/matcher.rs`, backed by `fancy-regex`. There is no separate fallback engine or native C dependency. The adapter implements `grep_matcher::Matcher` so `grep-searcher` continues to handle file scanning, line numbers, context, and binary detection. `compile_regex` returns rgapi's own exported `RegexMatcher` type.
+
+The matcher centralizes smart-case analysis and line-ending policy. Line-safe patterns can search a whole buffer; patterns that can consume line endings or inspect neighbouring lines are evaluated on individual lines. Notebook multiline matching instead uses the whole cell. Filename/path regexes match their whole input. Backtracking-limit errors propagate through the existing result channels, including directory-pruning errors; cancellation must not hide a queued error from async consumers.
+
 Python discovery and `paths=True` results contain absolute `pathlib.Path` objects. Structured search rows hold base-relative string labels with `/` separators. Traversal uses `ignore::WalkParallel`, so result order is not part of the API contract. Search results are structured rows; collected result lists use rg-style `str()` and notebook display. `SearchLine.lnhash` is computed with the same CRC-32-based line-content hash format as exhash (`lineno|hash|`, low 12 bits of CRC-32 over the line's UTF-8 bytes, encoded as two Base64url characters); `lnhashs=True` only changes row display, not `line_number` or matching behavior. Path regexes filter returned/searched paths; `skip_dir` and `skip_dir_re` prune traversal through `ignore::WalkBuilder::filter_entry`. Depth, size, filesystem, hidden, and ignore options use `ignore::WalkBuilder` settings. The `ignore` walker follows every root it is given. Discovery checks each root with `symlink_metadata`. The worker sends each unfollowed link root as a result before walking the other roots. This also supports dangling roots, which the underlying walker would reject. `ls` sets `walk_root_links`, which gives a root link to a directory to the walker. The walker then reports paths under the link. Other discovery roots use absolute paths without canonicalizing. Content searches retain canonical root resolution. `rg_iter` exposes the same parallel search stream that `rg` collects by default; `paths=True` and `count=True` consume that stream with different reducers. Text search skips binary files and invalid UTF-8 content.
 
 Streaming engine: `walk.rs` owns the generic machinery. `StreamIter<T>` is the worker-thread-plus-bounded-channel iterator (`sync_channel(8192)`, so producers block rather than buffer without limit when a consumer lags), and `spawn_walk` owns the shared scaffold: walker config, panic catching, cancel flag, and worker thread. `rg_iter` (`T = SearchLine`), `block_iter` (`T = SearchBlock`), `nb_iter` (`T = NbCell`), and `find_iter` (`T = PathBuf`, the path walk) plug entry closures into that engine. Block search reads each file once, searches it once, groups nonblank lines into blocks, maps matching lines to their blocks, and expands context by block index. Each `SearchBlock` carries numeric boundaries plus hashes for its first and last source lines. Python keeps both and chooses the displayed address without another file read.

{rgapi-0.2.0 → rgapi-0.2.2}/PKG-INFO

@@ -1,16 +1,16 @@
 Metadata-Version: 2.4
 Name: rgapi
-Version: 0.2.0
+Version: 0.2.2
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: Implementation :: CPython
 Requires-Dist: fastcore>=2.2.29
+Requires-Dist: fastws-cli>=0.0.20 ; extra == 'dev'
 Requires-Dist: fastship>=0.0.12 ; extra == 'dev'
 Requires-Dist: maturin>=1.0,<2.0 ; extra == 'dev'
 Requires-Dist: pytest ; extra == 'dev'
 Provides-Extra: dev
 License-File: LICENSE
 Summary: Python API for ripgrep-style file walking and searching
-Home-Page: https://github.com/AnswerDotAI/rgapi
 Author-email: Jeremy Howard <j@fast.ai>
 License-Expression: Apache-2.0
 Requires-Python: >=3.10
@@ -23,7 +23,7 @@ Project-URL: Repository, https://github.com/AnswerDotAI/rgapi
 
 `rgapi` provides `fd`-style file discovery and `rg`-style text search from Python without starting a shell command.
 
-It uses the same `ignore`, `grep-regex`, and `grep-searcher` crates that ripgrep uses for walking, regex matching, and file scanning. Walking and searching run in parallel by default. Most expensive work stays in Rust.
+It uses ripgrep's `ignore` and `grep-searcher` crates for walking and file scanning, with the pure-Rust `fancy-regex` engine for regex matching. Walking and searching run in parallel by default. Most expensive work stays in Rust.
 
 ## Overview
 
@@ -108,6 +108,16 @@ Set `min_depth` and `max_depth` to bound recursion. `max_filesize` skips files a
 
 Search is case-sensitive by default, matching `rg`. Use `smart_case=True` for `rg --smart-case` behaviour. Use `case_sensitive=False` to force case-insensitive matching.
 
+Patterns support backreferences and lookaround everywhere, including filename and path filters:
+
+```python
+rg(r"\b(\w+)\s+\1\b", ".")  # repeated words
+rg(r"(?<=TODO: )\w+", ".")   # word following TODO:
+fd(".", pattern=r"^(\w+)_\1\.py$")  # names such as utils_utils.py
+```
+
+All regex parameters use `fancy-regex`; no engine-selection flag or C library is required. Ordinary patterns use its non-backtracking engine. Patterns requiring backtracking have a bounded backtracking budget; exceeding it raises `ValueError`, rather than silently skipping a file or path.
+
 Each `SearchLine` has these fields:
 
 ```text
@@ -127,7 +137,9 @@ Pass `lnhashs=True` to `rg` or `rg_iter` to display hash addresses instead of li
 
 For other result forms, use `rg(..., paths=True)` to return unique matched paths or `rg(..., count=True)` to count match spans. `paths` and `count` cannot both be set.
 
-`^` and `$` match at the start and end of each line. A line ends at `\n` or `\r\n`. `$` also matches before a `\r` that has no `\n` after it. `rg`, `rgstr`, and `nbrg` without `multiline=True` raise `ValueError` for a pattern that contains a literal `\n` or `\r`.
+`^` and `$` match at the start and end of each line. A line ends at `\n` or `\r\n`. `$` also matches before a `\r` that has no `\n` after it. `rg`, `rgstr`, and `compile` reject patterns containing a literal `\n` or `\r`. `nbrg` allows them with `multiline=True`.
+
+Content searches and compiled content matchers are confined to individual lines: whitespace classes, backreferences and lookaround cannot cross a line ending, and `\A`/`\z` refer to each line. With `nbrg(..., multiline=True)`, `\A`/`\z` refer to the whole cell, while `^`/`$` still anchor lines. Filename and path filters match the full name or path instead.
 
 ### Path results
 
@@ -159,6 +171,8 @@ Set `timeout_ms` on `rg`, `fd`, `walk`, or `ls` to stop at a deadline and return
 
 `complete` is true exactly when `stop_reason` is `None`. `count=True` returns a plain integer without a completion flag. It cannot be combined with `timeout_ms`.
 
+Deadlines stop result collection and request worker cancellation. Workers cannot interrupt a regex evaluation or filesystem call already in progress; a timeout is not a hard per-match execution limit.
+
 ### Context lines
 
 `before_context`, `after_context`, and `context` correspond to `rg -B`, `rg -A`, and `rg -C`. Files containing NUL bytes or invalid UTF-8 are skipped.
@@ -190,7 +204,9 @@ nbrg("read_csv", ".")                  # cells whose source matches, across all
 nbrg("read_csv", ".", cell_context=1)  # also include neighbouring cells as context
 ```
 
-Notebook discovery, parsing, and matching run together in one parallel Rust pass. Matching uses `rg`'s regex engine with the same `case_sensitive` and `smart_case` behaviour. `nbrg` accepts the discovery filters from `fd` and `rg`, including `include`, `exclude`, `glob`, `hidden`, `max_depth`, and `skip_dir`.
+Notebook discovery, parsing, and matching run together in one parallel Rust pass. Matching uses the same `fancy-regex` engine as `rgapi.rg`, including backreferences and lookaround, with the same `case_sensitive` and `smart_case` behaviour. `nbrg` accepts the discovery filters from `fd` and `rg`, including `include`, `exclude`, `glob`, `hidden`, `max_depth`, and `skip_dir`.
+
+Matching is line-oriented by default. Use `nbrg(r"(foo)\n\1", ".", multiline=True)` to match across lines within a cell. Matches never cross cell boundaries.
 
 `nbrg` returns `NbResults`, a list of `NbCell`. Each `NbCell` has:
 
@@ -248,7 +264,7 @@ async for row in rga_iter("TODO", "."): ...
 
 Walking and searching use Rust threads, without `asyncio.to_thread` or the event loop's executor. A callback uses `loop.call_soon_threadsafe` to complete the awaited future or supply results to the generator's queue. The event loop remains unblocked, with normal `contextvars` behaviour.
 
-Cancellation stops the Rust workers within about one row. This includes timeouts from `asyncio.wait_for` or `asyncio.timeout`. It also includes task cancellation, such as starlette cancelling a disconnected client's request.
+Cancellation is checked between rows and files; it does not interrupt a regex evaluation or filesystem call already in progress. This includes timeouts from `asyncio.wait_for` or `asyncio.timeout`. It also includes task cancellation, such as starlette cancelling a disconnected client's request.
 
 Wrap an async iterator in `contextlib.aclosing` when leaving its loop early. `break` alone delays generator finalization until garbage collection. The context manager provides prompt cleanup:
 

{rgapi-0.2.0 → rgapi-0.2.2}/README.md

@@ -2,7 +2,7 @@
 
 `rgapi` provides `fd`-style file discovery and `rg`-style text search from Python without starting a shell command.
 
-It uses the same `ignore`, `grep-regex`, and `grep-searcher` crates that ripgrep uses for walking, regex matching, and file scanning. Walking and searching run in parallel by default. Most expensive work stays in Rust.
+It uses ripgrep's `ignore` and `grep-searcher` crates for walking and file scanning, with the pure-Rust `fancy-regex` engine for regex matching. Walking and searching run in parallel by default. Most expensive work stays in Rust.
 
 ## Overview
 
@@ -87,6 +87,16 @@ Set `min_depth` and `max_depth` to bound recursion. `max_filesize` skips files a
 
 Search is case-sensitive by default, matching `rg`. Use `smart_case=True` for `rg --smart-case` behaviour. Use `case_sensitive=False` to force case-insensitive matching.
 
+Patterns support backreferences and lookaround everywhere, including filename and path filters:
+
+```python
+rg(r"\b(\w+)\s+\1\b", ".")  # repeated words
+rg(r"(?<=TODO: )\w+", ".")   # word following TODO:
+fd(".", pattern=r"^(\w+)_\1\.py$")  # names such as utils_utils.py
+```
+
+All regex parameters use `fancy-regex`; no engine-selection flag or C library is required. Ordinary patterns use its non-backtracking engine. Patterns requiring backtracking have a bounded backtracking budget; exceeding it raises `ValueError`, rather than silently skipping a file or path.
+
 Each `SearchLine` has these fields:
 
 ```text
@@ -106,7 +116,9 @@ Pass `lnhashs=True` to `rg` or `rg_iter` to display hash addresses instead of li
 
 For other result forms, use `rg(..., paths=True)` to return unique matched paths or `rg(..., count=True)` to count match spans. `paths` and `count` cannot both be set.
 
-`^` and `$` match at the start and end of each line. A line ends at `\n` or `\r\n`. `$` also matches before a `\r` that has no `\n` after it. `rg`, `rgstr`, and `nbrg` without `multiline=True` raise `ValueError` for a pattern that contains a literal `\n` or `\r`.
+`^` and `$` match at the start and end of each line. A line ends at `\n` or `\r\n`. `$` also matches before a `\r` that has no `\n` after it. `rg`, `rgstr`, and `compile` reject patterns containing a literal `\n` or `\r`. `nbrg` allows them with `multiline=True`.
+
+Content searches and compiled content matchers are confined to individual lines: whitespace classes, backreferences and lookaround cannot cross a line ending, and `\A`/`\z` refer to each line. With `nbrg(..., multiline=True)`, `\A`/`\z` refer to the whole cell, while `^`/`$` still anchor lines. Filename and path filters match the full name or path instead.
 
 ### Path results
 
@@ -138,6 +150,8 @@ Set `timeout_ms` on `rg`, `fd`, `walk`, or `ls` to stop at a deadline and return
 
 `complete` is true exactly when `stop_reason` is `None`. `count=True` returns a plain integer without a completion flag. It cannot be combined with `timeout_ms`.
 
+Deadlines stop result collection and request worker cancellation. Workers cannot interrupt a regex evaluation or filesystem call already in progress; a timeout is not a hard per-match execution limit.
+
 ### Context lines
 
 `before_context`, `after_context`, and `context` correspond to `rg -B`, `rg -A`, and `rg -C`. Files containing NUL bytes or invalid UTF-8 are skipped.
@@ -169,7 +183,9 @@ nbrg("read_csv", ".")                  # cells whose source matches, across all
 nbrg("read_csv", ".", cell_context=1)  # also include neighbouring cells as context
 ```
 
-Notebook discovery, parsing, and matching run together in one parallel Rust pass. Matching uses `rg`'s regex engine with the same `case_sensitive` and `smart_case` behaviour. `nbrg` accepts the discovery filters from `fd` and `rg`, including `include`, `exclude`, `glob`, `hidden`, `max_depth`, and `skip_dir`.
+Notebook discovery, parsing, and matching run together in one parallel Rust pass. Matching uses the same `fancy-regex` engine as `rgapi.rg`, including backreferences and lookaround, with the same `case_sensitive` and `smart_case` behaviour. `nbrg` accepts the discovery filters from `fd` and `rg`, including `include`, `exclude`, `glob`, `hidden`, `max_depth`, and `skip_dir`.
+
+Matching is line-oriented by default. Use `nbrg(r"(foo)\n\1", ".", multiline=True)` to match across lines within a cell. Matches never cross cell boundaries.
 
 `nbrg` returns `NbResults`, a list of `NbCell`. Each `NbCell` has:
 
@@ -227,7 +243,7 @@ async for row in rga_iter("TODO", "."): ...
 
 Walking and searching use Rust threads, without `asyncio.to_thread` or the event loop's executor. A callback uses `loop.call_soon_threadsafe` to complete the awaited future or supply results to the generator's queue. The event loop remains unblocked, with normal `contextvars` behaviour.
 
-Cancellation stops the Rust workers within about one row. This includes timeouts from `asyncio.wait_for` or `asyncio.timeout`. It also includes task cancellation, such as starlette cancelling a disconnected client's request.
+Cancellation is checked between rows and files; it does not interrupt a regex evaluation or filesystem call already in progress. This includes timeouts from `asyncio.wait_for` or `asyncio.timeout`. It also includes task cancellation, such as starlette cancelling a disconnected client's request.
 
 Wrap an async iterator in `contextlib.aclosing` when leaving its loop early. `break` alone delays generator finalization until garbage collection. The context manager provides prompt cleanup:
 

rgapi-0.2.2/py/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "rgapi-py"
+version.workspace = true
+edition.workspace = true
+publish = false
+
+[lib]
+crate-type = ["cdylib", "rlib"]
+test = false
+
+[dependencies]
+rgapi = { path = ".." }
+grep-matcher.workspace = true
+pyo3 = { version = ">=0.29", features = ["extension-module"] }
+
+[build-dependencies]
+pyo3-build-config = "0.29"

rgapi-0.2.2/py/build.rs

@@ -0,0 +1 @@
+fn main() { pyo3_build_config::add_extension_module_link_args(); }