fuzzgpu 0.2.0__tar.gz → 0.4.0__tar.gz

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/Cargo.lock

@@ -308,9 +308,9 @@ dependencies = [
 
 [[package]]
 name = "either"
-version = "1.17.0"
+version = "1.18.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9e5e8f6c15a24b9a3ee5efec809ccd006d3b30e8b3bb63c39af737c7f87daa1d"
+checksum = "252afb9ae5eaa683babdc6a068b3f5726eb19e05070c731f9b2a23a7c3e8ed34"
 
 [[package]]
 name = "equivalent"
@@ -372,7 +372,7 @@ dependencies = [
 
 [[package]]
 name = "fuzzgpu-core"
-version = "0.2.0"
+version = "0.4.0"
 dependencies = [
  "bytemuck",
  "criterion",
@@ -386,7 +386,7 @@ dependencies = [
 
 [[package]]
 name = "fuzzgpu-python"
-version = "0.2.0"
+version = "0.4.0"
 dependencies = [
  "fuzzgpu-core",
  "log",
@@ -518,15 +518,6 @@ dependencies = [
  "hashbrown 0.17.1",
 ]
 
-[[package]]
-name = "indoc"
-version = "2.0.7"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "79cf5c93f93228cf8efb3ba362535fb11199ac548a09ce117c9b1adc3030d706"
-dependencies = [
- "rustversion",
-]
-
 [[package]]
 name = "is-terminal"
 version = "0.4.17"
@@ -654,9 +645,9 @@ dependencies = [
 
 [[package]]
 name = "log"
-version = "0.4.33"
+version = "0.4.34"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0ceec5bc11778974d1bcb055b18002eba7f4b3518b6a0081b3af5f21666da9ad"
+checksum = "f9f8bd3e56ce4dfc153cf470fffbfa98c7620958b312ca5c3a4b8d5181fd13c6"
 
 [[package]]
 name = "matrixmultiply"
@@ -674,20 +665,11 @@ version = "2.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cf8baf1c55e62ffcace7a9f06f4bd9cd3f0c4beb022d3b367256b91b87513d98"
 
-[[package]]
-name = "memoffset"
-version = "0.9.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "488016bfae457b036d996092f6cb448677611ce4449e970ceaf42695203f218a"
-dependencies = [
- "autocfg",
-]
-
 [[package]]
 name = "naga"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "23bf0a141a9ab6f07dbb492db53245e464bc9db42f407772d9ae03d83a2c1033"
+checksum = "a616d2fb8c89516ac2723a581f69d6c18576046bed761bd6b305e5618e6ae130"
 dependencies = [
  "arrayvec",
  "bit-set 0.10.0",
@@ -711,9 +693,9 @@ dependencies = [
 
 [[package]]
 name = "naga-types"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "658200ddc25c6c7b860747516d132d1b284c0fafb7a380233acee9a72fb30e11"
+checksum = "590afbf58a6f4f62873cd5cff4468061844bafa1cdf399cc954537c22d768d49"
 dependencies = [
  "hashbrown 0.17.1",
  "indexmap",
@@ -723,9 +705,9 @@ dependencies = [
 
 [[package]]
 name = "ndarray"
-version = "0.16.1"
+version = "0.17.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "882ed72dce9365842bf196bdeedf5055305f11fc8c03dee7bb0194a6cad34841"
+checksum = "520080814a7a6b4a6e9070823bb24b4531daac8c4627e08ba5de8c5ef2f2752d"
 dependencies = [
  "matrixmultiply",
  "num-complex",
@@ -775,9 +757,9 @@ dependencies = [
 
 [[package]]
 name = "numpy"
-version = "0.23.0"
+version = "0.29.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b94caae805f998a07d33af06e6a3891e38556051b8045c615470a71590e13e78"
+checksum = "6a5b15d63a5ff39e378daed0e1340d3a5964703ea9712eb09a0dc66fade996f4"
 dependencies = [
  "libc",
  "ndarray",
@@ -785,6 +767,7 @@ dependencies = [
  "num-integer",
  "num-traits",
  "pyo3",
+ "pyo3-build-config",
  "rustc-hash 2.1.3",
 ]
 
@@ -888,9 +871,9 @@ checksum = "d6790f58c7ff633d8771f42965289203411a5e5c68388703c06e14f24770b41e"
 
 [[package]]
 name = "ordered-float"
-version = "5.3.0"
+version = "5.5.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b7d950ca161dc355eaf28f82b11345ed76c6e1f6eb1f4f4479e0323b9e2fbd0e"
+checksum = "8c7c9e0d9b23589f26070720bac724174bfec1083e82f7854cdd0267518343c0"
 dependencies = [
  "num-traits",
 ]
@@ -1030,37 +1013,32 @@ dependencies = [
 
 [[package]]
 name = "pyo3"
-version = "0.23.5"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7778bffd85cf38175ac1f545509665d0b9b92a198ca7941f131f85f7a4f9a872"
+checksum = "4688ddedf473e32662b9b067670129a8afb8c18e351482c70d62ba4a88171e8b"
 dependencies = [
- "cfg-if",
- "indoc",
  "libc",
- "memoffset",
  "once_cell",
  "portable-atomic",
  "pyo3-build-config",
  "pyo3-ffi",
  "pyo3-macros",
- "unindent",
 ]
 
 [[package]]
 name = "pyo3-build-config"
-version = "0.23.5"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "94f6cbe86ef3bf18998d9df6e0f3fc1050a8c5efa409bf712e661a4366e010fb"
+checksum = "f41027e41b4bd03f6e60f9f417fe24a6341a6bb744edd62b6f709f2a52ea30e9"
 dependencies = [
- "once_cell",
  "target-lexicon",
 ]
 
 [[package]]
 name = "pyo3-ffi"
-version = "0.23.5"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e9f1b4c431c0bb1c8fb0a338709859eed0d030ff6daa34368d3b152a63dfdd8d"
+checksum = "e591a95526fead067432c3b3a33fc74770b87b1e04e73671090d9c2055a2b327"
 dependencies = [
  "libc",
  "pyo3-build-config",
@@ -1068,9 +1046,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros"
-version = "0.23.5"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fbc2201328f63c4710f68abdf653c89d8dbc2858b88c5d88b0ff38a75288a9da"
+checksum = "73225868fc1cd84eef2c3c230ddb91273bf1de46aeb8a4248da76d32a0924a1c"
 dependencies = [
  "proc-macro2",
  "pyo3-macros-backend",
@@ -1080,13 +1058,12 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros-backend"
-version = "0.23.5"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fca6726ad0f3da9c9de093d6f116a93c1a38e417ed73bf138472cf4064f72028"
+checksum = "571575aa3749fa6216757dd47d2a3e7ef360f329a40f0666a9fbd14889024952"
 dependencies = [
  "heck",
  "proc-macro2",
- "pyo3-build-config",
  "quote",
  "syn 2.0.119",
 ]
@@ -1411,9 +1388,9 @@ dependencies = [
 
 [[package]]
 name = "target-lexicon"
-version = "0.12.16"
+version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "61c41af27dd6d1e27b1b16b489db798443478cef1f06a660c96db617ba5de3b1"
+checksum = "adb6935a6f5c20170eeceb1a3835a49e12e19d792f6dd344ccc76a985ca5a6ca"
 
 [[package]]
 name = "tempfile"
@@ -1485,12 +1462,6 @@ version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b4ac048d71ede7ee76d585517add45da530660ef4390e49b098733c6e897f254"
 
-[[package]]
-name = "unindent"
-version = "0.2.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7264e107f553ccae879d21fbea1d6724ac785e8c3bfc762137959b5802826ef3"
-
 [[package]]
 name = "version_check"
 version = "0.9.5"
@@ -1604,9 +1575,9 @@ dependencies = [
 
 [[package]]
 name = "wgpu"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6d8f4bd44d92da5270f03409dba9f952dab24f128e05d6a554926101d1bf9114"
+checksum = "527ccdf43dd5b2e8676eed9984ce00e2bbb0a1b85b70c1969dcb6cd2eb55ab9e"
 dependencies = [
  "arrayvec",
  "bitflags",
@@ -1634,9 +1605,9 @@ dependencies = [
 
 [[package]]
 name = "wgpu-core"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "08763620e76fc980bca7bf84de82568614487a53172dd968d89187282eb87fa2"
+checksum = "14c018fce9b6270aa203c2fdd56f3cce996713534bd757e4ea58c8560b121f14"
 dependencies = [
  "arrayvec",
  "bit-set 0.10.0",
@@ -1668,36 +1639,36 @@ dependencies = [
 
 [[package]]
 name = "wgpu-core-deps-apple"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3283b6da70d7fc892de95840215aa36381b5d0cbc479e88ee2b173c7b992b225"
+checksum = "061f3d319a40d39d00b1ecc2c33b89fe21d4e6fe01859df3500a3a8ecccd6b68"
 dependencies = [
  "wgpu-hal",
 ]
 
 [[package]]
 name = "wgpu-core-deps-emscripten"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "85dcf2b2e5c2afb9eda64c570a87a4e570304bf39e52eddbaaf222d655b656ad"
+checksum = "d98b86cf4abf524a902dd35f18ca6a3f08fc2ae9847c8f10b48e30491b1f0b86"
 dependencies = [
  "wgpu-hal",
 ]
 
 [[package]]
 name = "wgpu-core-deps-windows-linux-android"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f76bc9c1c186ff3d9054e0d224c93c8c1c79d6653907c5249a5c1ea1a2cb1e43"
+checksum = "7586165fd5f6d881cb9ce4bb71f40d6caab2c0f1837e3fc1d9788a197fb6004f"
 dependencies = [
  "wgpu-hal",
 ]
 
 [[package]]
 name = "wgpu-hal"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cf765132d8d5f50e192e7880464890c13f4e7457aafe8e5466e8174586e9f101"
+checksum = "b6b7fb58561a792bc237628ba0792e332de418fefe145f13b5ed8201e6d52f58"
 dependencies = [
  "android_system_properties",
  "arrayvec",
@@ -1751,9 +1722,9 @@ dependencies = [
 
 [[package]]
 name = "wgpu-naga-bridge"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c9eaac644e5008925c78567d272b9d66ef83da55a53cc17fc7daade7bb6e66e5"
+checksum = "d2f62e73117bb7a62bfd9c5a5841438a823f6566c6442a808ee269d2d055c081"
 dependencies = [
  "naga",
  "wgpu-types",
@@ -1761,9 +1732,9 @@ dependencies = [
 
 [[package]]
 name = "wgpu-types"
-version = "30.0.0"
+version = "30.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1a9c93c2b35edde326df60ffdee4c0f5864eac3011d6768b70d43f028ad93565"
+checksum = "99dad6f1fbdbbdb4c278a6508b059d44688f5cebddf78d005a46a31340269286"
 dependencies = [
  "bitflags",
  "bytemuck",

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/Cargo.toml

@@ -6,4 +6,4 @@ resolver = "2"
 wgpu = "30.0"
 pollster = "0.4"
 bytemuck = { version = "1.14", features = ["derive"] }
-pyo3 = { version = "0.23", features = ["extension-module"] }
+pyo3 = { version = "0.29", features = ["extension-module"] }

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fuzzgpu
-Version: 0.2.0
+Version: 0.4.0
 Classifier: Development Status :: 4 - Beta
 Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: MIT License
@@ -37,7 +37,8 @@ Project-URL: Repository, https://github.com/kuntal-devrat/fuzzgpu
 
 *Cross-platform GPU compute via WebGPU (`wgpu`) & Multi-Core CPU parallelism with Rayon. Zero CUDA dependencies.*
 
-[![PyPI Version](https://img.shields.io/badge/pypi-v0.2.0-blue.svg?style=flat-square)](https://pypi.org/project/fuzzgpu/)
+[![PyPI 
+Version](https://img.shields.io/badge/pypi-v0.4.0-blue.svg?style=flat-square)](https://pypi.org/project/fuzzgpu/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg?style=flat-square)](https://opensource.org/licenses/MIT)
 [![Rust](https://img.shields.io/badge/rust-1.87+-orange.svg?style=flat-square)](https://www.rust-lang.org)
 [![Cross Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux%20%7C%20WASM-lightgrey.svg?style=flat-square)](https://github.com/kuntal-devrat/fuzzgpu)
@@ -60,7 +61,24 @@ No NVIDIA CUDA drivers or complex toolkits required.
 
 ---
 
-## What's New in v0.2.0
+## What's New in v0.4.0
+
+### Production hardening
+- **WRatio score_cutoff parity** — WRatio now threads a rescaled running cutoff through its token/partial-token steps exactly as rapidfuzz does; high cutoffs no longer leak scaled token ratios (verified against rapidfuzz 3.14.5).
+- **`process.cdist` correctness** — the generic path now zeroes below-cutoff scores for scorers without `score_cutoff` support, and the fast path honors `score_multiplier` (cutoff is applied to the raw score, then scaled) — matching rapidfuzz.
+- **Randomized rapidfuzz differential suite** — new `tests/test_rapidfuzz_differential.py` pins every scorer, cutoff, alignment, distance module, editops, and `process` helper to rapidfuzz 3.14.5 across a seeded 200-pair corpus (runs on every push).
+- **Gotoh sentinel** — `NEG_INF` hardened to `i64::MIN/4` so pathological gap penalties can no longer beat it.
+- **CPU-only wheel** — `maturin build --no-default-features` now ships a GPU-free variant, built and smoke-tested in CI.
+- **Flaky stress asserts gated** — wall-clock budgets in `test_stress.py` are disabled via `FUZZGPU_SKIP_PERF_ASSERTS=1` (correctness checks always run).
+- **Accurate GPU timeout message** — readback timeouts now report the actual configured timeout.
+
+### Includes all v0.3.0 fixes
+All fixes from v0.3.0 are included.
+
+---
+
+<details>
+<summary><b>Previous (v0.3.0)</b></summary>
 
 ### Production hardening
 - **Kernel `get()` panics eliminated** — all four GPU kernels (`GpuLevenshteinKernel`,
@@ -164,15 +182,14 @@ The full Python layer is now byte-identical to rapidfuzz 3.14.5 over a 169,744-p
 ## Benchmark Results
 
 *Hardware: Intel(R) Iris(R) Xe Graphics — integrated GPU (Vulkan) + Intel Core i7 (Rayon, all cores)*
-*Versions: fuzzgpu 0.2.0 · rapidfuzz 3.14.5 · python-Levenshtein 0.27.4*
+*Versions: fuzzgpu 0.4.0 · rapidfuzz 3.14.5 · python-Levenshtein 0.27.4*
 *Median of 7 runs after warmup. Reproduce: `python benchmarks/bench_compare.py`*
 
 > **GPU class note:** These numbers are from an **integrated GPU** (iGPU), which shares memory
 > bandwidth with the CPU and has a ~1 ms dispatch round-trip. On a **discrete GPU** (dGPU) the
 > GPU columns improve significantly — expect 3–10× better GPU throughput and GPU routing kicking
 > in at much smaller batch sizes (threshold drops from ~500 pairs to ~64 pairs automatically).
-> Concurrency gains (multiple Python threads, after the wgpu fix ships) add a further 2–4×
-> on top for server workloads regardless of GPU class.
+> Concurrent Python-thread dispatch is supported on all backends.
 
 ### Levenshtein Batch (1 query × N candidates, 10-char strings)
 | Batch Size | `fuzzgpu` (GPU) | `fuzzgpu` (CPU) | `rapidfuzz` | vs RF (GPU) | vs RF (CPU) |
@@ -190,7 +207,9 @@ The full Python layer is now byte-identical to rapidfuzz 3.14.5 over a 169,744-p
 | **10,000** | 1.39 ms | 1.49 ms | 9.93 ms | 7.14× | 6.67× |
 | **50,000** | 10.19 ms | 9.91 ms | 66.57 ms | 6.54× | 6.72× |
 
-> **Note:** rapidfuzz's `DamerauLevenshtein` uses Optimal String Alignment (OSA). fuzzgpu implements the **unrestricted** Lowrance-Wagner (1975) algorithm which allows non-adjacent transpositions. For example: `damerau("ca", "abc") == 2` (fuzzgpu) vs `3` (rapidfuzz OSA). Use `fuzzgpu.distance.OSA` for OSA-compatible semantics.
+> **Note:** Both fuzzgpu and rapidfuzz's `DamerauLevenshtein` implement the **unrestricted**
+> Lowrance-Wagner (1975) algorithm (non-adjacent transpositions allowed): `damerau("ca", "abc") == 2`
+> in both. For the restricted variant use `fuzzgpu.distance.OSA` (`== 3`, also matching rapidfuzz).
 
 ### Jaro-Winkler Batch (p = 0.1)
 | Batch Size | `fuzzgpu` (GPU) | `fuzzgpu` (CPU) | `rapidfuzz` | vs RF (GPU) | vs RF (CPU) |
@@ -229,7 +248,7 @@ pip install fuzzgpu
 ```toml
 # Rust
 [dependencies]
-fuzzgpu-core = "0.2.0"
+fuzzgpu-core = "0.4.0"
 ```
 
 ---
@@ -277,8 +296,8 @@ from fuzzgpu.fuzz import (
 
 ratio("fuzzy was a bear", "fuzzy was a bear")           # 100.0
 partial_ratio("hello", "oh hello there")               # 100.0
-score, src, dst, length = partial_ratio_alignment("hello", "oh hello there")
-# (100.0, 0, 3, 5)  ← window starts at char 3 of the longer string
+score, src, dst = partial_ratio_alignment("hello", "oh hello there")
+# ScoreAlignment(score=100.0, src_start=0, src_end=5, dest_start=3, dest_end=8)
 
 token_sort_ratio("new york mets", "mets new york")      # 100.0
 token_set_ratio("fuzzy was a bear", "fuzzy bear")       # 100.0
@@ -325,8 +344,8 @@ fuzzgpu.set_cpu_only(True)       # force CPU-only mode
 
 ```toml
 [dependencies]
-fuzzgpu-core = "0.2.0"                                        # GPU + CPU fallback
-# fuzzgpu-core = { version = "0.1.7", default-features = false } # CPU-only
+fuzzgpu-core = "0.4.0"                                        # GPU + CPU fallback
+# fuzzgpu-core = { version = "0.4.0", default-features = false } # CPU-only
 ```
 
 ```rust
@@ -443,7 +462,7 @@ fuzzgpu/
 │   └── process.py           # rapidfuzz.process-compatible helpers
 ├── fuzz/                    # libFuzzer targets + stable self-harness
 ├── benchmarks/              # Comparative benchmark scripts
-├── tests/                   # Python pytest suite (174 tests)
+├── tests/                   # Python pytest suite (278 tests incl. rapidfuzz differential)
 └── docs/GPU_TESTING.md      # Fault injection & GPU test conventions
 ```
 
@@ -471,7 +490,7 @@ cargo test --workspace
 | `FUZZGPU_DEBUG` | Log GPU→CPU fallback decisions |
 | `FUZZGPU_SIMD` | Force ISA: `portable\|neon\|avx2\|avx512` |
 | `FUZZGPU_READBACK_TIMEOUT_MS` | GPU readback timeout (default 10000 ms) |
-| `FUZZGPU_SKIP_DISPATCH_LOCK` | Bypass serialization lock (repro only) |
+| `FUZZGPU_SKIP_DISPATCH_LOCK` | Opt-in GPU dispatch serialization (safety valve for the rare gfx-rs/wgpu#10085 crash class on Intel D3D12; dispatch is fully concurrent by default) |
 | `FUZZGPU_REQUIRE_GPU` | In tests: fail instead of skip when no GPU |
 | `WGPU_BACKEND` | Force wgpu backend: `vulkan\|metal\|dx12` |
 | `PROPTEST_CASES` | Override proptest case count |

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/README.md

@@ -8,7 +8,8 @@
 
 *Cross-platform GPU compute via WebGPU (`wgpu`) & Multi-Core CPU parallelism with Rayon. Zero CUDA dependencies.*
 
-[![PyPI Version](https://img.shields.io/badge/pypi-v0.2.0-blue.svg?style=flat-square)](https://pypi.org/project/fuzzgpu/)
+[![PyPI 
+Version](https://img.shields.io/badge/pypi-v0.4.0-blue.svg?style=flat-square)](https://pypi.org/project/fuzzgpu/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg?style=flat-square)](https://opensource.org/licenses/MIT)
 [![Rust](https://img.shields.io/badge/rust-1.87+-orange.svg?style=flat-square)](https://www.rust-lang.org)
 [![Cross Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux%20%7C%20WASM-lightgrey.svg?style=flat-square)](https://github.com/kuntal-devrat/fuzzgpu)
@@ -31,7 +32,24 @@ No NVIDIA CUDA drivers or complex toolkits required.
 
 ---
 
-## What's New in v0.2.0
+## What's New in v0.4.0
+
+### Production hardening
+- **WRatio score_cutoff parity** — WRatio now threads a rescaled running cutoff through its token/partial-token steps exactly as rapidfuzz does; high cutoffs no longer leak scaled token ratios (verified against rapidfuzz 3.14.5).
+- **`process.cdist` correctness** — the generic path now zeroes below-cutoff scores for scorers without `score_cutoff` support, and the fast path honors `score_multiplier` (cutoff is applied to the raw score, then scaled) — matching rapidfuzz.
+- **Randomized rapidfuzz differential suite** — new `tests/test_rapidfuzz_differential.py` pins every scorer, cutoff, alignment, distance module, editops, and `process` helper to rapidfuzz 3.14.5 across a seeded 200-pair corpus (runs on every push).
+- **Gotoh sentinel** — `NEG_INF` hardened to `i64::MIN/4` so pathological gap penalties can no longer beat it.
+- **CPU-only wheel** — `maturin build --no-default-features` now ships a GPU-free variant, built and smoke-tested in CI.
+- **Flaky stress asserts gated** — wall-clock budgets in `test_stress.py` are disabled via `FUZZGPU_SKIP_PERF_ASSERTS=1` (correctness checks always run).
+- **Accurate GPU timeout message** — readback timeouts now report the actual configured timeout.
+
+### Includes all v0.3.0 fixes
+All fixes from v0.3.0 are included.
+
+---
+
+<details>
+<summary><b>Previous (v0.3.0)</b></summary>
 
 ### Production hardening
 - **Kernel `get()` panics eliminated** — all four GPU kernels (`GpuLevenshteinKernel`,
@@ -135,15 +153,14 @@ The full Python layer is now byte-identical to rapidfuzz 3.14.5 over a 169,744-p
 ## Benchmark Results
 
 *Hardware: Intel(R) Iris(R) Xe Graphics — integrated GPU (Vulkan) + Intel Core i7 (Rayon, all cores)*
-*Versions: fuzzgpu 0.2.0 · rapidfuzz 3.14.5 · python-Levenshtein 0.27.4*
+*Versions: fuzzgpu 0.4.0 · rapidfuzz 3.14.5 · python-Levenshtein 0.27.4*
 *Median of 7 runs after warmup. Reproduce: `python benchmarks/bench_compare.py`*
 
 > **GPU class note:** These numbers are from an **integrated GPU** (iGPU), which shares memory
 > bandwidth with the CPU and has a ~1 ms dispatch round-trip. On a **discrete GPU** (dGPU) the
 > GPU columns improve significantly — expect 3–10× better GPU throughput and GPU routing kicking
 > in at much smaller batch sizes (threshold drops from ~500 pairs to ~64 pairs automatically).
-> Concurrency gains (multiple Python threads, after the wgpu fix ships) add a further 2–4×
-> on top for server workloads regardless of GPU class.
+> Concurrent Python-thread dispatch is supported on all backends.
 
 ### Levenshtein Batch (1 query × N candidates, 10-char strings)
 | Batch Size | `fuzzgpu` (GPU) | `fuzzgpu` (CPU) | `rapidfuzz` | vs RF (GPU) | vs RF (CPU) |
@@ -161,7 +178,9 @@ The full Python layer is now byte-identical to rapidfuzz 3.14.5 over a 169,744-p
 | **10,000** | 1.39 ms | 1.49 ms | 9.93 ms | 7.14× | 6.67× |
 | **50,000** | 10.19 ms | 9.91 ms | 66.57 ms | 6.54× | 6.72× |
 
-> **Note:** rapidfuzz's `DamerauLevenshtein` uses Optimal String Alignment (OSA). fuzzgpu implements the **unrestricted** Lowrance-Wagner (1975) algorithm which allows non-adjacent transpositions. For example: `damerau("ca", "abc") == 2` (fuzzgpu) vs `3` (rapidfuzz OSA). Use `fuzzgpu.distance.OSA` for OSA-compatible semantics.
+> **Note:** Both fuzzgpu and rapidfuzz's `DamerauLevenshtein` implement the **unrestricted**
+> Lowrance-Wagner (1975) algorithm (non-adjacent transpositions allowed): `damerau("ca", "abc") == 2`
+> in both. For the restricted variant use `fuzzgpu.distance.OSA` (`== 3`, also matching rapidfuzz).
 
 ### Jaro-Winkler Batch (p = 0.1)
 | Batch Size | `fuzzgpu` (GPU) | `fuzzgpu` (CPU) | `rapidfuzz` | vs RF (GPU) | vs RF (CPU) |
@@ -200,7 +219,7 @@ pip install fuzzgpu
 ```toml
 # Rust
 [dependencies]
-fuzzgpu-core = "0.2.0"
+fuzzgpu-core = "0.4.0"
 ```
 
 ---
@@ -248,8 +267,8 @@ from fuzzgpu.fuzz import (
 
 ratio("fuzzy was a bear", "fuzzy was a bear")           # 100.0
 partial_ratio("hello", "oh hello there")               # 100.0
-score, src, dst, length = partial_ratio_alignment("hello", "oh hello there")
-# (100.0, 0, 3, 5)  ← window starts at char 3 of the longer string
+score, src, dst = partial_ratio_alignment("hello", "oh hello there")
+# ScoreAlignment(score=100.0, src_start=0, src_end=5, dest_start=3, dest_end=8)
 
 token_sort_ratio("new york mets", "mets new york")      # 100.0
 token_set_ratio("fuzzy was a bear", "fuzzy bear")       # 100.0
@@ -296,8 +315,8 @@ fuzzgpu.set_cpu_only(True)       # force CPU-only mode
 
 ```toml
 [dependencies]
-fuzzgpu-core = "0.2.0"                                        # GPU + CPU fallback
-# fuzzgpu-core = { version = "0.1.7", default-features = false } # CPU-only
+fuzzgpu-core = "0.4.0"                                        # GPU + CPU fallback
+# fuzzgpu-core = { version = "0.4.0", default-features = false } # CPU-only
 ```
 
 ```rust
@@ -414,7 +433,7 @@ fuzzgpu/
 │   └── process.py           # rapidfuzz.process-compatible helpers
 ├── fuzz/                    # libFuzzer targets + stable self-harness
 ├── benchmarks/              # Comparative benchmark scripts
-├── tests/                   # Python pytest suite (174 tests)
+├── tests/                   # Python pytest suite (278 tests incl. rapidfuzz differential)
 └── docs/GPU_TESTING.md      # Fault injection & GPU test conventions
 ```
 
@@ -442,7 +461,7 @@ cargo test --workspace
 | `FUZZGPU_DEBUG` | Log GPU→CPU fallback decisions |
 | `FUZZGPU_SIMD` | Force ISA: `portable\|neon\|avx2\|avx512` |
 | `FUZZGPU_READBACK_TIMEOUT_MS` | GPU readback timeout (default 10000 ms) |
-| `FUZZGPU_SKIP_DISPATCH_LOCK` | Bypass serialization lock (repro only) |
+| `FUZZGPU_SKIP_DISPATCH_LOCK` | Opt-in GPU dispatch serialization (safety valve for the rare gfx-rs/wgpu#10085 crash class on Intel D3D12; dispatch is fully concurrent by default) |
 | `FUZZGPU_REQUIRE_GPU` | In tests: fail instead of skip when no GPU |
 | `WGPU_BACKEND` | Force wgpu backend: `vulkan\|metal\|dx12` |
 | `PROPTEST_CASES` | Override proptest case count |

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/crates/fuzzgpu-core/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "fuzzgpu-core"
-version = "0.2.0"
+version = "0.4.0"
 edition = "2021"
 description = "GPU-accelerated fuzzy string matching engine"
 license = "MIT"

{fuzzgpu-0.2.0 → fuzzgpu-0.4.0}/crates/fuzzgpu-core/src/fuzz.rs

@@ -618,20 +618,27 @@ pub fn wratio(s1: &str, s2: &str, score_cutoff: f64) -> f64 {
         len2 as f64 / len1 as f64
     };
 
-    let mut end_ratio = ratio_with_cutoff(s1, s2, score_cutoff);
+    // rapidfuzz threads the *mutated* cutoff forward: each step rescales the
+    // running cutoff (already divided by the previous step's scale) by its own
+    // scale factor, so a low end_ratio at a high user cutoff escalates the
+    // effective cutoff past 100 and suppresses the token/partial-token terms.
+    // Reusing the original score_cutoff at every step instead let those terms
+    // leak through (WRatio score_cutoff parity).
+    let mut cutoff = score_cutoff;
+    let mut end_ratio = ratio_with_cutoff(s1, s2, cutoff);
 
     if len_ratio < 1.5 {
-        let c = score_cutoff.max(end_ratio) / UNBASE_SCALE;
-        return end_ratio.max(token_ratio(s1, s2, c) * UNBASE_SCALE);
+        cutoff = cutoff.max(end_ratio) / UNBASE_SCALE;
+        return end_ratio.max(token_ratio(s1, s2, cutoff) * UNBASE_SCALE);
     }
 
     let partial_scale = if len_ratio <= 8.0 { 0.9 } else { 0.6 };
 
-    let c = score_cutoff.max(end_ratio) / partial_scale;
-    end_ratio = end_ratio.max(partial_ratio(s1, s2, c) * partial_scale);
+    cutoff = cutoff.max(end_ratio) / partial_scale;
+    end_ratio = end_ratio.max(partial_ratio(s1, s2, cutoff) * partial_scale);
 
-    let c = score_cutoff.max(end_ratio) / UNBASE_SCALE;
-    end_ratio.max(partial_token_ratio(s1, s2, c) * UNBASE_SCALE * partial_scale)
+    cutoff = cutoff.max(end_ratio) / UNBASE_SCALE;
+    end_ratio.max(partial_token_ratio(s1, s2, cutoff) * UNBASE_SCALE * partial_scale)
 }
 
 /// QRatio: rapidfuzz's quick ratio — 0 for empty inputs (FuzzyWuzzy
@@ -647,12 +654,8 @@ pub fn qratio(s1: &str, s2: &str, score_cutoff: f64) -> f64 {
 /// Uses the shared-query Myers SIMD fast path when the query is ASCII ≤ 64 bytes
 /// and all candidates are ASCII — same acceleration as `levenshtein_batch_auto`.
 pub fn ratio_batch(query: &str, candidates: &[&str]) -> Vec<f64> {
-    // The indel model (substitution cost=2) does not have a direct Myers
-    // bit-vector path — Myers gives edit distance with unit substitution cost.
-    // We approximate with the standard Levenshtein batch and then convert:
-    //   indel_dist = 2 * lev_dist when lev path only uses ins/del (no subs).
-    // For general strings we fall back to the per-pair indel_distance.
-    // The Rayon parallel path is correct for all inputs.
+    // Per-pair ratio over Rayon; each pair takes the ASCII Myers / char-DP
+    // fast paths inside indel_distance.
     candidates.par_iter().map(|c| ratio(query, c)).collect()
 }
 
@@ -1154,4 +1157,17 @@ mod tests {
         let r = wratio("new york mets", "new york", 0.0);
         assert!((0.0..=100.0).contains(&r));
     }
+
+    #[test]
+    fn test_wratio_cutoff_threads_rescaled_cutoff() {
+        // rapidfuzz WRatio rescales the running cutoff at each step, so a high
+        // user cutoff suppresses the token terms even when a perfect substring
+        // exists. Regression: fuzzgpu returned 57.0/85.5 where rapidfuzz
+        // returns 0.0 for these inputs.
+        assert_eq!(wratio("new york yankees", "a", 90.0), 0.0);
+        assert_eq!(wratio("a", "ab", 95.0), 0.0);
+        assert_eq!(wratio("hello world", "world", 95.0), 0.0);
+        // The cutoff=0 path is unaffected: the unscaled partial ratio still wins.
+        assert!((wratio("new york yankees", "a", 0.0) - 60.0).abs() < 1e-9);
+    }
 }