PyPcre 0.6.4__tar.gz → 0.6.6__tar.gz

{pypcre-0.6.4 → pypcre-0.6.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PyPcre
-Version: 0.6.4
+Version: 0.6.6
 Summary: Modern, GIL-friendly, Fast Python bindings for PCRE2 with auto caching and JIT of compiled patterns.
 Author-email: ModelCloud <qubitium@modelcloud.ai>
 License-Expression: Apache-2.0
@@ -47,7 +47,8 @@ Fast, free-threaded Python bindings for `PCRE2` with a stable `stdlib.re`-compat
 
 
 ## Latest News 🚀
-* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
+* 10/21/2026 **[0.6.5](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.5)**: Inlined the generation-validated module hot lookup for faster module-level `match`, `search`, `fullmatch`, `findall`, and `finditer`, while preserving legacy-backend, cache, thread-default, and free-threaded fallbacks. 🚀⚡🧵
+* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, added generation-validated module caches and exact literal-capture substitution paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
 * 08/25/2026 **[0.6.2](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.2)**: Hardened packaging metadata and fixed FreeBSD/WSL CI failures on PCRE2 10.42 runtimes, including legacy replacement-template and JIT-fallback compatibility. 🛠️🐧🪟
 * 08/25/2026 **[0.6.1](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.1)**: Workaround for PCRE2 10.46/10.47 JIT start-optimization regressions, new verifying clobber suite, second memory/thread-safety sweep, Python 3.15 `sre_parse` removal compatibility, aligned `Pattern` `pos`/`endpos` signatures, and `setuptools` metadata fixes. 🛡️🧵🐍
 * 08/09/2026 **[0.6.0](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.0)**: `findall`, `finditer`, `sub`/`subn`, `split`, and `match`/`search`/`fullmatch` hot-path speedups up to **46x** vs `stdlib.re` and **48x** vs `regex`, with free-threaded `findall` reaching **13.8x** on 8 threads. 🚀⚡
@@ -255,7 +256,8 @@ numbers = pattern.findall(b"line 1\nline 22")
 
 - Module helpers: `prefixmatch`, `match`, `search`, `fullmatch`, `finditer`,
   `findall`, `split`, `sub`, `subn`, `compile`, `escape`, `purge`,
-  `template`, `parallel_map`, `configure`, `configure_threads`,
+  `template`, `parallel_map`, `configure`, `configure_match_limits`,
+  `configure_threads`,
   `configure_thread_pool`, `shutdown_thread_pool`, `set_cache_limit`,
   `get_cache_limit`, and `clear_cache`.
 - `compile()` returns a `Pattern` object with the familiar matching helpers
@@ -270,6 +272,12 @@ numbers = pattern.findall(b"line 1\nline 22")
   `NOFLAG`.
 - Errors are raised as `pcre.PcreError`; `error` and `PatternError` are kept as
   compatibility aliases.
+- Buffer-protocol subjects (`mmap.mmap`, `bytearray`, `array.array`,
+  `memoryview`) are accepted anywhere `bytes` is. Each call takes an immutable
+  snapshot of the buffer before matching, so concurrent writers (or another
+  process writing a shared mapping) cannot corrupt a match in progress. That
+  snapshot is a full copy: searching a 200 MB `mmap` allocates 200 MB per call.
+  For very large inputs, match over slices or chunks instead of the whole map.
 
 ### Common examples 🧪
 
@@ -367,10 +375,25 @@ the conversion without repeating the flag.
   workers' high-level helper caches on their next cache-backed call. Backend
   scratch buffers remain thread-scoped and are released when that thread exits.
   Oversized patterns and templates are not retained.
+- 🛑 `pcre.configure_match_limits(match_limit=..., depth_limit=..., heap_limit=...)`
+  bounds the work any single match may do, process-wide (PCRE2's
+  `pcre2_set_match_limit` / `depth_limit` / `heap_limit`; the heap limit is in
+  KiB). A match that exceeds a limit raises `PcreError` (`MATCHLIMIT`,
+  `DEPTHLIMIT` or `HEAPLIMIT`) instead of running unbounded, which is useful
+  for untrusted patterns or subjects. `0` restores the PCRE2 build default.
+  PCRE2 counts the match limit per start position, so a search over `n`
+  characters can still take up to `n × match_limit` steps.
 - 🧵 `pcre.configure_threads()`, `pcre.configure_thread_pool()`,
   `shutdown_thread_pool()`, `Flag.THREADS`, and `Flag.NO_THREADS` are available
   if you want to opt into or restrict threaded execution.
 
+## Documentation 📚
+
+- [`lastindex` replay and JIT stack fallback](docs/lastindex-and-jit-stack.md) —
+  how `Match.lastindex`/`lastgroup` are reconstructed, why the replay must use
+  the original start offset and options, and how an exhausted JIT stack falls
+  back to the interpreter instead of raising.
+
 ## Building 🏗️
 
 The extension links against an existing `libpcre2-8` installation. Install the development headers for your platform before building,

{pypcre-0.6.4 → pypcre-0.6.6}/PyPcre.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PyPcre
-Version: 0.6.4
+Version: 0.6.6
 Summary: Modern, GIL-friendly, Fast Python bindings for PCRE2 with auto caching and JIT of compiled patterns.
 Author-email: ModelCloud <qubitium@modelcloud.ai>
 License-Expression: Apache-2.0
@@ -47,7 +47,8 @@ Fast, free-threaded Python bindings for `PCRE2` with a stable `stdlib.re`-compat
 
 
 ## Latest News 🚀
-* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
+* 10/21/2026 **[0.6.5](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.5)**: Inlined the generation-validated module hot lookup for faster module-level `match`, `search`, `fullmatch`, `findall`, and `finditer`, while preserving legacy-backend, cache, thread-default, and free-threaded fallbacks. 🚀⚡🧵
+* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, added generation-validated module caches and exact literal-capture substitution paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
 * 08/25/2026 **[0.6.2](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.2)**: Hardened packaging metadata and fixed FreeBSD/WSL CI failures on PCRE2 10.42 runtimes, including legacy replacement-template and JIT-fallback compatibility. 🛠️🐧🪟
 * 08/25/2026 **[0.6.1](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.1)**: Workaround for PCRE2 10.46/10.47 JIT start-optimization regressions, new verifying clobber suite, second memory/thread-safety sweep, Python 3.15 `sre_parse` removal compatibility, aligned `Pattern` `pos`/`endpos` signatures, and `setuptools` metadata fixes. 🛡️🧵🐍
 * 08/09/2026 **[0.6.0](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.0)**: `findall`, `finditer`, `sub`/`subn`, `split`, and `match`/`search`/`fullmatch` hot-path speedups up to **46x** vs `stdlib.re` and **48x** vs `regex`, with free-threaded `findall` reaching **13.8x** on 8 threads. 🚀⚡
@@ -255,7 +256,8 @@ numbers = pattern.findall(b"line 1\nline 22")
 
 - Module helpers: `prefixmatch`, `match`, `search`, `fullmatch`, `finditer`,
   `findall`, `split`, `sub`, `subn`, `compile`, `escape`, `purge`,
-  `template`, `parallel_map`, `configure`, `configure_threads`,
+  `template`, `parallel_map`, `configure`, `configure_match_limits`,
+  `configure_threads`,
   `configure_thread_pool`, `shutdown_thread_pool`, `set_cache_limit`,
   `get_cache_limit`, and `clear_cache`.
 - `compile()` returns a `Pattern` object with the familiar matching helpers
@@ -270,6 +272,12 @@ numbers = pattern.findall(b"line 1\nline 22")
   `NOFLAG`.
 - Errors are raised as `pcre.PcreError`; `error` and `PatternError` are kept as
   compatibility aliases.
+- Buffer-protocol subjects (`mmap.mmap`, `bytearray`, `array.array`,
+  `memoryview`) are accepted anywhere `bytes` is. Each call takes an immutable
+  snapshot of the buffer before matching, so concurrent writers (or another
+  process writing a shared mapping) cannot corrupt a match in progress. That
+  snapshot is a full copy: searching a 200 MB `mmap` allocates 200 MB per call.
+  For very large inputs, match over slices or chunks instead of the whole map.
 
 ### Common examples 🧪
 
@@ -367,10 +375,25 @@ the conversion without repeating the flag.
   workers' high-level helper caches on their next cache-backed call. Backend
   scratch buffers remain thread-scoped and are released when that thread exits.
   Oversized patterns and templates are not retained.
+- 🛑 `pcre.configure_match_limits(match_limit=..., depth_limit=..., heap_limit=...)`
+  bounds the work any single match may do, process-wide (PCRE2's
+  `pcre2_set_match_limit` / `depth_limit` / `heap_limit`; the heap limit is in
+  KiB). A match that exceeds a limit raises `PcreError` (`MATCHLIMIT`,
+  `DEPTHLIMIT` or `HEAPLIMIT`) instead of running unbounded, which is useful
+  for untrusted patterns or subjects. `0` restores the PCRE2 build default.
+  PCRE2 counts the match limit per start position, so a search over `n`
+  characters can still take up to `n × match_limit` steps.
 - 🧵 `pcre.configure_threads()`, `pcre.configure_thread_pool()`,
   `shutdown_thread_pool()`, `Flag.THREADS`, and `Flag.NO_THREADS` are available
   if you want to opt into or restrict threaded execution.
 
+## Documentation 📚
+
+- [`lastindex` replay and JIT stack fallback](docs/lastindex-and-jit-stack.md) —
+  how `Match.lastindex`/`lastgroup` are reconstructed, why the replay must use
+  the original start offset and options, and how an exhausted JIT stack falls
+  back to the interpreter instead of raising.
+
 ## Building 🏗️
 
 The extension links against an existing `libpcre2-8` installation. Install the development headers for your platform before building,

{pypcre-0.6.4 → pypcre-0.6.6}/PyPcre.egg-info/SOURCES.txt

@@ -33,12 +33,14 @@ tests/test_api_parity.py
 tests/test_basic.py
 tests/test_bench_string.py
 tests/test_benchmark.py
+tests/test_build_hardening.py
 tests/test_bytes.py
 tests/test_c_api_audit.py
 tests/test_cache.py
 tests/test_cache_global.py
 tests/test_cache_global_inproc.py
 tests/test_cache_scope_safety.py
+tests/test_capture_limits.py
 tests/test_clobber.py
 tests/test_clobber_thread.py
 tests/test_clobber_verify.py
@@ -54,9 +56,12 @@ tests/test_jit.py
 tests/test_jit_anchoring.py
 tests/test_jit_start_optimize_workaround.py
 tests/test_lastindex_ovector_fastpath.py
+tests/test_lastindex_replay_and_jit_stack_fallback.py
 tests/test_lastindex_semantics.py
 tests/test_literal_capture_findall_fastpath.py
 tests/test_literal_capture_split_fastpath.py
+tests/test_literal_capture_sub_fastpath.py
+tests/test_match_limits.py
 tests/test_memory.py
 tests/test_mmap.py
 tests/test_module.py

{pypcre-0.6.4 → pypcre-0.6.6}/README.md

@@ -24,7 +24,8 @@ Fast, free-threaded Python bindings for `PCRE2` with a stable `stdlib.re`-compat
 
 
 ## Latest News 🚀
-* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
+* 10/21/2026 **[0.6.5](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.5)**: Inlined the generation-validated module hot lookup for faster module-level `match`, `search`, `fullmatch`, `findall`, and `finditer`, while preserving legacy-backend, cache, thread-default, and free-threaded fallbacks. 🚀⚡🧵
+* 10/20/2026 **[0.6.4](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.4)**: Fixed free-threaded races in `Match.lastindex` replay and PCRE2 version initialization, reduced Python dispatch overhead across module-level and compiled-pattern `match`, `search`, `fullmatch`, `findall`, `finditer`, `split`, `sub`, and `subn` hot paths, added generation-validated module caches and exact literal-capture substitution paths, and prevented native Windows MSVC builds from inheriting MinGW `pkg-config` flags. CI now uses interpreter-scoped pip/pytest bootstraps and the matching FreeBSD Python pip package. 🛡️🧵🚀⚡🪟🐧
 * 08/25/2026 **[0.6.2](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.2)**: Hardened packaging metadata and fixed FreeBSD/WSL CI failures on PCRE2 10.42 runtimes, including legacy replacement-template and JIT-fallback compatibility. 🛠️🐧🪟
 * 08/25/2026 **[0.6.1](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.1)**: Workaround for PCRE2 10.46/10.47 JIT start-optimization regressions, new verifying clobber suite, second memory/thread-safety sweep, Python 3.15 `sre_parse` removal compatibility, aligned `Pattern` `pos`/`endpos` signatures, and `setuptools` metadata fixes. 🛡️🧵🐍
 * 08/09/2026 **[0.6.0](https://github.com/ModelCloud/PyPcre/releases/tag/v0.6.0)**: `findall`, `finditer`, `sub`/`subn`, `split`, and `match`/`search`/`fullmatch` hot-path speedups up to **46x** vs `stdlib.re` and **48x** vs `regex`, with free-threaded `findall` reaching **13.8x** on 8 threads. 🚀⚡
@@ -232,7 +233,8 @@ numbers = pattern.findall(b"line 1\nline 22")
 
 - Module helpers: `prefixmatch`, `match`, `search`, `fullmatch`, `finditer`,
   `findall`, `split`, `sub`, `subn`, `compile`, `escape`, `purge`,
-  `template`, `parallel_map`, `configure`, `configure_threads`,
+  `template`, `parallel_map`, `configure`, `configure_match_limits`,
+  `configure_threads`,
   `configure_thread_pool`, `shutdown_thread_pool`, `set_cache_limit`,
   `get_cache_limit`, and `clear_cache`.
 - `compile()` returns a `Pattern` object with the familiar matching helpers
@@ -247,6 +249,12 @@ numbers = pattern.findall(b"line 1\nline 22")
   `NOFLAG`.
 - Errors are raised as `pcre.PcreError`; `error` and `PatternError` are kept as
   compatibility aliases.
+- Buffer-protocol subjects (`mmap.mmap`, `bytearray`, `array.array`,
+  `memoryview`) are accepted anywhere `bytes` is. Each call takes an immutable
+  snapshot of the buffer before matching, so concurrent writers (or another
+  process writing a shared mapping) cannot corrupt a match in progress. That
+  snapshot is a full copy: searching a 200 MB `mmap` allocates 200 MB per call.
+  For very large inputs, match over slices or chunks instead of the whole map.
 
 ### Common examples 🧪
 
@@ -344,10 +352,25 @@ the conversion without repeating the flag.
   workers' high-level helper caches on their next cache-backed call. Backend
   scratch buffers remain thread-scoped and are released when that thread exits.
   Oversized patterns and templates are not retained.
+- 🛑 `pcre.configure_match_limits(match_limit=..., depth_limit=..., heap_limit=...)`
+  bounds the work any single match may do, process-wide (PCRE2's
+  `pcre2_set_match_limit` / `depth_limit` / `heap_limit`; the heap limit is in
+  KiB). A match that exceeds a limit raises `PcreError` (`MATCHLIMIT`,
+  `DEPTHLIMIT` or `HEAPLIMIT`) instead of running unbounded, which is useful
+  for untrusted patterns or subjects. `0` restores the PCRE2 build default.
+  PCRE2 counts the match limit per start position, so a search over `n`
+  characters can still take up to `n × match_limit` steps.
 - 🧵 `pcre.configure_threads()`, `pcre.configure_thread_pool()`,
   `shutdown_thread_pool()`, `Flag.THREADS`, and `Flag.NO_THREADS` are available
   if you want to opt into or restrict threaded execution.
 
+## Documentation 📚
+
+- [`lastindex` replay and JIT stack fallback](docs/lastindex-and-jit-stack.md) —
+  how `Match.lastindex`/`lastgroup` are reconstructed, why the replay must use
+  the original start offset and options, and how an exhausted JIT stack falls
+  back to the interpreter instead of raising.
+
 ## Building 🏗️
 
 The extension links against an existing `libpcre2-8` installation. Install the development headers for your platform before building,

{pypcre-0.6.4 → pypcre-0.6.6}/pcre/__init__.py

@@ -26,6 +26,7 @@ from .pcre import (
     clear_cache,
     compile,
     configure,
+    configure_match_limits,
     findall,
     finditer,
     fullmatch,
@@ -109,6 +110,7 @@ __all__ = [
     "clear_cache",
     "purge",
     "configure",
+    "configure_match_limits",
     "configure_threads",
     "configure_thread_pool",
     "cache_strategy",

{pypcre-0.6.4 → pypcre-0.6.6}/pcre/cache.py

@@ -7,7 +7,6 @@
 
 from __future__ import annotations
 
-import os
 from enum import Enum
 from threading import RLock, local
 from typing import Any, Callable, TypeVar, cast
@@ -113,10 +112,10 @@ def _env_flag_is_true(value: str | None) -> bool:
     return value[0] not in {"0", "f", "F", "n", "N"}
 
 
-if _env_flag_is_true(os.getenv("PYPCRE_CACHE_PATTERN_GLOBAL")):
-    _CACHE_STRATEGY = _CacheStrategy.GLOBAL
-else:
-    _CACHE_STRATEGY = _CacheStrategy.THREAD_LOCAL
+# The C extension latches the strategy when it is imported (honouring -E/-I
+# and the PCRE2_CACHE_PATTERN_GLOBAL alias via Py_GETENV); mirror it rather
+# than re-reading the environment so the two layers can never disagree.
+_CACHE_STRATEGY = _CacheStrategy(_pcre2.get_cache_strategy())
 
 
 def cache_strategy(strategy: str | None = None) -> str: