Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Planned

Inline {...:validator(...)} syntax and async validation pipelines (currently deferred in API documentation).
composed_type extensions: pattern +, inheritance, and flattening nested parse results into the parent (see #7 <https://github.com/eddiethedean/formatparse/issues/7>_).

0.8.2 - 2026-05-22

Added

ty type checking: _formatparse.pyi stubs, stricter typing in formatparse/, and [tool.ty] configuration (tests use targeted overrides).
findall and findall_iter (including :meth:FormatParser.findall_iter) accept optional max_matches to cap how many non-overlapping matches are returned on untrusted input.

Changed

Dependencies: PyO3 0.28 (with API migration), fancy-regex 0.17, lru 0.18, rayon 1.12; Sphinx / myst-parser / RTD theme bumps for docs; GitHub Actions artifact actions v7/v8.
MSRV: Rust 1.83 (required by PyO3 0.28).
Pattern LRU cache emits a :mod:warnings message when a hash hit fails full verification (e.g. in-place mutation of converter.pattern on a reused extra_types dict).

Fixed

:class:BidirectionalPattern now passes extra_types to :func:compile so custom-type regexes are strict at compile time, not only on :meth:~BidirectionalPattern.parse.
:meth:FormatParser.parse and :meth:FormatParser.search re-resolve the compiled parser through the pattern cache when call-time extra_types differ from compile-time fingerprints, so custom converters use their @with_pattern regex instead of the \S+ fallback (and no longer raise ValueError on non-matching text that the loose regex would accept).
:class:BidirectionalPattern validation constraints are derived from compiled field metadata (field_constraints), fixing incorrect constraint extraction for nested brace specs and custom type names.

0.8.1 - 2026-05-15

Changed

Python package layout: split the monolithic formatparse module into focused submodules (api, validation, bidirectional, etc.) with unchanged from formatparse import … surface.
Rust internals: phased SOLID refactor—CompiledFields on FormatParser, split types/regex and parser/matching modules, pattern compilation in formatparse-core, unified builtin conversion via builtin_convert, and format_parser / findall_iter module splits. No intended public API or match-semantics changes.

Maintenance

Quality: ruff, mypy (check target 3.9+), cargo clippy -D warnings, and rustfmt clean across the workspace.

0.8.0 - 2026-05-15

Added

Multiline fields (:ml) for captures that may span newlines, with greedy/non-greedy boundaries like plain string fields (#8). Width, precision, alignment, and fill are supported for :ml the same way as for string fields (#70, #75).
Indent-block fields (:blk): same boundary rules as :ml, then dedent by removing the largest common prefix of spaces and tabs on each line (blank lines do not set the margin; tabs count as single characters) (#69).
Pattern line continuations: a backslash immediately before end-of-line continues the format pattern on the next line (\r\n or \n); doubled backslashes keep a literal newline; leading spaces and tabs on the continued line are stripped (#68).
findall_iter for incremental iteration over findall-style matches (#13).
parse_batch to parse many strings with one compiled parser (#14).
:brace field type for capturing a literal {...} payload in the input (#15).
Datetime strftime: merge adjacent strftime fragments that share a field name into a single datetime conversion (#4).
Composition: embed a compiled FormatParser as a custom type so a field is parsed by a child parser and returns a nested ParseResult, via composed_type / extra_types (#7).
Post-parse validators: ValidationError, MultipleValidationErrors, apply_validators, validator decorator, parse / compile / ValidatedParser with validators=, and built-ins in_range / non_empty_str (#10).
Validation pipeline (ValidationPipeline): ordered per-field validators, validation_mode strict / collect / lenient, parse(..., pipeline=...), whole-result hooks for cross-field checks, parse_with_validation, and built-ins is_valid_email / is_valid_url (heuristic checks, not full RFC compliance or security audits) (#11, #74–#78).
Regex lookaround assertions after the type token for integer (:d and related) and float (:f / :F) fields: append (?=…), (?!…), (?<=…), or (?<!…) groups so the numeric capture stays zero-width outside the field span (issues #9 <https://github.com/eddiethedean/formatparse/issues/9>, upstream parse#209 <https://github.com/r1chardj0n3s/parse/issues/209>). Compiled format patterns use the fancy-regex engine; simple literal positive lookarounds may be rewritten as non-capturing groups for correct anchoring. Lookarounds are rejected inside strftime %… type tails for this release.
Nested format patterns in a field’s format specification (e.g. {outer:{inner:d}}): the inner {…} is compiled as its own pattern, the outer capture is parsed again, and nested values appear as ParseResult objects under named (issue #12 <https://github.com/eddiethedean/formatparse/issues/12>; upstream parse#206 <https://github.com/r1chardj0n3s/parse/issues/206>). Brace-balanced scanning applies in the spec after :; nesting depth is capped at 10; findall and FindallIter use the Python match path when nested fields are present (same rule as nested dict field names).
Input line continuations for :ml and :blk captures: a backslash before end-of-line in the matched text joins lines (same odd/even backslash rules as pattern continuations in #68); leading spaces and tabs on the continued line are stripped (#80).

Fixed

Empty string input can match patterns whose fields use the default string conversion (#) (#16).
Post-parse validation: _validator_field_value no longer returned None for valid fixed indices (incorrect return indentation under the bounds check).
ValidationPipeline collect mode: all hooks still run when field validators fail; field and hook failures are merged into a single MultipleValidationErrors (field errors first in validator key order, then hook errors in registration order) (#11).
Float f / F with .0 precision (e.g. {:02.0f}): accept integer-shaped captures such as 20, matching str.format output for those specs (#84; upstream parse#159 <https://github.com/r1chardj0n3s/parse/issues/159>_).
CI: Skip pympler-based memory tests on PyPy (SummaryTracker / asizeof raises KeyError on PyPy 3.11).
CI / ReDoS guard: Raise the post-regex-compile wall-clock cap from 200ms to 500ms so slow shared runners (e.g. macOS Intel) do not spuriously reject valid patterns; documentation updated accordingly.
Pattern LRU cache: after a hash hit, the entry is checked against the full normalized pattern and extra_types fingerprint so a colliding key cannot return the wrong compiled parser.
Fixed-width string fields with literals beside the field (e.g. "{s:<5.5} " vs "abc "): when width and precision are equal, the compiled fragment is exactly prec characters so trailing (or leading) pattern space is not pulled into the capture (#97). Right-aligned fields with that same equal width and precision also accept an opaque fixed-width capture (including trailing spaces) so post-capture validation matches parse for those cells (#97).
String width/precision with literal newline after the field (e.g. " {s:<4.4}\n" vs " \n"): bounded string fragments no longer use DOTALL . for content, so a trailing \n in the input is not consumed as part of the field (#95). The same bounded runs also exclude VT, FF, NEL (U+0085), U+2028 LINE SEPARATOR, and U+2029 PARAGRAPH SEPARATOR so those line boundaries cannot be absorbed under (?s) when they appear as literals after the field.
Integer and radix fields with both width and precision (e.g. "{:2.2d}{:2.2d}", "#{:2.2x}{:2.2x}"): digit runs use inclusive min/max bounds (parse semantics: width = minimum digits, precision = maximum) so adjacent fields no longer steal digits with a greedy + (#82; upstream parse#107 <https://github.com/r1chardj0n3s/parse/issues/107>_). Patterns that previously matched too loosely on short inputs may now return no match.
String fields with alignment + precision before a fixed-width integer (e.g. "{n:>10.10}{x:02d}"): leading fill in the regex is non-greedy so the slice for the width/precision content is left for the following digit field (#88; related parse#218 <https://github.com/r1chardj0n3s/parse/issues/218>_).
Default string fields next to literals (e.g. "/{name}"): an empty capture is allowed when it matches str.format output such as "/" for name="" (#83; upstream parse#136 <https://github.com/r1chardj0n3s/parse/issues/136>_). Applies to full-string parse / compile().parse; search / findall still use .+? for those segments so unanchored matching does not stop early.
Integer d: leading spaces and tabs before decimal digits are accepted (e.g. parse("{a:d}", " 0")) for parity with padded str.format output (#81; upstream parse#133 <https://github.com/r1chardj0n3s/parse/issues/133>_).

Changed

compile() uses the same pattern LRU cache as parse / search / findall when cache keys match (#29).
PyO3: extension-module is an explicit Cargo feature enabled by Maturin for wheels and maturin develop; plain cargo test / cargo clippy on formatparse-pyo3 link libPython without extra linker configuration.

Documentation

Expanded the custom types / extra_types user guide (#17).

Maintenance

CI, Makefile, and publish workflow: pass --features extension-module when Maturin is run from formatparse-pyo3/ so builds match root pip install / maturin develop --manifest-path (Maturin does not read the repo-root pyproject from that working directory).
Pytest: skip tests/test_indent_block.py when the installed _formatparse lacks :blk / multiline validation (skip message points to maturin develop / editable install).
Rust: migrate off deprecated ToPyObject::to_object (#45); remove several crate-level Clippy allows (manual_strip, dead_code, wrong_self_convention, if_same_then_else, too_many_arguments, type_complexity) (#44, #46–#50).
CI: Codecov Action files input (replaces deprecated file); Dependabot bumps for Rust crates and GitHub Actions.
Formatting: Ruff-format touched Python tests and package sources.
CI: Standalone Rust · formatparse-core (Linux) job (separate row in Actions) plus matrix Summary table; matrix splits Rust compile / Rust test / Python build (maturin build + pip install wheel, no pytest until after cargo test) / Python test (pytest).
CI: Ubuntu PyPy 3.11 job; python-tests for formatparse-pyo3 is a blocking step on Ubuntu CPython 3.11 (no continue-on-error).
Rust: clearer expect messages for UTF-8 invariants in line-continuation helpers and ResultsIterator.

0.7.0 - 2026-05-14

PyPI publish (post-tag workflow fix)

Publish workflow: build sdist from repo root with --manifest-path (fixes “pyproject.toml not found” when run from formatparse-pyo3/ only).
Remove [tool.maturin] python-source = "." so maturin sdist does not require a repo-root _formatparse Python package (that check failed for this PyO3 cdylib layout on 1.12.x and 1.13.x). Publish and [build-system].requires use maturin>=1.12.6,<2.0.
Linux wheel for CPython 3.8: use manylinux2014 + --manylinux 2014 when manylinux_2_28 no longer ships that interpreter (avoids exit 127); pin manylinux2014_x86_64:2026.05.01-2 because latest removed /opt/python/cp38-cp38 after manylinux#1882.

Added

formatparse_core::count_capturing_groups for validating custom with_pattern regexes, including correct handling of (?P<name>...) named capture groups.
Richer ParseResult.__repr__ for REPL debugging.

Fixed

Pattern cache keys for extra_types now incorporate converter pattern and regex_group_count so cached parsers are not incorrectly reused.
Results negative indexing aligned with Python list semantics.
Zero-fill and right-align width.precision parsing parity with the reference parse library (#40).
Trailing fill preserved for left-aligned string fields with width (#39).
formatparse.__version__ stable when installed from sdist/wheel (PEP 440) and fallback from Cargo.toml in source checkouts (#38).
FormatParser pickling behavior documented and clarified for extra_types.

Changed

findall return type documentation: fast path returns Results; with extra_types, evaluate_result=False, or nested dict fields, a plain list is returned (matches runtime).
Security documentation: clarify post-compile timing check vs match-time behavior; document uncapped findall match count within input limits.

Maintenance

CI: load pytest-cov when PYTEST_DISABLE_PLUGIN_AUTOLOAD=1; bump cargo-audit in the main CI Ubuntu step for advisory DB compatibility.
Dependency updates (e.g. lru for RustSec advisories), formatting, and Clippy cleanups.