CHANGELOG.md

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.

1.1.0 - 2026-08-08

The singleton RegexSolver becomes an instantiable RegexSolverClient, with AsyncRegexSolverClient beside it, every operation moves from Term onto the client, and the SDK covers the whole API rather than the seven endpoints it knew about. The API moved with it, to a new contract served under /v1, so 1.0.x no longer reaches an endpoint that exists; see Compatibility below.

Added

• RegexSolverClient(api_token, base_url=..., auto_batch=True, max_terms_per_request=None) and AsyncRegexSolverClient, taking the same arguments. Several clients, each with its own token, can exist in one process; both work as context managers and release their session on close() / aclose(). The synchronous client drives the asynchronous one on a shared background event loop.
• The operations the API gained since 1.0.3: complement, concat, determinize, repeat, get_cardinality, get_dot, get_length, is_empty, is_empty_string, is_total, is_deterministic and get_account_limits.
• Keyword options: response_format (REGEX, FAIR or ANY), deterministic and execution_timeout on the operations that return a term; execution_timeout alone on analyze operations and determinize.
• generate_strings(term, limit, offset, ...), which pages through the language instead of returning a fixed count from the start, taking path_order, character_order, seed, min_length, max_length and charset. Paging is only consistent over a deterministic FAIR, hence determinize() and is_deterministic().
• Term.matches(string), evaluated locally with re — anchored \A(?:...)\Z, compiled with re.DOTALL — rather than by the API. It raises on a FAIR whose pattern is not known yet, and returns False for the empty language, which the engine writes as [].
• An exception hierarchy under RegexSolverError, so a caller can catch the case it handles instead of matching on a message. ApiError carries status_code and body, and splits per status down to RegexSyntaxError, TimeoutExceededError, QuotaExceededError and the rest.
• A rate limiter, shared by every client holding the same token: a 429 sets a deadline from Retry-After, requests wait for it, and the operation is retried within a five-minute budget.
• Automatic batching: concat, intersection and union split a call carrying more terms than the account allows per request and fold the results back into one, each constituent request counting against the monthly quota. Disable with auto_batch=False, or lower the split with max_terms_per_request.
• A per-term cache of what the API has already returned — cardinality, length, pattern, dot, and the empty, empty-string, total and deterministic flags — so asking twice costs one request.
• Term.serialize() / Term.deserialize(), round-tripping the regex=<pattern> / fair=<payload> form, plus __eq__, __hash__, get_value(), to_dto() and from_dto().
• AccountLimits, Cardinality (Integer, BigInteger, Infinite), Length, ResponseFormat, PathOrder and CharacterOrder as exported models.
• generate-api.sh, which regenerates regexsolver/_generated/ from the specification the API publishes at https://api.regexsolver.com/openapi.json; .openapi-generator-ignore protects the hand-written files.
• CI running flake8 and mypy, then the tests on Python 3.10 through 3.14, on pushes to main, on pull requests, and before a release is published.
• A CHANGELOG.md, this file, a pull request template, and Dependabot updates.

Changed

• The HTTP layer is an asyncio client generated from the OpenAPI specification, on aiohttp, instead of hand-written requests calls, with regexsolver/ the hand-written surface over regexsolver/_generated/.
• Term is an abstract base class with RegexTerm and FairTerm subclasses rather than a pydantic model; Term.regex() and Term.fair() are unchanged.
• Term.get_pattern() and Term.get_fair() read the term's own value and its cache, returning None when the other format has not been resolved yet. Resolving a pattern is client.get_pattern(term).
• The minimum supported Python version is 3.10, declared in requires-python, up from 3.7.
• The package is built from pyproject.toml alone, with the test dependencies under the test extra.
• The tests drive the clients with unittest.mock over the generated API instead of matching URLs with requests_mock and loading fixtures from tests/assets/.

Removed

• RegexSolver, with get_instance() and initialize().
• The operation methods on Term: intersection(), union(), subtraction(), is_equivalent_to(), is_subset_of(), generate_strings() and get_details().
• Details, along with the regexsolver.details module, which returned cardinality, length and the empty and total flags in one response.
• setup.py, requirements.txt and test-requirements.txt.

Compatibility

• RegexSolver.get_instance().initialize(token) becomes RegexSolverClient(token), and a term method becomes a client method taking the terms as arguments: term1.union(term2) is client.union(term1, term2), is_equivalent_to is equivalent, is_subset_of is subset, subtraction is difference, and get_details is get_cardinality(), get_length(), is_empty() and is_total().
• ApiError no longer prefixes its message with The API returned the following error: ; the message text differs and status_code is what to branch on.
• The endpoints moved from https://api.regexsolver.com/api/* to /v1/*, and /api/analyze/details is split into one /v1/analyze/* endpoint per property. 1.1.0 is the lowest version that works against the API.
• The serialized regex= / fair= form is unchanged, so a term persisted by 1.0.x deserializes.

1.0.3 - 2024-08-11

Fixed

• Length.minimum is optional, so a response without one is parsed instead of raising a validation error.

1.0.2 - 2024-08-10

Fixed

• Cardinality.value defaults to None, so an infinite cardinality, which carries no value, is parsed instead of raising a validation error.

1.0.1 - 2024-08-09

Changed

• The README documents the full API.

1.0.0 - 2024-08-01

Initial release.