Prepare wikipedia 1.5.0

[goldsmith]

Fixes incorrect page matches, hanging requests, and errors when loading page
data. Adds regression tests, wheels, and CI for Python 3.10–3.14.

Changes

• Check exact titles before using search suggestions.
• Add HTTPS, connection reuse, configurable timeouts and bounded retries.
• Fix redirects, disambiguation, sections, HTML and missing page properties.
• Keep page objects tied to their original language. Limit module caches to
  256 entries each for five minutes and protect cached lists from mutation.
• Update packaging, dependencies and documentation.

Compatibility

Python 3.10 is the new minimum. Existing function names and positional arguments
are preserved. auto_suggest=True remains the default, but exact disambiguation
pages now raise instead of being replaced by a suggestion.

Requests default to a 5-second connect timeout and a 15-second read timeout,
with at most two retries. RequestError covers HTTP/API failures;
HTTPTimeoutError remains available. Missing lists return [] and absent
coordinates return None.

Tests

143 offline tests pass on Python 3.10–3.14 on macOS, including all 32 legacy
tests. Clean wheel/source installs and four LangChain integration tests pass.
Live checks passed on English, French and Japanese Wikipedia. Linux and Windows
results will be checked in CI.

Remaining work

This overlaps with Akber's packaging and HTTP PRs, listed below. Trusted
Publishing and Dependabot can be added on top. The remaining differences need
review before combining the changes.

Backlog

All 163 issues and 45 PRs open on September 28 are listed below. Related issues
are grouped. Some reports still need their original examples checked; this is
not a list of issues to close automatically.

45 pull requests

PR	This branch	Notes
#355	To compare	Akber’s HTTP and title-lookup PR overlaps with this branch. Compare timeout, retry, title-resolution and User-Agent changes before combining them.
#354	To combine	Akber’s packaging PR overlaps with this branch. Trusted Publishing and Dependabot are still to add; Python and dependency versions need agreement.
#351	Included unchanged	Niki’s equality fix, with original authorship preserved.
#343	Handled here	Exact titles are tried first; auto_suggest=True remains the default.
#336	Left for later	Formatting and type hints can be a separate PR.
#333	Left for later	Proxy environment variables are documented. A new setter can follow separately.
#310	Handled here	Exact lookup now happens before search, including when search returns nothing.
#305	Handled here	Missing titles get one suggestion fallback. Unrelated formatting is left out.
#296	Handled here	Credits are retained in the README and changelog.
#287	Patch not used	Indexes the first result before checking for an empty response. Both cases are tested here.
#277	Left for later	Including subsections would change section()’s documented behavior.
#261	Adapted	Tim Gates’s spelling correction is included.
#253	Patch not used	Uses comparison instead of assignment. Title lookup is fixed separately here.
#236	Patch not used	maxlag errors get bounded retries here. A configurable maxlag parameter is still deferred.
#235	Left for later	Error construction is fixed. Input trimming needs a separate decision.
#230	Patch not used	Language validation should not require CSV files or numpy.
#219	Partly adapted	Anna Katrine Jørgensen’s geosearch and coordinate fixes; revision history is deferred.
#215	Patch not used	Whitespace-only license change. The license is unchanged.
#206	Left for later	Existing proxy environment variables work. A new proxy API is deferred.
#205	Partly adapted	rowan08’s missing-property handling, with additional response checks. Registry changes are left out.
#195	Handled here	Dependency bounds are updated and tested.
#187	Left for later	New revision-timestamp property deferred.
#182	Left for later	Needs a signature that preserves existing positional calls.
#179	Left for later	Language-link lookup needs its own API and pagination tests.
#173	Left for later	Category traversal needs pagination and namespace handling.
#172	Handled here	Old examples are replaced.
#170	Handled here	Dead badges are removed.
#160	Adapted	Christopher Beland’s HTTPS proposal is included.
#159	Left for later	Custom endpoints need a separate support decision. Tests can replace the internal transport.
#157	Adapted	Michael E. Rose’s absent-coordinate handling; malformed data still raises an error.
#147	Left for later	Exposing arbitrary pageprops is deferred.
#137	Adapted	Emmanuel Bizieau’s timeout proposal, applied to every request.
#132	Adapted	mkasprz’s handling of unavailable image URLs, with response validation.
#131	Handled here	One fallback replaces recursive suggestion attempts.
#128	Already on master	Explicit html.parser selection is retained.
#121	Handled here	Tests now use isolated fixtures and run against installed packages.
#115	Already on master	The page-parameter fix is retained; TOC handling is updated.
#112	Already on master	Explicit html.parser selection is retained.
#110	Left for later	Pagination needs to preserve existing positional calls.
#109	Handled here	Documentation and examples are updated.
#101	Already on master	The section-request parameter fix is retained.
#100	Left for later	Translation lookup needs a separate API and tests.
#92	Left for later	Extra disambiguation context would change the return type.
#84	Already on master	The section-request parameter fix is retained.
#64	Adapted	David Lutz’s connection-reuse proposal, without a new public session API.

163 issues, grouped by problem

Reports	What this PR does / what remains	Issues
Wrong page or missing page (29)	Exact titles are checked before search. Tests cover normalization, typos and missing pages; the original examples still need confirmation.	#346, #344, #342, #340, #321, #320, #316, #314, #307, #304, #295, #291, #273, #266, #258, #256, #227, #217, #204, #199, #192, #190, #186, #108, #104, #102, #85, #58, #54
Suggestion ranking still needs work (2)	Partly addressed. When the exact title is absent, suggestions can still outrank better search results.	#284, #259
Connection failures, busy servers and invalid responses (20)	HTTPS, timeouts, retries and clearer errors are added. Outages, blocked requests and server settings still need investigation.	#353, #334, #303, #294, #292, #281, #271, #242, #220, #202, #197, #175, #155, #154, #152, #142, #130, #122, #98, #96
Missing links, images and other page data (8)	Missing properties and pagination are tested, including repeated tokens and failed requests. The original hanging and page-size examples need confirmation.	#290, #285, #133, #93, #87, #78, #71, #65
Disambiguation options (2)	Options now use unique article titles instead of display labels.	#244, #79
Language and geosearch (1)	Geosearch clears its cache after a language change. Existing pages keep their original language.	#218
Sections (5)	TOC parsing, empty sections and heading matching are fixed. section() still stops at the next heading.	#168, #119, #105, #95, #77
Redirects and page IDs (2)	Page-ID redirects follow MediaWiki’s response and respect redirect=False.	#164, #118
Empty queries (10)	Error construction is fixed. Trimming or rejecting empty input consistently is still deferred.	#262, #233, #223, #213, #203, #200, #185, #162, #135, #89
User agents (3)	Application names and contact details are documented. set_user_agent() remains optional.	#350, #322, #126
Text formatting and collection order (8)	Current limits are documented. Article layout, rich markup and citation extraction are unchanged.	#330, #288, #237, #183, #145, #140, #114, #106
Parser warnings (5)	The parser fix was already on master and is retained.	#323, #279, #264, #225, #107
Search length limit (1)	The API error is exposed. The upstream search limit is unchanged.	#318
Packaging and old badges (3)	Packaging and CI are updated; stale badges are removed. Old installation failures may still need reproduction.	#270, #254, #167
A new PyPI release (1)	Builds and release notes are prepared. Nothing has been published to PyPI.	#251
Old Python or copied-install syntax errors (2)	Python 3.10+ installs are tested. Old interpreters and manually copied installations remain unverified.	#81, #67
Async and independent client configuration (3)	Deferred. Configuration remains global; async support needs a separate design.	#338, #337, #146
Infoboxes, tables and link labels (5)	Deferred. Existing text and page properties are retained.	#315, #212, #178, #111, #221
Categories and category members (3)	Category names are supported. Category traversal is deferred.	#265, #180, #134
Other languages and MediaWiki sites (6)	Translation lookup and custom endpoints are deferred. Existing pages keep their original language.	#188, #156, #144, #127, #117, #88
Proxy configuration (2)	Requests’ proxy environment variables are documented. A new setter is deferred.	#313, #184
More ways to identify or inspect pages (6)	Linked IDs, URL lookup, page status, history and redirect lists are deferred.	#232, #231, #181, #166, #113, #90
Other new query features (5)	Random-page filtering, more search results, user information and batch queries are deferred.	#228, #139, #125, #97, #70
A page with a disambiguation companion (1)	An article can have a separate disambiguation page without being one itself. Detection uses pageprops.	#306
Search rankings (1)	Search ordering comes from MediaWiki. Exact-title lookup does not change search() rankings.	#289
ODD_ERROR_MESSAGE (1)	The constant remains available. Redirect handling no longer relies on it.	#272
Handling missing or ambiguous pages (3)	Callers should handle PageError and DisambiguationError and choose from the returned options.	#224, #151, #136
Unicode input (1)	Pass a Unicode title, not an already URL-encoded string.	#189
Project-list inclusion (1)	Requires a maintainer reply; no code change.	#174
Geosearch radius (1)	MediaWiki’s radius constraint is unchanged. Invalid parameters raise an API error.	#169
Summary URL (1)	Use page(title).url and page(title).summary.	#161
Simple English Wikipedia (1)	Use set_lang("simple").	#158
Reports needing a reproducible example (8)	Need a small example, package/Python versions and the actual error response.	#329, #328, #325, #324, #319, #302, #257, #226
Authorization-header error (1)	Need response and environment details. This package does not add access-token authentication.	#348
Missing exception attributes (1)	Exception attributes are tested. The reported failure still needs a reproducible example.	#250
Missing fullurl (1)	Malformed metadata raises RequestError. The original cause remains unconfirmed.	#239
French disambiguation report (1)	Need the pageprops response to check whether the page was actually a disambiguation page.	#201
Console encoding (1)	Unicode repr() is fixed and tested. Output-stream encoding and the original screenshot remain unverified.	#83
Reports outside this package (7)	No package change proposed. These reports concern websites or unrelated content.	#345, #341, #332, #327, #326, #283, #275