v0.4.3 — DuckDB 2.0 named-argument compatibility

Compatibility release. No behaviour changes on the v1.5 line; the same source tree now also builds and binds correctly on DuckDB 2.0.

What this fixes

urlpattern_init, url_build and url_modify take named arguments (url_build(protocol := 'https')). DuckDB 2.0 changed how such a function is declared and how it learns its argument names, and only the first half was a compile error.

• Declaration. SimpleFunction::varargs is gone. 2.0 splits it into two parameter kinds: AddArgs (VAR_POSITIONAL, *args) and AddKwargs (VAR_KEYWORD, **kwargs). These functions take keyword arguments, so they now declare AddKwargs plus SetRequiresExpressionNames(true) — matching how DuckDB's own union_value is declared.
• Argument names. On v1.5 a named argument's name is an alias on the bound child (GetAlias()). On 2.0 aliases are empty and names arrive via BindFunctionInput::GetArgumentNames(). A bind body that kept reading GetAlias() would compile on 2.0 and then treat every argument as positional — silently wrong. Both sources are now resolved once into a single list, with positional arguments normalised to the empty string so the bind bodies read identically on both lines.

Verification

Both lines, one source tree, on a 36-core host:

	DuckDB 2.0 b15358c371d	v1.5 line b155d6f63ce
build	0 errors	0 errors
tests	162 assertions, all passed	162 assertions, all passed
url_build(protocol := 'https', hostname := 'example.com', pathname := '/api')	https://example.com/api	identical
url_modify('https://example.com/api', pathname := '/v2')	https://example.com/v2	identical
urlpattern_init(pathname := '/users/:id')	init:A=/users/:id;	identical

b15358c371daf10dad799a43b6167f6ccdce84d8 is the community registry's current default DuckDB (duckdb/community-extensions#3079).

Negative control, because a clean build is not evidence here: declaring *args instead of **kwargs compiles with zero errors and then fails at load with *"A function signature can only have one 'args' parameter". The correctness claim above rests on the queries, not the build.

Closes #14. See #15 for the full discussion, including two C++ traps (GetAlias() returning Identifier on 2.0 in a non-template overload compiled on both lines, and two-phase lookup rejecting the name Identifier inside a discarded if constexpr branch).

v0.4.2 — v1.5.6 line marker (retroactive)

Marks the commit the community registry has been serving as urlpattern 0.4.2.

This tag is retroactive: the descriptor published version 0.4.2 from this commit,
but no v0.4.2 tag existed, so the v1.5-line release point was only findable as a
bare SHA. Created ahead of DuckDB 2.0.0 (2026-10-21), after which the registry's
ref field targets the v2.0 line and v1.5 builds move to the new variegata
field — which needs a findable commit for any v1.5.6 bugfix release.

No code change; this is the same tree the registry already serves.

v0.4.1 — DuckDB v1.5.x compatibility

Compatibility + CI update over 0.4.0:

• Use BaseExpression::GetAlias() accessor for DuckDB main/v1.5.5 compatibility.
• Bump duckdb + extension-ci-tools to v1.5-variegata; build stable leg against the v1.5.4 release tag; re-enable Windows.
• Regression coverage for the issue #5 fixes (percent-decode, init ';').

Fixes the missing community build for DuckDB v1.5.5 (the previous listing pinned a pre-compat commit). Test suite green (162 assertions).

v0.4.0 — Correctness & hardening (issue #5)

Correctness & hardening (issue #5)

• Query params are now percent-decoded. ParseQueryString previously returned raw substrings, so url_search_params('...?q=hello%20world')['q'] yielded hello%20world. It now decodes %XX and +→space, matching WHATWG URLSearchParams.
• init: serialization no longer truncates on ;. A component value containing ; (e.g. search := 'a;b') was silently cut at the delimiter; the encoding is now escaped/length-safe.
• Case-insensitivity self-test. The provider compensates for an inverted set_case_sensitive in DuckDB's RE2 wrapper; a self-test now asserts the compensation so a future upstream fix can't silently invert ignore_case.
• Bounded pattern cache. The per-local-state pattern cache is now size-capped (LRU) so a query with many distinct large patterns can't grow memory unbounded.
• urlpattern_extract NULL semantics. No-match / group-absent / empty-value are now distinguishable instead of all returning ''.

(Regression tests confirming the percent-decode and init: ';' fixes have since landed on main and will roll into the next release.)

DuckDB v1.5.4

Built against DuckDB v1.5.4; ReDoS-immune by construction (patterns compile to RE2, linear-time). Stable builds green on all platforms.

Known follow-up

The duckdb-next-build leg (DuckDB main/v1.6) does not yet compile — BaseExpression::alias became a protected Identifier and needs GetAlias(). Tracked separately; does not affect this v1.5.4 release.

v0.3.1

Update for DuckDB v1.5 compatibility.

v0.3.0 - DuckDB v1.4.4 Support

What's New

• Updated to DuckDB v1.4.4
• Updated extension-ci-tools to v1.4.4

Supported Platforms

• Linux AMD64
• Linux ARM64
• Linux AMD64 (musl)
• macOS AMD64
• macOS ARM64

Installation

INSTALL urlpattern FROM community;
LOAD urlpattern;

v0.1.0 - Initial Release

DuckDB URLPattern Extension v0.1.0

Initial release of the URLPattern extension for DuckDB, implementing the WHATWG URLPattern API.

Features

• Custom URLPATTERN type with validation and implicit casting from VARCHAR
• Pattern matching: urlpattern_test(pattern, url) - test if a URL matches a pattern
• Group extraction: urlpattern_extract(pattern, url, group_name) - extract named groups
• Full match results: urlpattern_exec(pattern, url) - returns STRUCT with all match details
• Pattern accessors: urlpattern_pathname(), urlpattern_protocol(), urlpattern_hostname(), urlpattern_port(), urlpattern_search(), urlpattern_hash()
• Pattern caching for improved performance on repeated patterns

Quick Start

-- Load the extension
LOAD urlpattern;

-- Test if URLs match a pattern
SELECT urlpattern_test('https://example.com/users/:id', 'https://example.com/users/123');
-- true

-- Extract named groups
SELECT urlpattern_extract('https://example.com/users/:id', 'https://example.com/users/123', 'id');
-- '123'

-- Get full match results
SELECT urlpattern_exec('/posts/:slug', '/posts/hello-world');
-- {matched: true, pathname: '/posts/hello-world', groups: {slug: 'hello-world'}, ...}

Supported Platforms

Platform	Status
Linux (amd64)	✅
Linux (arm64)	✅
Linux (musl)	✅
macOS (amd64)	✅
macOS (arm64)	✅
Windows	❌ (MinGW compatibility issue)
WASM	❌ (emscripten compiler issue)

Installation

Download the appropriate extension for your platform from the release assets, or build from source.

Documentation

See the documentation for full API reference.