Search by

code-evolve / label-filter

spungin

A small glob-like pattern language for filtering lists of objects by their labels. Parsed to an IR, never compiled to a regex.

v0.4.0 2026-11-06 14:31 UTC

This package is auto-updated.

Last update: 2026-11-06 14:32:35 UTC


README

A small pattern language for filtering lists of objects by their labels. Parsed to an intermediate representation and matched directly, never compiled to a regular expression, so a half-typed pattern cannot throw and cannot backtrack.

PHP 7.4 or later. No dependencies, and none for the tests either.

composer require code-evolve/label-filter
use CodeEvolve\LabelFilter\LabelFilter;

$filter = LabelFilter::compile('\b\app&*.pdf');

if ($filter->error !== null) {
    // Show the reason beside the input. A refused pattern filters nothing out,
    // so the list the user is looking at stays exactly as it was.
    echo $filter->error;
}

$kept = array_filter($labels, fn ($label) => $filter->matches($label));

Two things that are deliberate, because they are what a filter box needs:

  • A pattern never throws. Every input compiles; an unusable one sets error and matches everything, which keeps the list visible while someone is still typing.
  • compile() once, matches() per row. The pattern is parsed a single time.

The language, in seven lines

apple                 contains, case-insensitive: the default
apple*pie             * is any run of characters
`apple  apple`        anchored to the start / to the end
[0-9].pdf             one character from a set: [a-z], [A-Za-z], [0-9A-F]
*.||md|txt||          a group, so a shared part is written once
\b\app                whole word: app, app-1, my app, app_data: not apple
\-\test               not: hide every label the pattern finds
a&b                   and: every item must match

The specification is docs/syntax.md, which ships with this package and is also readable at code-evolve/label-filter. Seven characters are reserved and are written with a backslash to mean themselves: *, |, &, a backtick, \, [ and ].

A trailing & or | is ignored, so apple& and apple| mean apple while the next item is being typed. An escaped \& or \| at the end is a literal and is kept.

One language, five implementations, one fixture

This is the PHP port. The same language is implemented in TypeScript, Rust, Python and Go, and all five run against the same file of cases, so they cannot disagree quietly.

php tests/conformance.php     # from this package: 96 spec cases, 568 assertions

The fixture is conformance/cases.json: cases lifted from numbered sections of the specification, patterns that must be refused, and fragments of junk that must not crash anything.

Where this package is developed

The source of record is the php/ directory of code-evolve/label-filter, which holds all five implementations, the specification they share and the fixture they are all tested against. Issues and pull requests belong there.

The repository Packagist reads is generated from that directory, file for file, and carries nothing of its own.

Licence

MIT. See LICENSE.txt.