Skip to content
PHIL
Français

Changelog

phil/translation-memory

All notable changes to phil/translation-memory will be documented in this file.

Updates should follow the Keep a CHANGELOG principles.

0.1.0-alpha - 2026-11-12

Added

  • LanguageTag: BCP 47 tags read and written in their canonical case, read from a locale such as fr_CA, and their parent, without the last subtag.
  • Text: the words of inline HTML — tags left out, character references decoded, white space normalized.
  • Fingerprint: the SHA-256 of a source with its white space normalized, read back from storage.
  • Segment and SegmentState: a key, a source, its fingerprint, a state from untranslated to reviewed, a target and the source it was written for; translate(), suggest(), review(), confirm(), clear(), withKey(), and withSource(), which makes a target stale when its source changes and translated again when the source comes back.
  • SourceChange: what changed in the source of a stale segment, as escaped HTML from phil/text-diff.
  • SentenceSplitter: sentences cut at ., !, ?, … and the CJK ends, with closing quotes, brackets and tags, known abbreviations and initials, tags never cut, the text given back whole.
  • UjoSegmenter, Granularity and TranslatableFields: UJO documents cut by text field or by sentence, nested blocks included, under keys made of the block identifier and the field path; translated segments put back, the white space around each sentence kept, only on the very source they were made from.
  • SegmentSync and SyncResult: stored segments brought up to date with a new cut of their source — kept, moved with their translation, stale, added, removed.
  • Variants: the regional variants that inherit from their parent language, fr-CA and en-CA by default, their chain, and the segments of a variant with the usable targets of its parent where it has none.
  • Memory\TranslationMemory, MemoryStore, InMemoryStore, MemoryUnit, MemoryMatch and MatchKind: exact matches by fingerprint, fuzzy matches in a window of lengths, sorted by trigrams and scored by their Levenshtein distance, one per target, best first, the units of a variant's parent after its own; segments learnt when translated or reviewed.
  • Memory\Similarity: the Levenshtein distance between code points, bounded, counted by PHP's levenshtein() on one byte per character when the two texts hold 256 different characters at most and in a band of the table otherwise, its score from 0 to 1, character trigrams and their Dice coefficient.
  • Glossary\Glossary, Term, TermHit, TermIssue and TermIssueKind: terms found as whole words, the longer of two overlapping ones winning, their inflected forms included; targets checked for the count of allowed or kept translations and their forms, and for refused ones outside an allowed translation.
  • Glossary\Contract\Inflector and RegularPlurals: the forms an inflector gives added to every term of a glossary; the regular plurals of English and French.
  • Check\Contract\Check, Checker, Checked, CheckIssue and Flags: quality checks of translated segments run in turn, each dismissed on a segment by its flag ignore-<check>; Punctuation, Spaces, Tags, Plurals — ICU plurals and the CLDR categories of whole numbers — and SameAsSource.
  • Check\Contract\TextWidth, AdvanceWidths, MaxSize and Glyphs: the rendered width of each line of a target in pixels from the advances of a font, checked against the segment's max-size flag or a limit; the characters the character map of a phil/fonts face does not cover.
  • Xliff\XliffWriter: XLIFF 2.1, one unit per segment named after its key, tags as <ph> codes with their HTML in <originalData>, states and phil: sub-states, the earlier source of a stale segment in a note.
  • Xliff\XliffReader: XLIFF 2.0, 2.1, 2.2 and 1.2 read into XliffFile and XliffUnit, inline codes turned back into their HTML, states mapped, document type declarations refused.
  • Xliff\XliffMerge, MergeResult, Skipped and SkipReason: an imported file merged into segments, its languages checked, units of a changed source or without target skipped, tags the source does not hold left out.
  • Tmx\TmxWriter and TmxReader: translation units written as TMX 1.4b, tags as <bpt>, <ept> and <ph>, and read back from TMX, document type declarations refused, 100 000 units at most.
  • Tbx\TbxWriter and TbxReader: a glossary written as TBX-Basic of ISO 30042:2019, and the terms of a pair of languages read from TBX of 2019 or 2008, statuses mapped, a variant read from its parent's section, document type declarations refused, 100 000 entries at most.
  • TranslationMemoryException, implemented by InvalidLanguage, InvalidSegment, InvalidTerm, InvalidXliff, InvalidTmx, InvalidTbx and InvalidConfiguration.
  • The documentation in doc/, one file per topic, indexed by the README.
  • CONTRIBUTING.md, SECURITY.md and CODE_OF_CONDUCT.md.