Skip to content

0.2.0 - 2026-10-18

Latest

Choose a tag to compare

@github-actions github-actions released this 18 Oct 11:52
v0.2.0
fe2424e

Release Notes

Added

  • Minimum supported Rust version declared: rust-version = "1.88" (set
    by the icu_* 2.3 dependencies), checked by an msrv CI job.
  • Tests: every fixture is also read inside a sentence (a word, a line
    break, parentheses, a comma or a period around it) and must leave the
    surrounding text byte-for-byte intact; spans may not start or end with
    whitespace or cross a line break; the random-input tests include
    non-breaking and other Unicode spaces.

Changed

  • Breaking: Language is #[non_exhaustive]; a downstream match
    needs a wildcard arm, so adding a language is no longer breaking.
  • Breaking: the generated CLDR tables (verbalize::data) are no
    longer public, and the doc-hidden check_rbnf is removed.
  • Options and Agreement are documented as struct literals built with
    ..Default::default(), since a 0.x minor release may add fields.
  • Idempotence is no longer promised for every string: a reading can
    expose a token the first pass saw only glued into a larger one
    (www.example.com/Nr. → … Nr. → … Nummer). It holds for every
    fixture input and the random digit-rich strings of the property tests.
  • Release workflow: the GitHub release is created only after cargo publish succeeds, so a failed publish no longer leaves a release for a
    version that is not on crates.io; re-running the job skips crates
    already published. Every action in release.yml is pinned to a commit
    SHA.
  • cargo xtask cldr-update swaps the new data in only after every file
    downloaded, so a failed download no longer deletes xtask/cldr;
    gen/check-data fail when the vendored CLDR version differs from
    xtask/cldr.lock. xtask no longer depends on verbalize.
  • tools/icu-diff compares negative values too, fails when ICU's public
    rulesets differ from the shipped ones or a ruleset gets no comparisons,
    reports comparisons made, and pins PyICU==2.16.2.

Fixed

  • A degree value or dose scheme followed by a non-breaking or other
    multi-byte space and a letter (270° + U+00A0 + Nord, 1-0-1 +
    U+00A0 + mg) panicked. A chemical count beyond u32
    (H5000000000O, H₅₀₀₀₀₀₀₀₀₀O) panicked in debug builds.
    de/en/ro.
  • Text after a reading is kept intact: Es hat 20° in Berlin was
    "zwanzig Gradin Berlin", 1,5 × 10³ Zellen was "… hoch dreiZellen";
    numbers, units, signs and abbreviations are never joined across a line
    break, so Pagina 21 at the end of a line no longer takes "de" from
    the next line, and a phone number keeps the parentheses around it. Two
    readings that touch are separated: Nr.2023 was
    "Nummerzweitausenddreiundzwanzig", ca.5 kg "zirkafünf Kilogramm".
    de/en/ro.
  • A unit after a comparison sign or scale word is read when more text
    follows (≤10 mg pro Tag → "kleiner oder gleich zehn Milligramm pro
    Tag"), and 2x., 3-mal,, ∞., 30 €/Monat. are read before
    punctuation. (§) is "(Paragraf)" with no stray space. de/en/ro.
  • Negative amounts with a scale word or three or more decimals said
    "minus" twice (-2,5 Mrd. €, USD -5.125); -1 Mio. / -1 mil.
    take the singular scale word. de/en/ro.
  • German: um 1 Uhr is "um ein Uhr" (was "eins Uhr"). After an article
    or preposition, N. before a lower-case adjective and its noun is an
    ordinal with the adjective's ending (der 3. große Sieg → "der dritte
    große Sieg", nach 2. langer Pause → "nach zweiter langer Pause"). An
    angle of one minute or degree is singular (30°1' → "dreißig Grad
    eine Minute", "one minute"). de/en.
  • Romanian: 1 ¼ is "unu și un sfert" (was "unu și sfert"); al II-lea5
    reads the ordinal and the number as separate words. ro.
  • Plural forms and ordinal suffixes of numbers above 2⁶⁴ follow the full
    value instead of one clamped to u64::MAX. en/ro.
  • CLI: piping into a command that exits early (… | head -1) no longer
    panics; a failing --threshold still exits 1. sqlite: paths open
    read-only, so a mistyped path is an error instead of an empty, passing
    database; columns are chosen by SQLite's TEXT affinity (VARCHAR),
    BLOB/NULL values are skipped, and tables like sqlitex are scanned.
    Symlinked directories are not followed (no loops). annotate escapes
    tab, newline, carriage return and backslash in its TSV output.
    --region must belong to --lang.
  • tools/corpus/check.sh --strict fails when no corpus (or only an empty
    directory) is present instead of passing having checked nothing.
  • Docs: Normalizer::with_options no longer says English and Romanian
    return the input verbatim; the README normalize example passes
    --lang de.

Install verbalize-cli 0.2.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/relu/verbalize/releases/download/v0.2.0/verbalize-cli-installer.sh | sh

Install prebuilt binaries via powershell script

powershell -ExecutionPolicy Bypass -c "irm https://github.com/relu/verbalize/releases/download/v0.2.0/verbalize-cli-installer.ps1 | iex"

Download verbalize-cli 0.2.0

File Platform Checksum
verbalize-cli-aarch64-apple-darwin.tar.xz Apple Silicon macOS checksum
verbalize-cli-x86_64-apple-darwin.tar.xz Intel macOS checksum
verbalize-cli-x86_64-pc-windows-msvc.zip x64 Windows checksum
verbalize-cli-aarch64-unknown-linux-gnu.tar.xz ARM64 Linux checksum
verbalize-cli-x86_64-unknown-linux-gnu.tar.xz x64 Linux checksum