material 3.7.0

ES2022 browser JS.

Changed

• The browser JavaScript compiles at ES2022 (#264).
  The published browser JavaScript now compiles at ES2022 instead of
  ES2020, as one whole-dist target; useDefineForClassFields: false keeps
  today's assignment semantics for class fields, where at ES2022 the
  TypeScript default flips them to Object.defineProperty definitions (the
  flag differs from that default in two shipped files). The floor is
  measured, not assumed: parsing all 642 JS files of the published 3.5.4
  package with acorn 8.18 finds 186 that fail ES2019, 185 of which parse at
  ES2020 — so the browser floor today is ES2020, carried by optional
  chaining (153 files) and ?? (127 files) — and exactly one file,
  dist/ssr/index.js, that already fails ES2020 and ES2021 and ships
  logical assignment and native private fields; no file fails ES2022. The
  README's browser-support section names the new floor, and
  npm-readme.md carries the same sentence. The emit is smaller: of the
  560 shipped tsc-emitted files, 545 are byte-identical and 15 shrink by
  4,223 bytes raw in all — 4,026 of them dist/elements/define.js, the
  only file with native private members and methods (it also gains two
  static blocks for its static class fields) — while the other 14 drop 9 to
  21 bytes each as ??= stays native. Element pages measured from the
  packed package — each component's define* element plus its CSS, in
  brotli — are all down and none grows: 15,695 bytes over all 40 rows,
  mean 392, range 289 to 463. The component initial and deferred rows
  (SIZES.md, gzip) are byte-identical to main's — the factory pages carry
  none of the element runtime, and every deferred value is unchanged — so
  the table needs no update. The packed-consumer rows that do move:
  all-js 127,444 → 127,428 bytes gzip (−16), form 20,393 → 20,387 (−6),
  button-initial 7,566 → 7,563 (−3), and button, slider,
  navigation-rail and text-field by 2 to 3; apart from the SSR entry
  below, no gzip row grows, and every budget stays green. (all-js brotli
  moves +160 bytes and button +6 bytes: cross-chunk dictionary drift;
  size and size:check are the gate and both pass.) The one row whose
  gzip grows is the SSR entry: 112,599 → 112,609 bytes gzip (+10; raw
  308,694 → 308,739, +45; brotli +77). Cause: the dist/ssr build now
  applies the explicit useDefineForClassFields: false, so its copy of the
  classes assigns instance fields in the constructor where Bun left them
  native — which is how the browser dist has always emitted them — and it
  stays inside size:check's SSR budget, with the SSR consumer and browser
  checks green.

  Compatibility: The browser JS now requires an ES2022-capable parser
  (it was ES2020); bundlers that parse ES2020 but not ES2022 should update.

Internal

• The published syntax floor is checked (#264).
  bun run syntax:check parses every .js, .mjs and .cjs file the
  package ships at ecmaVersion: 2022 with acorn 8, and fails with
  file:line:column on the first file that needs a later edition. It runs in
  the package job on the build's dist, so the floor cannot drift silently.
  test/build/syntax-floor.test.ts passes what the ES2022 target emits —
  private members, a static block, class fields, private-in — and fails a
  regex v fixture (ES2024, first asserted valid at ecmaVersion: 2024)
  with file, line and message; it reads .js, .mjs and .cjs in
  subdirectories and ignores declarations, CSS and JSON.
  test/build/ci-commands.test.ts keeps the command in the CI list.

• The size cost of the ES2022 target: a smaller package, and one row
  larger.
  Against 3.6.0's published package job: all-js 414,644 raw / 127,428
  gzip / 105,827 brotli against 414,669 / 127,444 / 105,667; the full
  stylesheet 628,272 / 58,669 / 42,164 and the inline SSR page 253,639 /
  5,325 / 2,906, byte-identical; and the packed tarball 862,802 bytes
  compressed, 4,830,984 unpacked in 1,582 files against 863,526 and
  4,837,005 in 1,582 files — 724 bytes smaller packed and 6,021 unpacked in
  the same number of files. Every one of the 40 adapters:size element
  pages is smaller, like-for-like: 15,695 bytes brotli in all, mean 392,
  range 289 to 463; every adapter import stays within the check's
  1,700-byte brotli ceiling over its element. The component rows that move
  (raw / gzip / brotli): form 65,250 / 20,387 / 18,174 against 65,257 /
  20,393 / 18,225; button-initial 22,463 / 7,563 / 6,778 against 22,466
  / 7,566 / 6,774; button 41,301 / 13,940 / 12,648 against 41,304 /
  13,942 / 12,642; slider 34,967 / 12,165 / 11,018 against 34,970 /
  12,167 / 11,036; navigation-rail 19,340 / 7,072 / 6,407 against
  19,346 / 7,075 / 6,412; text-field 30,836 / 9,517 / 8,571 against
  30,842 / 9,520 / 8,574; and the SSR entry 308,739 / 112,609 / 91,640
  against 308,694 / 112,599 / 91,563 (+45 / +10 / +77, the cause in the
  Changed section). base-lite-css is 12,999 / 1,644 / 1,306 against its
  1,750-byte budget, and the size table's 41 rows are byte-identical to
  main's, so SIZES.md needs no update. Measured on main at 5fc52c2, push
  run 41800107765, package job 126473784266; this release changes only the
  CHANGELOG and the version field. No ceiling moved, every budget check
  passed, and every README size row stands as printed.

---

npm: material@3.7.0 · Docs: md3.io · Full history: CHANGELOG.md

The history before 3.0.0 was developed in floor/mtrl; #numbers in commit subjects before 3.0.0 refer to pull requests there.