material 3.7.1

Lighter modals.

Changed

• The modal components ship less code (#267, #268).
  The shared modal layer and the focus helpers behind dialogs, sheets,
  drawers and pickers are slimmed, and the <dialog> cancel handling is
  centralized in one place, with no public API change and no event payload
  or DOM contract change.

  #267 removes onModalEscape's dependence on eventsFrom: a per-Window
  capture-listener reference count and the opening-event clock recorded on
  the Escape stack entry replace it, so dialog, bottom-sheet,
  side-sheet, drawer, timepicker and search tree-shake
  eventsFrom out of their bundles completely, while menu still
  tree-shakes onModalEscape. The layer.ts state — hidden elements,
  opening order and the per-window count — becomes one WeakMap, and the
  top-layer, inert, Escape and tab-wrap helpers are compacted;
  activeElementOf is untouched, so non-modal components do not move. The
  seven modal components drop their duplicate cancel listener and their
  handleCancel / onCancel / asking boilerplate, and the capture-phase
  cancel listener on <m-dialog> and <m-drawer> still runs first and
  can refuse closure with preventDefault() and
  stopImmediatePropagation().

  #268 removes the glue the components kept beside that layer: dialog
  routes beforeopen / beforeclose, open/close/afteropen/afterclose and
  its timer and focus-frame teardown through shared emit / ask /
  stopOpen helpers; bottom-sheet and side-sheet unify their
  trap/stack and scrim/backdrop handlers; drawer derives its layered
  state from the top-layer element instead of a per-module WeakSet;
  datepicker folds its Escape and eventsFrom references into one; and
  timepicker folds its click-outside handler into the dialog listener.

  Behaviour: No behaviour change from 3.7.0. The centralized cancel
  keeps the event cancelable and emits it only for the top-layer dialog
  and drawer, as before — not for every modal — and datepicker,
  timepicker, search, bottom-sheet and side-sheet still close from
  the Escape callback, so a capture listener cannot newly refuse Escape on
  them. The shared glue keeps an already-open non-top sheet from trapping
  at construction, and keeps the top-layer Tab listener across close and
  reopen. Both review fixes restore 3.7.0 behaviour that intermediate
  heads had changed; those heads never shipped.

  Size: The whole root bundle drops 342 bytes gzip — all-js 127,428
  → 127,086 (raw 414,644 → 412,325) — and the packed tarball drops 449
  bytes (862,803 → 862,354). Every component row that moves is smaller, in
  gzip bytes: dialog 12,598 → 12,440; bottom-sheet 7,361 → 7,185;
  side-sheet 6,839 → 6,653; drawer 8,481 → 8,365; datepicker 13,041 →
  12,995; timepicker 11,256 → 11,159; search 11,134 → 11,041; menu 12,697
  → 12,660; fab-menu 8,369 → 8,366; snackbar 12,236 → 12,223; tooltip
  6,165 → 6,149; list 6,839 → 6,838. No deferred row grows: fab-menu
  10,666 → 10,636; menu 2,122 → 2,119; select 10,650 → 10,623;
  split-button 17,245 → 17,210. Every budget check passes and no ceiling
  moved.

Internal

• The size cost of the modal diet.
  Against 3.7.0's published package job (main 9a98d851, push run
  41802819451, package job 126482889118): the packed rows, raw / gzip /
  brotli, are all-js 414,644 / 127,428 / 105,827 against 412,325 /
  127,086 / 105,325 (−2,319 / −342 / −502); the full stylesheet 628,272 /
  58,669 / 42,181, the inline SSR page 253,639 / 5,325 / 2,906 and the SSR
  entry 308,739 / 112,609 / 91,640 are byte-identical; base-lite-css is
  12,999 / 1,644 / 1,306 against its 1,750-byte budget; and the packed
  tarball is 862,354 bytes compressed / 4,827,107 unpacked in 1,582 files
  against 862,803 / 4,830,984 in the same number of files — 449 bytes
  smaller packed and 3,877 unpacked. The one brotli-only move is
  navigation-rail 6,407 → 6,419 (+12) with raw and gzip equal; as the
  ES2022 release reported, gzip is the gate and it passes.

  The size table's rows move with the code (minified / gzip / vs core, as
  SIZES.md prints them): dialog 38.7 KB / 12.3 KB / +9.3 KB against 37.5
  / 12.1 / +9.1; bottom-sheet 19.9 / 7.2 / +4.1 against 19.2 / 7.0 /
  +4.0; side-sheet 18.3 / 6.7 / +3.6 against 17.6 / 6.5 / +3.5; drawer
  23.4 / 8.3 / +5.2 against 22.9 / 8.2 / +5.1; datepicker 37.2 / 12.7 /
  +9.7 against 37.1 / 12.7 / +9.7; timepicker 31.6 / 11.0 / +8.0 against
  31.4 / 10.9 / +7.9; search 32.4 / 10.9 / +7.8 against 32.0 / 10.8 /
  +7.7; menu 38.6 / 12.4 / +9.4 against 38.4 / 12.4 / +9.3; and the all
  row 393.0 / 122.9 / +119.8 against 390.7 / 122.5 / +119.5, its +2.1 KB
  deferred figure unchanged. The exact gzip behind the table, from the
  budget lines: dialog 12,598 → 12,440; bottom-sheet 7,361 → 7,185;
  side-sheet 6,839 → 6,653; drawer 8,481 → 8,365; datepicker 13,041 →
  12,995; timepicker 11,256 → 11,159; search 11,134 → 11,041; menu 12,697
  → 12,660; fab-menu 8,369 → 8,366; snackbar 12,236 → 12,223; tooltip
  6,165 → 6,149; list 6,839 → 6,838; and no deferred row grows (fab-menu
  10,666 → 10,636; menu 2,122 → 2,119; select 10,650 → 10,623;
  split-button 17,245 → 17,210). The budget's all scenario is 125,800 →
  125,455 bytes gzip against its unchanged 127,200-byte budget.

  The 40 adapters:size element pages (brotli) are 907 bytes smaller in
  all: 14 rows down, 23 identical, and three up by cross-chunk dictionary
  drift (slider +22, navigation-bar +4, list +3); every import stays
  within the check's 1,700-byte brotli ceiling over its element.

  Measured on main at 1cd4f82e, push run 41939743447, package job
  126957651455; this release changes only the CHANGELOG and the version
  field. SIZES.md already carries the merged tree's rows, measured at
  af69fe66, and no README size row moves: the README prints all-js at
  127.1 kB, which is the 127,086 the package job measured. No ceiling
  moved, and every budget check passed.

---

npm: material@3.7.1 · 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.