Skip to content

Chore: Move the Compose fork to 1.11.1-ext.4 - #808

Merged
b-re-w merged 7 commits into
developfrom
feat/compose-ext4
Nov 15, 2026
Merged

b-re-w merged 7 commits into
developfrom
feat/compose-ext4

Conversation

@b-re-w

@b-re-w b-re-w commented Nov 14, 2026

Copy link
Copy Markdown
Member

Moves the Compose fork from 1.11.1-ext.3 to 1.11.1-ext.4.

What moved

  • REVISION and MINGW_REVISION pin the fork's tag v1.11.1-ext.4, cadc5984ef96ed306ea12a7e967bcf945e7b1bdd.
  • WINDOW_REVISION pins the tip of the fork's extended when this started, a13a6289029d11033c892b94bc200e67aeed93d7 (fork Feat: One executable gate for the Kotlin/Native Windows renderer #235).
  • Both commits' gradle.properties publish COMPOSE and COMPOSE_MATERIAL3 as 1.11.1-ext.4, and both extended/skiko/build-skiko-awt.sh publish skiko-awt as 1.11.1-ext.4, so COMPOSE_AS, MATERIAL3_AS and SKIKO_AWT_AS are 1.11.1-ext.4.
  • Every live coordinate moves to ext.4: the renderer module.yaml files (desktop, macos, linux, windows), COMPOSE_VERSION in Transform3DLayers.kt and its staged Android copy (scripts/stage-android-kotlin.sh), the troubleshooting guides (en and ko), the scripts, the test comments and the workflow comment. The historical ext.3 CHANGELOG entry keeps its version; ext.4 gets its own entry under Unreleased.
  • compose-fork.changes and compose-fork-mingw.changes are regenerated from 73ac849 to the new pin. The one path they stop listing is extended/window/scene/.../DrawnEditMenu.kt, which fork Fix: Repoint fork dependencies on every build-compose.sh run #221 removed on purpose.

What the pins hold

The tag (since b97666a8): Java 11 class files for the design systems and window modules, graalvm-linux at Java 17 (fork #226); skiko-awt with one capability, the desktop modules naming the fork's skiko-awt (fork #167); the Windows edit menu C glue dxc_native_set_edit_menu, which the X11 and AppKit C answer themselves, and the scene leaving the text menu to Compose (fork #221); FR-82, an X11 window keeping its size in dp across a scale change (fork #216); F1 to F12 and the Menu key (fork #219); the Windows new size drawn before it is shown after maximise, restore or snap (fork #222); one menu on a right click in a macOS text field (fork #224); CI fixes #217, #218, #220.

The window pin beyond the tag (git log v1.11.1-ext.4..a13a6289 --first-parent): fork #232 (Windows restore and snap show the frame drawn ahead through the swapchain, the probe gated on a rate; #223 stays open), fork #235 (the macOS native view's frame committed in the frame's own transaction, #233), fork #234 (an application can declare a node's accessible role to the Kotlin/Native macOS window), and CI changes #229 and #231.

The Windows edit menu feature (#762) and the FR-82 CI checks (#726) are not here; they merge develop after this.

Glue

  • renderer/scripts/fork-dependencies.sh: since fork Docs: Point the single-executable rule at FR-22.6's format table #167 the desktop modules name org.thisisthepy.compose.skiko:skiko-awt:1.11.1-ext.4, a JVM jar published as a root alone. The check asked it for a skiko-awt-desktop publication, so require-fork-artifacts.sh desktop refused a complete ~/.m2. A root with no available-at variant now counts as published when its own files are; a multiplatform root without the target's publication is still refused. fork-dependencies.test.sh covers both (red before the fix).
  • scripts/tests/x11-key-codes.test.sh: its sed range ran past the one-line X11_DIGITS into the new X11_FUNCTION_KEYS (fork Fix: Run the window publish's kotlin.bat without Git's Unix tools on its PATH #219). Each table is read to its first closing parenthesis now, and the F keys and the Menu key are checked.
  • No new dxc_native_* stand-ins were needed: desktop-c-symbols.test.sh passes, 32 symbols across 3 desktops.

Artifacts

1.11.1-ext.4 is on Maven Central for the tag's modules (foundation root, desktop, macosarm64, iossimulatorarm64; window common and graalvm-macos; skiko-awt), but the window pin is past the tag and the renderer checks ~/.m2. So the repo's own scripts published ext.4 into ~/.m2 (ext.3 untouched): build-compose.sh --target desktop and --target macosArm64 at cadc5984, publish-skiko-awt.sh and publish-window.sh at a13a6289. iOS and Android use JetBrains' Compose and need none of it.

Gate (local, macOS arm64, head a06986c)

Step Command Result
a scripts/stand-in-renderer.sh --if-refused target/stand-in-renderer exit 0
b cargo fmt --all -- --check exit 0
b cargo clippy --workspace -- -D warnings exit 0
b cargo test --workspace exit 0, 980 passed, 0 failed, 32 ignored; design_system_parity 2/2 at cadc5984
c every scripts/tests/*.test.sh, plus renderer/window-modules-no-awt, linux-build, linux-metadata 86/86 exit 0. Skipped parts: appkit-mouse (this window server does not deliver a posted HID click), fork-artifacts-before-amper bash 4 case, web-stack-placement (no host wasm), x11-window-layout Linux offsetof check
d renderer/scripts/require-fork-artifacts.sh desktop macos exit 0
d ./kotlin test -m desktop exit 0, 1180/1180
e ./kotlin build -m ios -p iosSimulatorArm64 (fresh tasks, konanc ran) exit 0
f ./kotlin test -m macos exit 0, 45/45, none skipped
g dx build --package sample-minimal --platform android (app dir wiped) exit 0, APK with lib/arm64-v8a/libmain.so

What only CI can prove

The Compose and Windows pins move to the fork's v1.11.1-ext.4 tag
(cadc5984ef96ed306ea12a7e967bcf945e7b1bdd) and the window pin to the tip
of the fork's extended branch (a13a6289029d11033c892b94bc200e67aeed93d7).
Both commits publish every line as 1.11.1-ext.4, so COMPOSE_AS,
MATERIAL3_AS and SKIKO_AWT_AS move with them, and every live coordinate
follows: the renderer module.yaml files, COMPOSE_VERSION in
Transform3DLayers.kt and its staged Android copy
(stage-android-kotlin.sh), the troubleshooting guides in English and
Korean, the scripts, the test comments and the workflow comment.

ext.4 builds the design systems and window modules into Java 11 class
files (fork #226), publishes skiko-awt with one capability and points
the desktop modules at it (fork #167), and carries the Windows edit menu
C glue (fork #221), FR-82's X11 size in dp across a scale change (fork
#216), F1 to F12 and the Menu key (fork #219), the Windows maximise and
restore frame (fork #222) and one macOS text menu (fork #224). The window
pin adds fork #232 (Windows restore and snap show the frame drawn ahead,
probe gated on a rate, #223 open), #235 (macOS native view frame in the
frame's own transaction, #233), #234 (declared accessible roles on the
Kotlin/Native macOS window) and CI changes #229 and #231.

compose-fork.changes and compose-fork-mingw.changes are regenerated from
73ac849 to the new pin. The one path they no longer list is the scene's
DrawnEditMenu.kt, which fork #221 removed on purpose. The historical
CHANGELOG entry for ext.3 keeps its version and ext.4 gets its own.
Since 1.11.1-ext.4 (fork #167) the fork's desktop modules (ui-desktop,
foundation-desktop and the rest) name org.thisisthepy.compose.skiko:
skiko-awt at the fork version. skiko-awt is a JVM jar published as a
root alone, with no per-target publications, and publish-skiko-awt.sh
puts it in ~/.m2. fork-dependencies.sh still asked every root sibling
for a publication per served target (skiko-awt-desktop), so
require-fork-artifacts.sh desktop refused a complete repository and
every renderer build script stopped before the toolchain ran.

A root whose metadata has no variant pointing at a target publication
("available-at") is now published when its own files are. A
multiplatform root without the target's publication is still refused.
fork-dependencies.test.sh covers both; the first case failed before
this change with skiko-awt named as missing.
The test read X11_DIGITS from the fork's X11Keys.kt with a sed range
that ends at the next line holding a closing parenthesis. The digits sit
on one line, so the range used to end at the file's end; the window pin
now has X11_FUNCTION_KEYS after them (fork #219), and the range ran on
into that table: 12 digits, and digit 5 read as 97. Each table is now
read up to the first closing parenthesis, its opening line included.

The shared table's new F1 to F12 entries and the Menu key are checked
as well, since the X11 windows depend on them as they do on the rest.
The fork's X11 window numbers F1 to F12 since fork #219, which the ext.4 pin
brings. a_key_with_no_number_claims_no_meaning used F1 as its example of a key
the window does not number, so on Linux it failed with "Expected <-1>, actual
<122>" (batch CI run 43639736378, Test the linux renderer; the only failed
test of that job).

The example is now Scroll Lock, which the window still leaves unnumbered, and a
new test checks both ends of the function row reach the scene as F1 and F12.
Not compiled here: the Linux module builds only on Linux.
The window pin moves to the tip of the fork's extended branch, which
still publishes every line as 1.11.1-ext.4. It adds fork #239 (for
#233): the macOS window holds its backdrop blur still while a native
view slot shows, so the view and the frame of one turn are composited
together, place() works only where a slot changed, and
setFrameAndVisibleRegion sets both in one call. It also brings fork
#236 (the application owns a Windows window's caption strip) and #237
(quoted booleans in the publish workflows).
@b-re-w

b-re-w commented Nov 15, 2026

Copy link
Copy Markdown
Member Author

Merge evidence (compose-rust boost, w1:p1T).

  • CI: batch run 43666750736 on ci/batch-2026-11-15b (this branch at 0152794 with the window pin a187dd1e, plus develop and other branches): 76 jobs passed, including the Linux renderer x64/arm64, the GraalVM images on Linux, macOS and Windows, the Windows static renderer kit and its application checks, the mingw Compose build, the samples, and the heavy suites. One job failed, "hand-check under a scripted resize (Windows, x64)": its three release links passed, and the step then exited 1 on its last diagnostic line (a grep for .pdata section headers under pipefail that found no match). That step is on develop as it is and only ran here because this dispatch asked for the measure suite; it is not the pin. Fixed separately on Test: Drag the current Windows window pin on every edge, judged by edge markers (#17, #353) #684, which owns that workflow now.
  • Local gate (earlier heads of this branch): cargo 980, script tests 86/86, desktop JVM 1204, macOS module 46, forced iOS compile, dx Android build.
  • Since then only develop was merged in (b6feb6e, CHANGELOG.md kept both sides); the pin and fork script tests pass on it.

Known and tracked, not fixed by this move: no wheel or trackpad scroll on the Kotlin/Native macOS window (fork PR #240, needs the next fork tag).

@b-re-w
b-re-w merged commit 7f8e052 into develop Nov 15, 2026
1 check passed
@b-re-w
b-re-w deleted the feat/compose-ext4 branch November 15, 2026 15:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant