feat(windows): complete WinUI 3 backend - #8733
Merged
Merged
Conversation
Lands #8722 with its four gate blockers resolved. Replaces the windows-winui scaffold with real WinUI 3 / Fluent rendering through a vendored Windows Reactor snapshot (microsoft/windows-rs 65066a7109c214f317ed66261cfb7518160b8aaf), upstream licenses preserved. The vendored tree moved from `crates/perry-ui-windows-winui/vendor/` to `third_party/windows-winui/`, and this was forced rather than chosen. Cargo's `is_excluded` is `!explicit_member && excluded`, and `explicit_member` matches any `members` entry that is a path PREFIX of the candidate -- so while the snapshot sat under a member directory, `exclude` lost and all 13 crates were real workspace members. Verified empirically: adding `exclude = ["crates/perry-ui-windows-winui/vendor"]` in place left cargo reporting 92 packages with all 13 still members; after the move it reports 79, matching the existing architecture baseline exactly, so no baseline refresh was needed. As members they entered the build graph, so `cargo test --workspace` compiled them and Perry's `[workspace.lints]` would have applied to upstream code. GC roots, the part that mattered: `Node` DOES hold JS values -- every widget callback is stored as a raw closure pointer via `js_nanbox_get_pointer` -- so it gets scanner coverage following #8713's pattern, not a verdict. `AppState` does not: `String`, two `f64` dimensions, an `i64` that indexes NODES rather than an address, two `Option<(f64,f64)>` and a `PresenterKind`, so it takes a `not_a_gc_pointer` verdict. Three further roots the census had not flagged are scanned too (`ON_ACTIVATE`, `ON_TERMINATE`, `PENDING_TIMERS`), and the scanner is armed on the Fluent path, where winui shadows `app_create` and would otherwise have left `perry-ui-windows`' own tables unscanned. The four oversized files are generated upstream bindings and are allowlisted as an immutable snapshot. `build_and_run.rs` was NOT allowlisted -- the block this PR added moved to `link/winui_assets.rs`, 2110 -> 1970. No version bump; the only root Cargo.toml change is the `exclude` entry.
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (296)
📝 WalkthroughWalkthroughThis PR adds an opt-in ChangesPerry WinUI Backend Integration
Vendored Windows Reactor Toolkit
Estimated code review effort: 5 (Critical) | ~240 minutes Sequence Diagram(s)sequenceDiagram
participant Compiler as Perry compile/link
participant WinUIAssets as winui_assets.rs
participant AppBackend as perry-ui-windows-winui::app
participant Reactor as windows_reactor::App
participant Win32 as perry-ui-windows (fallback)
Compiler->>WinUIAssets: deploy_winui_runtime_assets()
WinUIAssets-->>Compiler: Bootstrap DLL, resources.pri copied
AppBackend->>AppBackend: app_run()
alt Fluent backend active and WinUI runtime available
AppBackend->>Reactor: run(root_factory)
Reactor-->>AppBackend: WinUI window rendered
else WinUI unavailable
AppBackend->>Win32: app_run()
Win32-->>AppBackend: Win32 window rendered
end
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This was referenced Aug 24, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Lands #8722 with its four gate blockers resolved.
The vendored tree moved — and this was forced, not chosen
I verified this rather than accepting it. Cargo's
is_excludedis!explicit_member && excluded, andexplicit_membermatches anymembersentry that is a path prefix of the candidate. While the snapshot sat undercrates/perry-ui-windows-winui/vendor/, the member entry for its parent won and all 13 crates were genuine workspace members.Empirically, on the PR head:
windows-*as membersexclude = ["crates/perry-ui-windows-winui/vendor"]added in placethird_party/windows-winui/+ excluded79 matches the existing architecture baseline exactly, so no baseline refresh was needed. This mattered substantively, not cosmetically: as members they entered the build graph, so
cargo test --workspacecompiled 73k lines of upstream code and Perry's[workspace.lints]policy would have applied to it.GC roots — the part that actually mattered
Nodeholds JS values. Every widget callback is stored as a raw closure pointer viajs_nanbox_get_pointer(Common::on_click, pluscallbackon Button/TextField/SecureField/Toggle/Slider). So it gets scanner coverage following #8713's pattern — not anot_a_gc_pointerverdict, which would have been false.AppStatedoes not. Its fields areString, twof64dimensions, ani64that indexesNODESrather than an address, twoOption<(f64,f64)>, and aPresenterKind. Rule S fired on the scalars; the verdict is accurate.Three further roots the census had not flagged are scanned as well —
ON_ACTIVATE,ON_TERMINATE,PENDING_TIMERS, all raw closure pointers. And the scanner is armed on the Fluent path, where winui shadowsapp_create, which would otherwise have leftperry-ui-windows' own live tables unscanned.The other two
Four oversized files are generated upstream bindings (up to 26k lines) and are allowlisted as an immutable snapshot that must stay byte-identical for re-diffing.
build_and_run.rswas deliberately not allowlisted — the block this PR added moved intolink/winui_assets.rs, byte-verbatim apart from twopub(super)bumps. 2110 → 1970.Validation
lint-job checkers passcargo check -p perryand-p perry --tests: 0 errors, 0 warningsCargo.tomlchange is theexcludeentry with its rationaleKnown residual — flagged, not fixed
render_handleoperates on a clone of the node and copies the unboxed callback pointer intomoveclosures owned by the Reactor element tree;start_runtime_pumpandapp_rundo the same for timers andon_exit. Those copies live inside boxed Rust closures no scanner can reach or rewrite, so an evacuating collection leaves them stale. Scanning the tables cannot fix this — the real fix is re-reading a scanned slot at invoke time, the indirectionperry-ui-macosgets from its handle-keyed maps, which is a ~15-site refactor. This is a pre-existing property of the PR's design rather than something introduced here, and it is documented at both scanners. Worth a follow-up issue.Not verified
WinUI runtime behaviour was not tested — it cannot be built or run from macOS, and a real cross-compile fails on a C build script (
perry-audio-miniaudioneeds Windows headers). The scanner and type definitions were validated by compiling the real definitions against the realperry_ffi::GcRootVisitorin a scratch crate, which proves match exhaustiveness and signatures but says nothing about the backend working. Someone on Windows should confirm that. The three relocatedwinui_asset_teststype-check but were not executed.Summary by CodeRabbit
New Features
Bug Fixes
Documentation