Acknowledge _NET_WM_SYNC_REQUEST only after the resized frame on X11

[treeform]

Summary

On X11 the _NET_WM_SYNC_REQUEST counter was set at the top of every pollEvents, before that poll's events were handled. If the window manager's sync request arrived at the end of one poll and its ConfigureNotify in the next, windy acknowledged the frame while the window was still drawn at the old size. KWin then composited that frame, which is the one-frame black stripe on the right and bottom edge reported in #156.

This tracks the handshake per window:

• sync request → SyncRequested, remember the counter value
• ConfigureNotify → SyncConfigured
• next poll (after a frame at the new size went by) → XSyncSetCounter, back to SyncIdle

The event loop also waits up to 10 ms for the ConfigureNotify that follows a sync request, so the frame drawn after the poll is already the new size. A request that never gets a configure is never acknowledged, which is what the EWMH protocol expects.

This follows the intent of #156 by @levovix0. Left out on purpose: the temporary vsync disable after a resize and the cached size getter, which are unrelated to the sync handshake (the cache also made size= wait its full timeout, since blockUntil never polls events).

Test

tests/test_x11.nim plays the window manager side under a bare Xvfb: it sends the _NET_WM_SYNC_REQUEST client message, resizes with XResizeWindow, and reads the counter back with XSyncQueryCounter. It checks:

• a resize without a request never touches the counter
• request and configure in one poll: no ack until the next poll
• request in one poll, configure in a later poll: no ack until the configure was handled and a frame went by, and the waiting poll gives up quickly
• the full 64-bit value round-trips and a newer request wins
• an idle window's counter is left alone

Against the old unconditional ack the test fails at the split case, which is exactly the reported bug. CI runs it on Ubuntu under xvfb-run with Mesa llvmpipe.

Validation

• nim r tests/test_x11.nim under Xvfb (Docker, linux/amd64, Nim 2.2.4): passes
• same test against the old ack: fails at the split-poll case
• nim c tests/test.nim, examples/basic.nim, examples/callbacks.nim, examples/property_changes.nim on Linux: compile

Not verified on a real KWin session, I do not have one at hand. @levovix0 if you can try this on your setup that would confirm the visual fix.

🤖 Generated with Claude Code