[marler8997]

Since the zig 0.17 upgrade, dbe stops rendering when the rate is turned up to 1000 FPS with double buffering disabled.

Cause: dbe waits for socket data with io.operateTimeout. std 0.17's posix batch wait converts the time left to milliseconds with @divTrunc before calling poll, and returns error.Timeout when poll times out (its comments say spurious timeouts are OK). So a timeout always comes back up to 1ms early: measured, a 1ms timeout returns after 0–11µs, 2ms after ~1ms, 15ms after ~14ms. dbe rendered on every timeout, so at the 1ms frame setting it rendered ~56,000 frames/s and flooded the X server. 0.15's poll(timeout_ms) never woke early.

Fix:

• after a timeout, re-check whether it's time to render (continue) instead of rendering unconditionally
• round the timeout up (add just under 1ms) so the wait doesn't spin with poll(0) until the frame is due
• compute animation progress from nanoseconds instead of truncated milliseconds, which slowed the animation at high frame rates (pre-existing)

Verification

Instrumented build on Xvfb with double buffering disabled, logging frames per second and animation progress (should advance 0.5/s):

frame setting	before	after
1ms (1000 FPS)	~56,000–62,000 frames/s	938 frames/s, ~1% CPU
15ms	—	66 frames/s

With only the re-check (no rounding up) it was 999 frames/s but used 100% of a core spinning. Animation progress now advances 0.50/s at both settings. zig build install install-xauth test-non-interactive passes; x86_64 windows compiles.

https://ampcode.com/threads/T-01a14b88-aff8-717e-bb9f-364137832240