Repository navigation
clock: parity with cvc::world_clock's shared-clock API (lockstep, batch, followers) - #116
Merged
Merged
Conversation
Brings grl_snam.clock.WorldClock to parity with libcvc's one-clock-for-every-simulation work (transfix/libcvc#629): - lockstep mode; - advance_ticks (batch); - configure; - StepResult.first_tick; - realized_scale; - cap_scales_with_rate. Adds ticks_per_step, steps_due and ClockFollower (take/due_at/alpha), so a Python simulation follows a shared clock at its own whole-quantum period, as a C++ one does. All additions keep the existing API and behaviour. tests/test_clock.py mirrors the C++ cases (pacing invariance included) and checks tick-for-tick parity with pycvc.world_clock where pycvc is installed. 32 tests pass locally (Windows, cvcpkg python311 + torch-cp311, pycvc built from libcvc#629), along with the existing WorldClock tests in test_fog_demo.py.
transfix
added a commit
that referenced
this pull request
Oct 15, 2026
…a WorldClock The Python side of "every simulation on one world clock" (libcvc docs/roadmap/WORLD-CLOCK-SYNC-ROADMAP.md, step C5). It builds on the clock parity work in #116. clock.WorldClock: - a Condition notified on every tick change; - wait_for_tick_change(tick, timeout), so a follower thread sleeps until the driver moves the clock instead of polling (as cvc::world_clock does). sim_thread.SimThread: - SimThread(world, clock=clock) FOLLOWS a shared clock: one step every world.dt / fixed_dt quanta, run as the driver advances it, as cvc::nav::sim_thread does. Pause skips this sim's due steps; the clock's own pause pauses everything. - SimThread(world, hz=60) keeps its old behaviour on a private clock (quantum world.dt, speed hz * dt) that the thread drives. Pause, SetRate and Nudge act on that clock, so world time and the steps run always agree. - wait_caught_up(timeout), so a capture shows exactly each frame's steps. nav_native.NativeSimThread(world, clock=app.world_clock()) is the C++ worker following a shared pycvc clock, plus wait_caught_up. It needs the pycvc functions from transfix/libcvc#630; HAS_SIM_THREAD_CLOCK gates it. demos._common.SimPacer keeps its API but now drives a WorldClock at the suite quantum (0.01 s), with a ClockFollower for the navigator. It was a private accumulator; its .clock is exposed for a host. A stall is capped at 8 navigator steps, and the clock drops and counts the rest. Tests, run against a local libcvc build with #630: - test_clock: the tick wait wakes on advance, times out without one, and counts a rewind as a change; - test_swarm: SimThread on a shared lockstep clock runs tick // k steps on every frame, and none while paused; - test_sim_thread: NativeSimThread on app.world_clock(), the same check on the C++ worker; - test_sim_pacer: frame-rate independence, speed and the stall drop still hold. All 19 tests in the clock, sim_thread and SimThread selection pass, and so do test_sim_pacer and test_clock (17). ruff 0.16 and black 26.5.1 are clean.
transfix
added a commit
that referenced
this pull request
Oct 15, 2026
…a WorldClock (#117) The Python side of "every simulation on one world clock" (libcvc docs/roadmap/WORLD-CLOCK-SYNC-ROADMAP.md, step C5). It builds on the clock parity work in #116. clock.WorldClock: - a Condition notified on every tick change; - wait_for_tick_change(tick, timeout), so a follower thread sleeps until the driver moves the clock instead of polling (as cvc::world_clock does). sim_thread.SimThread: - SimThread(world, clock=clock) FOLLOWS a shared clock: one step every world.dt / fixed_dt quanta, run as the driver advances it, as cvc::nav::sim_thread does. Pause skips this sim's due steps; the clock's own pause pauses everything. - SimThread(world, hz=60) keeps its old behaviour on a private clock (quantum world.dt, speed hz * dt) that the thread drives. Pause, SetRate and Nudge act on that clock, so world time and the steps run always agree. - wait_caught_up(timeout), so a capture shows exactly each frame's steps. nav_native.NativeSimThread(world, clock=app.world_clock()) is the C++ worker following a shared pycvc clock, plus wait_caught_up. It needs the pycvc functions from transfix/libcvc#630; HAS_SIM_THREAD_CLOCK gates it. demos._common.SimPacer keeps its API but now drives a WorldClock at the suite quantum (0.01 s), with a ClockFollower for the navigator. It was a private accumulator; its .clock is exposed for a host. A stall is capped at 8 navigator steps, and the clock drops and counts the rest. Tests, run against a local libcvc build with #630: - test_clock: the tick wait wakes on advance, times out without one, and counts a rewind as a change; - test_swarm: SimThread on a shared lockstep clock runs tick // k steps on every frame, and none while paused; - test_sim_thread: NativeSimThread on app.world_clock(), the same check on the C++ worker; - test_sim_pacer: frame-rate independence, speed and the stall drop still hold. All 19 tests in the clock, sim_thread and SimThread selection pass, and so do test_sim_pacer and test_clock (17). ruff 0.16 and black 26.5.1 are clean.
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.
What
grl_snam.clock.WorldClockreaches parity with the C++cvc::world_clockand its shared-clock API in transfix/libcvc#629. One clock per process, one driver, every simulation a follower at its own whole-quantum period.New in
WorldClock:lockstepmode: exactlylockstep_stepsquanta peradvance(), whatever the wall says;advance_ticks(n)for batch runs;configure(...): a host picks a quantum every simulation divides before any of them starts; restarts at tick 0 and keeps the mode;StepResult.first_tick;realized_scale(): the speed actually achieved;cap_scales_with_rate.New helpers:
ticks_per_step(dt, fixed_dt)refuses a step that is not a whole number of quanta;steps_due;ClockFollowerwithtake,due_at,alphaandfor_step.Everything is additive: the existing constructor,
mode,seek_timeand the replay behaviour the fog, squad and finale recorders use are unchanged.This is step C5 of the world clock sync plan (libcvc
docs/roadmap/WORLD-CLOCK-SYNC-ROADMAP.md). The demos andSimThreadmove onto the clock in later PRs.Tests
tests/test_clock.pyhas 12 cases mirroring libcvc'sworld_clock_test:first_tick,configure, realized speed and the rate-scaled cap;ticks_per_stepand a brute-forcesteps_duesweep;pycvc.world_clock, tick for tick, where pycvc is installed. It skips otherwise.Run locally on Windows with cvcpkg python311 and torch-cp311, against pycvc built from libcvc#629: 32 passed, including the existing
WorldClocktests intest_fog_demo.py.