…ONT drop, feedback contract PVs, exerciser cleanup Six remaining findings from /code-review of 70dbea3, batched as one focused commit. ios_vortex.py (#3 + #10): * Kill `_refresh_after_store` entirely. ROI-bound putters now inline- await `_refresh_roi_explicit` using the new `value` arg directly — no fire-and-forget create_task (which had a task-handle discard / GC race), no 10ms sleep heuristic (which raced caproto's value- store on a loaded loop). Split into `_make_lo_putter` / `_make_hi_putter` so each can pass the new value explicitly. * Add `lower_ctrl_limit=0` / `upper_ctrl_limit=N_CHANNELS-1` on the ROI bound pvproperties. caproto now rejects out-of-range writes at the CA protocol layer with `CannotExceedLimits`. Operator who typos R{N}LO=-100 gets a clean error instead of silently getting a sum over [0..HI] (per no-silent-fallbacks). * HI<LO case stays a silent swap (acceptable transient during ROI reconfig) but logs a warning so the IOC's second-guessing is at least visible. ios_feedback.py (#7): * Add the 5 PVs `ios_devs.FeedbackLoop` declares but the IOC didn't serve: PID.DRVH, PID.DRVL, PID.DT, PID.MDT, PID.SCAN. Inert here (no calc, no scan-mode behavior — the loop uses constants for high/low limits and a fixed TICK_S) but their CA presence keeps the ophyd-side device-class walk happy when the resolver's `add_prefix=""` bug is fixed and direct-control starts walking m1b1.fbl.* for real. ios_scaler.py (#8 + #9): * Remove the `.CONT` (continuous-vs-one-shot) pvproperty. The IOC implements one-shot only; declaring CONT silently accepted writes while doing nothing. Per no-silent-fallbacks: better to fail loud ('no such PV') than silently ignore. * Switch S1..S4 from int (caproto LONG) to float (DOUBLE). At CLOCK_FREQ_HZ=1e7 the int32 path overflows when TP > 2^31 / 1e7 ≈ 215 s — realistic for trace-fluorescence scans. DOUBLE handles ~2^53 cleanly, which covers any realistic TP. configuration_service_ios.sh (#11 + #12 + #15-adjacent): * Wire scaler TP override: add `sclr_time` (top-level happi entry for .TP) to the resolve batch, register, caput `PHASE3_TP=0.5` before triggering CNT, sleep TP+0.5s, verify both T~=TP and S1~=1e7×TP. Test now exercises the TP-honoring path that was previously dead-code'd via the `PHASE3_TP=0.5` unused constant. * Switch scaler S1 check from bash integer compare (`-lt`) to awk numeric compare, since S1 is now float. Same change applied to the vortex R2 sum check for consistency. * Reset feedback SP to 0 BEFORE disabling at end-of-test. Without this, a re-run against the same pod starts with CVAL=5/SP=5, error=0, and the convergence test trivially passes regardless of P-loop correctness. Now re-runs see error=5 and actually exercise the math. Note: direct-control's fire-and-forget batch caput still masks the CA-layer rejection in HTTP responses (existing tech-debt). The IOC does enforce the new ctrl_limits — verified manually by reading the ios_vortex container logs (CannotExceedLimits raised, readback stays at prior value). HTTP success=true is misleading but the actual behavior is correct. All 38 Phase 1+2+3 exerciser checks still pass; new TP-override and feedback-reset paths verified end-to-end.

Uh oh!

Use AsyncAPI for websocket documentation? #3

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions