Repository navigation
Acquisition and sequencing: fine-acquire settle and budget, offset classes, auto-ontarget, pick at readout, NEXP>1 stop, offset wait - #503
Open
cfremling wants to merge 10 commits into
Conversation
While ACAM guides, a fine-acquire correction only re-points acamd's goal; the guide loop applies it at its next executed correction. The settle was timed from the send, so with short slicecam exposures the next sample median could be built from pre-move frames and the same correction applied twice (UT 2026-08-06, BD+28d4211: exposed 2.6 arcsec off slit). acamd tracks the request: offsetgoal marks the goal change pending; the guide loop clears it when it executes a correction computed from the changed goal (the pending state is captured before the goal is read, so a correction already in flight can never clear it). The flag is published; do_fineacquire waits for pending false on a status newer than its send before settling. A stop interrupts the wait; timeout FINE_ACQUIRE_MOVE_TIMEOUT (default 60 s, 0 disables) proceeds on the old timer. Also fixes the offset_acam_goal log line, which printed degrees labelled arcsec on the guiding path. Co-Authored-By: Claude Fable 5 <[email protected]>
After a correction has been applied, one frame already inside the goal is accepted without waiting for the full sample set. A frame can only accept, never reject: larger readings fall through to the existing median logic unchanged. SETTLE_FRAMES config default 2 -> 0; the wall-clock settle after the confirmed goal change covers the move.
A deliberate goal offset (the star-to-target offset after fine acquisition, put-on-slit, the GUI offset) armed a one-shot allowance that any sent correction cleared. A guide correction whose blocking pt_offset was in flight when the offset arrived cleared it as it completed, and the next frame's correction to the new goal was refused at the guiding cap, so the exposure was taken on the offset star. The cap selection also required acquire_mode == TARGET_GUIDE, which is a race of its own. Goal changes now carry a generation, and each correction is classified by the goal it was computed from: guide, fine-acquire, or deliberate offset. Only a deliberate-offset correction may use the larger ACQUIRE_TCS_MAX_PUTONSLIT_OFFSET cap; once sent it has made the change and clears goalshift_pending, unless a newer change arrived meanwhile. A correction computed from an older goal can neither use nor spend a newer goal's allowance. acamd.cfg.in: ACQUIRE_OFFSET_THRESHOLD=1.0, the value the deployed configuration uses. Co-Authored-By: Claude Opus 5.5 <[email protected]>
A target that kept landing just outside the goal was corrected again and again until the sequencer's acquisition timeout. Two slicecamd.cfg keys: FINE_ACQUIRE_MAX_ADJUSTMENTS=<n> (default 3, 0 = previous behaviour): at most <n> corrections are commanded in one run, and the <n>-th decides. It is sent only if the offset it removes is within FINE_ACQUIRE_RELAXED_FACTOR x goal, and the run converges once that move has executed and settled; otherwise the run fails at once without sending it. A correction acamd does not confirm within FINE_ACQUIRE_MOVE_TIMEOUT also fails a budgeted run. Nothing is sampled or evaluated once no correction can follow. FINE_ACQUIRE_RELAXED_FACTOR=<f> (default 2.0, at least 1). Also: a single frame inside the goal after a correction converges the run; FINE_ACQUIRE_SETTLE_SEC=1 (was 3.0) now that the settle starts when the move has executed; every lock writes the same "fine acquisition converged" line, with how it locked appended. Co-Authored-By: Claude Opus 5.5 <[email protected]>
?ONTARGET was invisible through tcsd because send_command translates
every non-whitelisted reply as a status code: the flag IS the reply, so
0 collided with TCS_SUCCESS ('success') and 1 fell through to
'tcs_undefined' ERROR. Whitelist it for verbatim passthrough.
sequencerd: config TCS_AUTO_ONTARGET (default no) spawns a watcher for
exactly the lifetime of move_to_target's TCSOP wait. It calls
ontarget() -- the operator's own path -- when, on two consecutive 1 Hz
reads: ?ONTARGET==1 AND motion==tracking AND |telescope-target| <
TCS_AUTO_ONTARGET_SEP arcsec. All three are required: shadow data
(UT 2026-08-25, 64 slews) shows the flag is position-agnostic, the
separation cannot see sub-threshold hops, and motion says nothing
about settle. Operator ack trailed the flag by median +9.7 s / p90
+18 s and never once preceded it. The operator's press is unchanged;
whichever arrives first wins.
When the last sub-exposure of a science target starts reading out, the target is completed and the next one is picked through the normal path (get_next, activate, publish, slew, acquire) while camerad reads out. A readout gate holds every camerad command for the new target (camera settings, expose) until camerad reports READY. There is no early pick in single-step or single-target mode, after a cancel, with a thread error set, or when the current row is a calibration. Readouts are counted from camerad's PIXELCOUNT messages, so the count cannot run ahead of the shutter; a PIXELCOUNT from the tail of the previous readout cannot end the next target's EXPOSE wait. camera_set runs in parallel with the slew and acquisition behind the readout gate and is joined just before the exposure; a target is never exposed unless its own camera settings were applied. The flexure compensator is set just before each exposure, re-applied during the readout between sub-exposures, and before a REPEAT exposes. Co-Authored-By: Claude Opus 5.5 <[email protected]>
… commanded angle TCS_AUTO_ONTARGET defaults to yes. The automatic ontarget now also requires |TCS cass angle - RINGGO angle| < TCS_AUTO_ONTARGET_CASS (0.05 deg) on the same two consecutive reads: ACAM's goal is computed from the cass angle at the acquire, about 8.5" per degree of ring still to turn. When everything but the ring has held for 20 s, one warning tells the operator to press ontarget once the ring is in place. Co-Authored-By: Claude Opus 5.5 <[email protected]>
stop: the sequencer accepts it for the whole series of the current target (the EXPOSE wait covers only the first sub-exposure; camerad not READY covers the rest) and no longer clears a cancel. camerad ends the series after the frame in progress, cut short if it is integrating with 5 s or more left, instead of starting the next sub-exposure. abort no longer starts the next sub-exposure either. A cancelled exposure's remaining PIXELCOUNT messages cannot set the READOUT wait later. camerad waits for the last frame's FITS write before acting on the next command after READY. Co-Authored-By: Claude Opus 5.5 <[email protected]>
A target offset sent while ACAM guides only changes acamd's goal; the guide loop's next correction makes the move. The sequencer exposed a fixed OFFSET_SETTLE_SEC (3 s) after sending it, which could be before the telescope had moved. It now waits until acamd's status shows the goal change raised after the offset was sent and cleared again (GOALSHIFT_PENDING), bounded at 30 s, after which it exposes anyway with a warning; a cancel ends the wait. OFFSET_SETTLE_SEC is now an extra wait after the move and defaults to 0. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
This was referenced Oct 29, 2026
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8c251201c2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+655
to
+660
| if ( this->last_acam_pubtime.load(std::memory_order_acquire) <= send_time ) return false; | ||
| if ( this->acam_goalshift_pending.load(std::memory_order_acquire) ) { | ||
| seen_pending = true; | ||
| return false; | ||
| } | ||
| return seen_pending; |
There was a problem hiding this comment.
If acamd's pending=true and subsequent pending=false reports are both processed before this waiter evaluates its predicate, seen_pending remains false even though the correction completed successfully. This can happen before the synchronous offset_acam_goal() call returns or while the waiting thread is not scheduled: handletopic_acamd() retains only the latest pending value, and condition-variable notifications do not queue transitions. The acquisition then waits the full move timeout and, with the configured correction budget, incorrectly reports failure. Record the raised and cleared publish times in the telemetry handler, as the sequencer's new offset wait does, so completion does not depend on observing the intermediate state in this thread.
Useful? React with 👍 / 👎.
Collaborator
Author
There was a problem hiding this comment.
Fixed in e3414e7: the acamd telemetry handler records the publish times of the latest status with GOALSHIFT_PENDING set and of the latest with it clear, and the wait completes once a set status published after the send is followed by a clear one, the same rule as the sequencer's wait for the target offset. It no longer depends on observing the intermediate state.
The wait for acamd to apply a fine-acquire correction completed only if it saw GOALSHIFT_PENDING set and then clear. Condition-variable notifications do not queue, so if both statuses were handled before the waiting thread re-checked, it waited out FINE_ACQUIRE_MOVE_TIMEOUT and, under a correction budget, failed the run. The acamd telemetry handler now records the publish times of the latest set and clear statuses, and the wait completes once a set status published after the send is followed by a clear one, as the sequencer's wait for the target offset does. Co-Authored-By: Claude Opus 5.5 <[email protected]>
6 tasks
This branch has not been deployed
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
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.
Acquisition and sequencing changes that depend on each other, so they are one PR: the sequencer's wait for the target offset needs acamd's goal-change reporting and runs in the pick-at-readout exposure path, and the later sequencer changes build on the earlier ones. One commit per change, for review:
GOALSHIFT_PENDINGin common/message_keys.h). The settle after a fine-acquire correction starts when the move has executed, not after a fixed timer.ACQUIRE_TCS_MAX_PUTONSLIT_OFFSET, so an in-flight guide correction can no longer consume that allowance and leave the exposure on the offset star (2026-08-26, 10-06, 10-19).FINE_ACQUIRE_MAX_ADJUSTMENTS,FINE_ACQUIRE_RELAXED_FACTOR).?ONTARGETthrough; sequencerd watches?ONTARGET, tracking and the separation). The operator's ontarget still works and wins when first.Config templates:
ACQUIRE_OFFSET_THRESHOLD=1.0;FINE_ACQUIRE_SETTLE_SEC=1,FINE_ACQUIRE_SETTLE_FRAMES=0,FINE_ACQUIRE_MOVE_TIMEOUT=60.0,FINE_ACQUIRE_MAX_ADJUSTMENTS=3,FINE_ACQUIRE_RELAXED_FACTOR=2.0;TCS_AUTO_ONTARGET=yes,TCS_AUTO_ONTARGET_SEP=5.0,TCS_AUTO_ONTARGET_CASS=0.05,OFFSET_SETTLE_SEC=0.To deploy: rebuild, then restart acamd, slicecamd, sequencerd, tcsd and camerad.
Tested on sky 2026-10-28/29 UT using branch cf/summit-20261020, 18 targets in 10:27–12:34 UT:
Jira: PAL-166, PAL-171, PAL-182, PAL-144, PAL-181, PAL-152, PAL-142.
Replaces #479, #483, #497, #498, #499, #500, #501 and #502. Independent of #485, #493, #494, #495, #496 and #504: each of those applies to main on its own, before or after this one.
🤖 Generated with Claude Code