Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Entries for 2.6.0 to 2.14.0 were generated from commit subjects. Older entries were converted from the former CHANGELOG.rst.

[2.14.0] - 2026-10-04

• Slim base dependencies for pyobs-gui, move the rest to the full extra

• BaseVideo: opt-in CORS for /video.raw (cors_origins)

[2.13.5] - 2026-10-04

• Add VirtualFileSystem.set_roots() to replace roots at runtime (#939)

• docs: add pyobs-auth to the project tiers doc

[2.13.4] - 2026-10-04

• Resolve stretch cuts from the input dtype, carry SRCDTYPE in raw frame meta

• Log missing science exptimes in darkbias at info level

[2.13.3] - 2026-09-29

• fix: AstroplanScheduler hanging forever when scheduled blocks can’t be pickled

[2.13.2] - 2026-09-29

• fix: end video streams quietly on any client disconnect, show IQ failure reason in subscribe warnings

• docs: refresh fleet open items (2026-09-29)

[2.13.1] - 2026-09-29

• Export Frame, FrameRecord, StartSource from pyobs.modules.camera

[2.13.0] - 2026-09-29

• specs: BaseVideo redesign merged (PR #926), drop #924/#925 from fleet-open-items

• BaseVideo: frame buffer, timestamped frames, new grab path, stream guiding (#926)

• specs: BaseVideo redesign design docs and plan (#924, #925)

• steering: remove closed pyobs-web-client #40-#47 from fleet-open-items

• steering: record the IResettable pyobs-core pin bug and its fleet-wide fix

• steering: remove IResettable driver-override issues from fleet-open-items

• specs: mark all eight IResettable sibling-repo issues as implemented

[2.12.0] - 2026-09-27

• CHANGELOG: Image.writeto() NAXIS fix (#922)

• Image.writeto(): let astropy regenerate NAXIS cards

• Add stack combine processors (MeanStack, MedianStack, SumStack, SigmaClipStack) (#921)

• Remove finished items from fleet-open-items (open items only)

• Update fleet-open-items snapshot for 2026-09-27

• Fix stale status line on data-pipeline-stack-reset plan doc

• Link phase 8 driver issues in iresettable.md and the plan doc

[2.11.0] - 2026-09-27

• Add IResettable, IDataPipeline, IDataStack to BaseCamera/BaseVideo (#919)

• Fall back to selected pipeline when BaseVideo image has no pipeline meta

• Add design docs and plan for IResettable, IDataPipeline and IDataStack

[2.10.7] - 2026-09-23

• Harden PushNotifier against FCM payload and outage failure modes

[2.10.6] - 2026-09-20

• Send valid LCO config states to the portal

[2.10.5] - 2026-09-19

• Fix ResolvableErrorLogger’s min_interval comparison and bookkeeping

• Log a failing archive query as a throttled warning, not a traceback

• Add ArchiveError and raise it for every archive failure

[2.10.4] - 2026-09-19

• fix: let class-passed child objects inherit vfs/timezone/observer

[2.10.3] - 2026-09-18

• ImageWatcher: default error_after to 1800s instead of requiring it

• docs: mark push-notification preferences client half done

[2.10.2] - 2026-09-18

• feat: rename IPushNotifications methods, add get_push_preferences

[2.10.1] - 2026-09-18

• feat: per-user push notification type preferences in PushNotifier

• docs: re-check fleet open items (2026-09-18)

[2.10.0] - 2026-09-17

• feat: extract DataSequenceMixin, add IDataSequence support to BaseSpectrograph

• docs: mark imagewatcher-retry-hardening landed, add pyobs-web-client#57

[2.9.2] - 2026-09-17

• fix: harden ImageWatcher retry behavior on failed file processing

• docs: mark push-notification-module implemented, update fleet-open-items

[2.9.1] - 2026-09-15

• fix: run firebase_admin.delete_app() in a thread in PushNotifier.close()

• docs: qfitswidget toolbar-overflow fix released (qfitswidget 1.1.3, pyobs-gui 2.4.2)

[2.9.0] - 2026-09-14

• docs: sync fleet-open-items.md changelog

• feat: implement PushNotifier module (module ERROR + log ERROR/CRITICAL to FCM)

• Drop two stale pyobs-web-client entries from fleet open items

• docs: fold ERROR/CRITICAL log alerts into PushNotifier v1, file #902

• docs: pyobs-gui scroll-fallback plan implemented, drop from open plans

• docs: track two new pyobs-gui plans; update telescopewidget-layout status

• docs: drop web-admin#95 – already fixed, released in v2.3.4, closed

• docs: close #859 as not worth it, drop from fleet tracker

• docs: reject gui-standalone-binary – real build came out several GB

• docs: close #899 – released in v2.8.9

[2.8.9] - 2026-09-14

• docs: record #899 fix merged to develop, drop from fleet tracker

• Route ACL-denial faults through the normal call_id/fault path (#899)

• Track pyobs-brot#71 (settle-timeout root cause) in fleet tracker

• Record BROT settle-loop staleness/resend plan; close pyobs-brot#61 and #68 in fleet tracker

• chore: untrack .claude/, keep it local-only

[2.8.8] - 2026-09-14

• docs: close #898, #846, #884; fleet maintenance pass

• feat: publish struct field schemas in disco#info’s block

• Surface call_id in ShellCommand’s remote-error formatting

• docs: close pyobs-web-client #48/#49, cross-file call_id issue (#54/pyobs-gui#167)

• docs: record pyobs-pipeline’s Keycloak login cutover (#17)

• docs: move iag50 2.x migration plan to pyobs-iag50’s own specs

• docs: close #896, mark general per-exptime-darks-on-lco-sites plan accepted

• docs: split #896 design into general pyobs-core plan + site-specific

• docs: cross-reference iag50cm per-exptime-darks plan for #896

• docs: drop closed #891 from fleet snapshot

• docs: drop landed #895 from fleet snapshot, mark plan merged

• fix: reschedule when mastermind skips a late start window (#897)

• removed graphify

• docs: refresh fleet open-items snapshot (2026-09-09)

[2.8.7] - 2026-09-09

• feat: return per-run success/failure counts from Reduction.call

• fix: pass a Quantity dark_exposure to ccdproc even when dark_scale=False

• Sketch a PushNotifier module design for issue #884’s deferred push notifications

[2.8.6] - 2026-09-08

• fix: publish DummyCamera capabilities after connecting, not before

[2.8.5] - 2026-09-08

• fix: re-raise weather station errors so poll loop backs off

• specs/steering: update fleet-open-items snapshot to 2026-09-08

[2.8.4] - 2026-09-08

• fix: tolerate unknown fields in LCO portal API responses

[2.8.3] - 2026-09-08

• fix: publish caps hash update when XMPP capabilities are set

[2.8.2] - 2026-09-08

• fix: register MoveRaDecEvent/MoveAltAzEvent in BaseTelescope.open()

• docs: fix stale pyobs-web-client plan links in fleet-open-items

[2.8.1] - 2026-09-07

• xmpp: add loopback ws listener on 5282, off the public path

• Mark mobile-app RN design superseded by pyobs-web-client Capacitor plan

• Record XMPP-core spike outcome: stanza.js chosen (issue #884)

• Add mobile-app design doc + ADRs 0016-0018 (mobile client, shared TS core)

• Drop pyobs-weather#6 from fleet open items: merged to develop

• Update fleet steering docs: close blocking-SDK punch list, refresh open items

[2.8.0] - 2026-09-04

• Address review: min_length on module_name, structural-failure test, policy note

• Add camera/filter-wheel model fields, fix nullable module_name bug

[2.7.2] - 2026-09-04

• feat: add description to ConfigFieldSchema

[2.7.1] - 2026-09-04

• feat: propagate AccessLevel through ConfigFieldSchema

[2.7.0] - 2026-09-04

• feat: add AccessLevel enum for GUI basic/expert config field tiers

• Add RoofCapability for plain open/close roof timing

• Add mean-distance dome rotate time to first-task duration estimates

• docs: close out pyobs-core#871 in fleet open-items snapshot

• fix: address review nits on Comm.unregister_event() cancellation (#871)

• fix: Comm.unregister_event() cancels already-scheduled handler tasks (#871)

• docs: close out pyobs-core#872 in fleet open-items snapshot

• docs: close out #872 fits-header-audit-followthrough plan as implemented

[2.6.1] - 2026-09-03

• Fix FilterHeaderMixin/FocuserHeaderMixin crash when chaining to an abstract get_fits_header_before

• docs: record 2.6.0 release + sibling pin-bump status in fits-header-audit plan

• docs: plan for Comm.unregister_event() task-cancellation fix (#871)

[2.6.0] - 2026-09-03

• Add FITS header fields for camera flip/gain, telescope tracking mode, guiding offsets, and weather verdict; add FilterHeaderMixin/FocuserHeaderMixin

• docs: drop closed pyobs-core#739 from fleet open-items snapshot

• docs: drop closed pyobs-core#863 from fleet open-items snapshot

• docs: drop pyobs-gui main-vs-sidebar-widgets + video-widget-split from fleet open-items

• docs: drop closed pyobs-web-admin#89 from fleet open-items snapshot

[2.5.0] - 2026-09-03

• New IResettable interface (reset()/full_reset()): a module can reset its per-acquisition settings (exposure time, image type, binning, window, data pipeline) or, with full_reset(), everything including hardware state (cooling). Implemented by BaseCamera, BaseVideo and BaseSpectrograph. Module.startup() now calls full_reset() automatically for any module implementing it, and the robotic scripts (Imaging, DarkBiasScript, the LCO default script, FlatField) call reset() before configuring a camera for a new task. See specs/design/iresettable.md.

• New IDataPipeline interface (set_pipeline()): a camera can run a named, YAML-configured pipeline on every grab before storing it – the pipeline result replaces the raw data, and every stored file carries a PIPELINE FITS header. Implemented by BaseCamera and BaseVideo via the new DataPipelineMixin, configured with new pipelines/ default_pipeline init parameters. See specs/design/idatapipeline.md.

• New IDataStack interface (grab_stack(count)): grabs count consecutive frames into one 3D cube product (or, with a pipeline selected, its combined result), blocking until the product is stored. Implemented by BaseCamera and BaseVideo, memory-capped via new max_stack_bytes/stack_frame_overhead/stack_timeout_margin init parameters. See specs/design/idatastack.md.

• New pyobs.images.processors.stack package: pipeline steps that collapse a grab_stack() cube into one float32 frame, one class per method (MeanStack, MedianStack, SumStack, SigmaClipStack) on a common CombineStack base. Non-stack images pass through unchanged, NaNs are ignored, the cube is processed in row blocks in a worker thread, and an uncertainty and mask extension can be enabled. See specs/design/combinestack.md.

• Image: fixed NAXIS1/NAXIS2 for 3D data (previously swapped), is_color now excludes stack cubes (CTYPE3 == 'FRAME'), and a new optional frames table (per-frame FRAME/DATE-OBS/EXPTIME, written/read as a FRAMES FITS extension). Also fixes BaseCamera.apply_meridian_flip() flipping the wrong axes on 3D data. See specs/design/idatastack.md.

• Image.writeto(): no longer fails on misplaced or stale NAXISn cards (e.g. a header passed to Image() without NAXIS cards, or NAXIS3 left over after collapsing a cube via the data setter). They are now regenerated from the data on write (#922).

• Every module implementing IFitsHeaderBefore/IFitsHeaderAfter now writes one HIERARCH <MODULE> VERSION <PACKAGE> card per loaded pyobs-* package into FITS headers, reusing loaded_pyobs_packages() (added for #759). Closes #739. See specs/design/package_versions_fits_header.md.

• Fixes Comm.unregister_event() not cancelling a handler dispatch that _send_event_to_module() had already scheduled via asyncio.create_task() before the unregister call – the task ran later regardless, against whatever the handler was bound to (observed in pyobs-gui as a libshiboken: Internal C++ object already deleted error on client disconnect). Closes #871.

[2.4.1] - 2026-09-03

• Fixes Time.night_obs() flipping to the next night around dawn: it anchored the night boundary at “nearest sunset + 12h”, which lands close to sunrise near the equinox – exactly when morning calibration scripts run (e.g. DarkBiasScript queried the following night, with no science frames yet, instead of the one that just ended). Anchored at local solar noon instead, matching the archive’s own night-labeling convention and not drifting with day length. Also affects Mastermind’s night resolution and the DAY-OBS FITS header, which share the same call.

[2.4.0] - 2026-09-03

• Breaking: TaskScheduler.schedule() (both OnDemandScheduler and AstroplanScheduler) gained a new instrument_capabilities: InstrumentCapabilities | None = None parameter, part of pyobs-core#864/#865’s plumbing for feeding pyobs-portal instrument capability data into script duration estimates (see specs/plans/2026-09-01-instrument-capability-duration-estimates.md). Any out-of-tree TaskScheduler subclass overriding schedule(tasks, projects, start, end) without the new keyword argument will raise TypeError once called by pyobs/modules/robotic/scheduler.py’s _schedule_worker, which now always passes it. Add the parameter to your override’s signature (a default of None is enough if you don’t use it) to stay compatible.

• Added InstrumentCapabilities (pyobs/robotic/instruments.py), mirroring pyobs-portal’s instrument JSON with module-name-keyed camera/telescope/dome/filter-wheel capability lookups, and threaded it through TaskData/TaskArchive/Task.estimate_duration(). PortalTaskArchive now polls and caches real capability data from GET /api/instruments/ (re-fetched only when pyobs-portal’s last_instrument_update marker moves in either direction; a fetch/parse failure keeps serving the last-good data rather than blocking the tasks/projects poll). ImagingScript, PointingScript, DarkBiasScript, and AutoFocusScript now use real portal-declared readout/filter-change/slew numbers where available, falling back to the previous flat fudge constants otherwise. See specs/plans/2026-09-01-instrument-capability-duration-estimates.md.

• Fixes the Scheduler’s get_schedule()/get_current_observation() permanently returning an empty schedule: forcing auto_update=False on the observation archive also silenced the poller that’s the only channel by which the scheduler process learns about observation-state changes written by other processes (e.g. Mastermind marking an observation IN_PROGRESS/COMPLETED). Added announce_updates to PortalObservationArchive instead, so the poller can stay on without double-logging its own writes.

• VFS gained rmdir() (LocalFile/VirtualFileSystem); ImageWatcher now removes now-empty parent directories left behind when it deletes a watched file.

• ImageWatcher only attempts FITS header parsing for filenames that look like FITS (.fits/.fitz/.fits.gz/.fits.fz); other watched/copied files no longer produce spurious astropy header-parse warnings.

[2.3.0] - 2026-09-02

• ImageWatcher gained a flatten=True (default) constructor option: set False to keep a non-templated destination file’s path relative to watchpath instead of collapsing it to just the basename, so a whole nested directory tree can be relocated while preserving its structure. Also added recursive watching (inotify mode via asyncinotify’s RecursiveWatcher, poll mode, and the initial open() scan), so subdirectories are picked up too, not just the watched root. Fixed two pre-existing bugs while rewriting the poll loop: poll_interval was stored but never actually awaited between iterations (a busy-loop), and a stray debug print() was left in the new-file branch. (#860)

• Issue #855: PolymorphicBaseModel’s custom serializer (used by Task/Script/ Constraint/Merit/Target and their subclasses) now honors exclude/include/ by_alias/exclude_none/exclude_unset/exclude_defaults on model_dump()/ model_dump_json() instead of silently ignoring them, for flat top-level field specs (the only pattern any current caller uses); a nested spec now raises NotImplementedError instead of being silently ignored or partially applied.

• Issue #844: Reduction’s minimum-darks-per-group threshold is now configurable via min_darks_per_group (default 3, unchanged), matching the existing min_flats option.

• Issue #856: PortalTaskArchive._update() no longer counts a bare updated_at bump (e.g. a no-op re-save) as a real content change, which previously triggered an avoidable re-download and reschedule.

• Issue #851: DummyCamera gained an optional roof reference (mirroring the existing telescope reference) and now skips flat-field/star simulation, producing dark frames, while its DummyRoof is parked; fails closed until the roof’s state has been received.

• Issue #848: Scheduler._update_schedule() now also reschedules on a project priority change or a same-ID task content change (e.g. priority), not just on task IDs appearing or disappearing; also fixes an assignment-order bug where self._projects was overwritten before it could ever be diffed. Follow-up review fixes: preserve the portal task FK when canceling (the cancel PUT was silently rejected every time otherwise), guard against a startup race between the observation- and task-archive pollers, and stop get_current_observation() from self-healing an in-progress observation out from under a running Mastermind.

• Issue #850: AltitudeLimitError no longer inherits from MotionError, so refusing a below-horizon target no longer counts toward motion-fault escalation (3 legitimate rejections within 10 minutes used to push the module into ERROR state, requiring a manual reset_error()).

• Issue #845: Module._on_module_opened and FocusSeries._run_focus_series now also catch the plain ValueError that Comm._resolve_proxy() raises (per its documented contract) when a proxy can’t be resolved, instead of leaking it to the caller.

• Issue #847: Scheduler._update_schedule() no longer skips rescheduling on portal task removal (the observation-archive schedule-cache gate it relied on is permanently empty for PortalObservationArchive by construction). PortalObservationArchive also self-heals observations whose task the portal no longer resolves, canceling them locally and on the portal instead of logging-and-skipping forever.

• Issue #849: DummyRoof.stop_motion() now reports PARKED rather than IDLE on a closed roof, matching init()/park().

[2.2.0] - 2026-09-01

• Issue #832: Calibration/Reduction now match dark masters to a science frame’s EXPTIME instead of always scaling whatever dark master find_master happened to return (see ADR 0015-dark-master-strict-exptime-matching-reference-scale-down-only.md). An exact exptime match (within dark_exptime_tolerance, default 1%) is used unscaled; below dark_min_exptime (default 5s) with no exact match, calibration falls back to bias-only; a reference master (dark_scale_exptime, default 600s) is scaled down to shorter science exptimes but never scaled up.

  Behavior change, upgrade check required: a site that takes darks at only one exptime and previously relied on always-scale-whatever’s-nearest now fails calibration (a caught, catchable ValueError, logged and returned uncalibrated – not a crash) for any science exptime longer than its one master, or with no reference configured. Set allow_unmatched_dark_scale=True to keep today’s behavior, or take darks at more than one exptime (DarkBiasScript’s exptimes/match_science_exptimes, issue #831) so exact matches exist. Reduction’s master-calibration-frame filename pattern also gained an exposure-time component ({EXPTIME|exptime}, shared by BIAS/SKYFLAT too); existing archives keep their old-pattern masters under their original filenames.

  A reference master is also no longer scaled up (only down), and a legacy master with no EXPTIME in its header is now used unscaled instead of raising KeyError on the allow_unmatched_dark_scale fallback path.

• Issue #831: DarkBiasScript can now take one morning dark series per exposure time actually used by the night’s science frames, instead of always a single fixed exptime. Adds science_exptimes_for_night() (derives per-instrument/binning exptimes from the night’s OBJECT frames; exptimes below dark_min_exptime are dropped since calibration treats them as bias-only) and exposes EXPTIME on the archive API (FrameInfo, list_frames/ list_options, both PyobsArchive and LocalArchive) so it can be queried. Enable via DarkBiasScript’s new exptimes/match_science_exptimes options.

• Added DummyStructuredConfig (pyobs.modules), implementing IStructuredConfig with one field of every ConfigFieldSchema type (str, int, a float with a unit, bool, enum, nested object) so pyobs-gui’s new StructuredConfigWidget (gui#154) has something to manually verify against without pyobs-iagvt’s FTS, the only other IStructuredConfig consumer.

• Fixes issue #830: http_request_with_retries no longer logs a WARNING on every failed retry attempt. Retries now stay quiet for the first 60s a URL has been failing (covering a typical short pyobs-portal restart), then warn at most once per minute until the request succeeds again. PortalTaskArchive/PortalObservationArchive’s poll loops got the same treatment for their log.error calls: the first failure still logs immediately at ERROR, but repeats during the same outage are throttled to at most once per minute instead of once per ~20s poll cycle. Both reuse a new pyobs.utils.http.LogThrottle helper.

[2.1.0] - 2026-08-31

• Added IRobotic (executor) / IRoboticScheduler (planner) interfaces, wired into Mastermind and Scheduler respectively (issue #825). IRobotic pushes RoboticState (current task, next-up, cant_run_reason) on every transition; IRoboticScheduler adds get_schedule(limit), returning pending/in-progress entries only, capped server-side regardless of the requested limit. TaskStartedEvent/TaskFinishedEvent/TaskFailedEvent now also carry obsnum. See specs/design/irobotic.md.

• Added DummyMastermind/DummyScheduler (pyobs.modules.robotic), simulating a robotic schedule so pyobs-gui’s new RoboticWidget/ScheduleWidget have real state transitions to develop and test against without live hardware.

• XmppComm now pre-creates a module’s own event pubsub nodes at the moment it declares them (send-only register_event()), instead of relying on the server’s lazy auto-create-on-first-publish. Closes the window where a peer connecting before a module’s first publish could never complete a subscribe (item-not-found) and had to fall back to an indefinitely retrying background task. State nodes are explicitly out of scope (see specs/plans/2026-08-28-precreate-pubsub-nodes.md’s “Why state nodes are excluded” section).

• Fixes issue #824: _retry_delay’s exponential backoff no longer raises OverflowError at high attempt counts (e.g. attempt 1024), and both event/state subscribe-retry background tasks now discard their tracked key/handler on an unexpected failure instead of leaving a permanently stuck “subscribed” marker behind with nothing actually subscribed and nothing retrying.

• Fixes issue #829: DummyCamera exposures were broken under the pinned photutils 3.0 –make_model_image looks up source positions by the model’s own parameter names (x_0/ y_0 for Moffat2D), not by the sources table’s column names (x_mean/y_mean); now passes params_map to bridge the two.

[2.0.0] - 2026-08-26

v2.0.0.dev98 (2026-08-25)

• Observation gained an archive_url field (str | None, default None), matching the computed link pyobs-portal#82 started returning from GET /api/observations/ (a deep link into the archive’s frontend for terminal observations; always None for the pending/ in_progress states Mastermind schedules on). Without it, the strict Observation model rejected the new key and PortalObservationArchive failed every download; the validation error was swallowed by _check_for_changes’s retry handler, so the marker-gated poll loop stalled forever without ever picking up schedule changes. The field is portal-only UI metadata that Mastermind tolerates but never consumes.

v2.0.0.dev97 (2026-08-25)

• No pyobs-core code changes – version bump only.

v2.0.0.dev96 (2026-08-25)

• Breaking: pyobs.robotic.storage.backend is renamed to pyobs.robotic.storage.portal, and BackendTaskArchive/BackendObservationArchive are renamed to PortalTaskArchive/PortalObservationArchive, following the pyobs-robotic-backend -> pyobs-portal rename (see ADR 0013, specs/adrs/0013-renaming-pyobs-robotic-backend.md). There is no deprecation shim – update the class: dotted path in any deployment YAML using the old names before upgrading.

v2.0.0.dev95 (2026-08-25)

• Added Field(description=...) to every field across all pyobs.robotic.scripts Script subclasses and their nested config models (e.g. skyflat priorities) – pyobs-robotic-backend’s script builder renders each field’s JSON-schema description as form help text, and none had one (InstrumentConfig.image_type only appeared to, via a leaked ImageType enum docstring, now replaced with an explicit member listing). A new parametrized test guards every Script/config field for a description going forward. Fixes #811. (#812)

v2.0.0.dev94 (2026-08-24)

• Fixed broken __module__ overrides on PolymorphicBaseModel subclasses (SkyFlatsBasePointing, SkyflatPriorities, Archive and its subclasses including LocalArchive/PyobsArchive): a stale short path left over from an earlier pyobs.robotic.utils reorg made serialization, which keys on self.__module__, produce unparseable dumps on reload (ModuleNotFoundError). Removed the overrides so __module__ reflects the real, importable path. Fixes #806.

• BaseVideo.raw_handler now caches the per-frame (meta, frame) bytes it builds (header assembly, JSON serialization, ascontiguousarray/tobytes() copy), keyed by _frame_num. With N simultaneous raw clients (guiding, a recorder, a viewer – a case the raw endpoint (specs/design/basevideo-raw-frame-streaming.md) already supports but didn’t dedupe for), only the first consumer to wake for a given frame does the work; the rest reuse the cached result instead of redoing it (#769). Costs nothing with 0 or 1 raw clients connected.

• Mastermind.get_fits_header_before() now emits a PROJECT FITS keyword from Task.project, alongside TASK/REQNUM/OBSNUM – closes the last open item of the project-based access-control work for pyobs-archive (pyobs-archive#42; superseded by that repo’s own more detailed plan, which already shipped the archive-side filtering).

• Script module-name fields (pyobs.robotic.scripts) are now tagged with their required pyobs.interfaces via Annotated metadata, letting pyobs-robotic-backend introspect FieldInfo.metadata to render interface-filtered module dropdowns instead of hand-maintaining a duplicate table (pyobs-robotic-backend#98). Fixes #808. (#809)

v2.0.0.dev93 (2026-08-23)

• ImageWatcher no longer blocks the event loop while processing watched files: per-file FITS parsing (astropy.io.fits.HDUList.fromstring) now runs via asyncio.to_thread(), and all blocking LocalFile I/O (open/read/write/close/remove/find/exists) is routed through the default executor, mirroring the existing listdir pattern – root cause of the 2026-08-20 “Event loop stalled” incident on MONET South, where RPC handling, inotify processing, and comm keepalive froze for seconds during a slow parse. LocalFile’s open()/makedirs moved from __init__ into __aenter__ (still validating the path synchronously at construction time), closing a gap for network-mounted watch paths; every VFS consumer inherits the fix, not just ImageWatcher. See specs/plans/2026-08-20-imagewatcher-event-loop-blocking.md. (#798)

• move_radec/move_altaz now reject non-finite (NaN/inf) RA/Dec/Alt/Az up front with InvalidArgumentError, instead of letting them reach a NaN altitude comparison that was silently always False and then Angle.to_string(), whose numpy 2.x vectorized formatter raised a RuntimeWarning that application.py’s filterwarnings promotes to a hard error –killing the RPC handler mid-slew. All to_string() call sites in the affected module were replaced with manual sexagesimal formatting, mirroring pyobs-gui’s fix for the same numpy 2.x issue. Fixes #802.

v2.0.0.dev92 (2026-08-23)

• BaseVideo gained an opt-in token parameter gating its MJPEG/raw/FITS endpoints behind either an Authorization: Bearer <token> header or an HMAC-signed session cookie issued by a new browser /login page (/logout clears it); /ping stays open. The cookie is stateless (expiry plus HMAC-SHA256 over the expiry, keyed by the token), all comparisons are constant-time, and an unauthenticated request is redirected (401 -> 303) to /login so browsers land on the form. HttpFile gained a public headers property so consumers (e.g. pyobs-gui’s VideoWidget) can read the Authorization header without reaching into a private attribute. See specs/plans/2026-08-21-basevideo-http-token-auth.md. (#799)

v2.0.0.dev91 (2026-08-21)

• Demoted routine retry/shutdown log messages from WARNING to INFO.

• Fixed BIASSEC/TRIMSEC FITS-header computation: the old guard fired (and warned) for any window not covering the full visible frame on both axes – exactly the common case of a subframe with no prescan/overscan at all (e.g. guided skyflats) – while the case it claimed to describe, prescan/overscan on both axes, slipped through silently and wrote an incomplete one-axis BIASSEC. The check is now based on whether the window actually extends beyond the visible frame: fully-inside windows log at INFO (nothing to compute), and true both-axis prescan/overscan windows warn that only one axis is supported. Header output is unchanged for the common cases.

• Application now logs module-creation failures (unconsumed config keys, broken __init__ chains, …) at ERROR level with the full traceback before re-raising, so they land in the configured log file / journald instead of only on stderr – which is /dev/null for daemonized modules (--pid-file, as started by pyobsd / pyobs-web-admin). Such a failure previously vanished from all logs: the log simply stopped after the last INFO line and the module just showed as stopped.

• Fixed BaseVideo’s index page using a root-absolute /video.mjpg image URL, which 404s when the module is served behind a reverse-proxy path prefix (e.g. /pyobs/fibercamera/) since the browser resolves it against the host root instead of the page’s own path – and since that request never reaches the module, the camera is never (re)activated. Now relative.

v2.0.0.dev90 (2026-08-20)

• BackendTaskArchive and BackendObservationArchive gate their 5s refresh on the backend’s last_task_update/last_observation_update marker again. The marker is now a DB-derived Max(updated_at) (pyobs-robotic-backend#84) that is truthful across gunicorn workers, so the archives no longer re-download (and re-compare) on every poll – the per-poll content comparison was fragile and misfired whenever runtime code mutated a serialized field (see the DynamicTarget.resolve() fix below), livelocking the scheduler. The content comparison is kept as the on_tasks_changed trigger, so no-op marker bumps never fire a scheduler run.

v2.0.0.dev89 (2026-08-20)

• DynamicTarget.resolve() no longer overwrites its declared name field with the picked star’s name. That mutation leaked into Task.model_dump() (which serializes the static target), so BackendTaskArchive’s content comparison saw a “change” on every poll while the scheduler resolved a dynamic target (e.g. the CSV-picker autofocus task) – the constant on_tasks_changed kept _need_update set and every scheduler run aborted itself with “Not using scheduler results, since update was requested”, so no schedule was ever committed. The picked star is still available via the resolved target (_target) and lands in the scheduled observation; the serialized name stays "(dynamic)".

v2.0.0.dev88 (2026-08-20)

• Task gained an updated_at field (str | None, default None), matching the field pyobs-robotic-backend#84 started returning from GET /api/tasks/ (DB-derived updated_at on Task). Without it, the strict Task model rejected the new key and BackendTaskArchive failed every download, leaving the scheduler and mastermind without a task list.

v2.0.0.dev87 (2026-08-20)

• BackendTaskArchive and BackendObservationArchive no longer gate their 5s refresh loop on the robotic-backend’s last_task_update/last_observation_update marker. That marker is computed from a per-process Django LocMemCache and is unreliable across gunicorn workers, so a stale or missing marker pinned the archive’s _last_update and the mastermind kept running stale tasks/schedules (edited task parameters, deactivated tasks, newly scheduled observations, window-expired observations) forever. Both archives now re-fetch unconditionally on every poll and detect real changes by comparing the downloaded content against the cached copy (full model dumps, including observation state), firing on_tasks_changed only when content actually changed; last_changed() now reports the local time a change was last observed. Fixes #789. (#790)

• PyobsArchive and LocalArchive gained an obsnum filter on list_frames() and list_options(): PyobsArchive now emits the OBSNUM query parameter (exact match on Frame.OBSNUM), and LocalArchive filters its local index on the OBSNUM FITS header, so observations can be matched to their archived frames via list_frames(obsnum=...). Needed by pyobs-robotic-backend#82. (#791)

v2.0.0.dev86 (2026-08-19)

• Fixed FitsHeaderMixin.add_requested_fits_headers() letting a misbehaving peer’s malformed IFitsHeaderBefore/IFitsHeaderAfter response crash the whole exposure: it only caught exc.RemoteError, but XmppComm.execute() only converts IqError/IqTimeout into that, so anything else raised while making or deserializing the RPC call escaped uncaught. Now catches Exception broadly (logging the peer and failure type, skipping just that peer), while CancelledError – a BaseException – still propagates, preserving cancellation semantics. Fixes #767.

• Project gained a public flag (default False), matching the field the robotic backend (pyobs-robotic-backend#79) will start returning from GET /api/projects/. Without it, the strict Project model rejected the new key and BackendTaskArchive silently kept running on stale or empty task/project data. Fixes #786.

• Fixed XmppComm crashing on payload-less pubsub notifications (retract stanzas, node purges, and nodes with deliver_payloads off all arrive without a <payload> element under <item>): _handle_event raised AttributeError on None.text, surfacing only as a noisy “Task exception was never retrieved” since the dispatching task was never awaited. Now guards the payload access and drops payload-less notifications, and the background task gets a done-callback so any future exception is retrieved and logged normally.

v2.0.0.dev85 (2026-08-19)

• http_request_paginated() now tolerates page drift mid-fetch: DRF’s page=N pagination 404s with “Invalid page.” once N exceeds the current page count, and on a live-growing dataset with non-unique ordering (e.g. Observation.Meta.ordering = ["start"]) a concurrent insert can shift page boundaries between fetching a “next” link and following it, crashing the whole request. Now stops pagination early with whatever was already fetched – but only for that specific error shape, and never on the first page, so other 404s still raise.

v2.0.0.dev84 (2026-08-19)

• Project gained a missing users field (per-user project visibility), matching the robotic backend’s ProjectSerializer. With extra="forbid" enabled (#762), the unmodeled field previously raised a validation error instead of being silently dropped.

v2.0.0.dev83 (2026-08-19)

• The TypeError raised when a leftover kwarg reaches object.__init__() unclaimed (end of the cooperative super().__init__(**kwargs) chain added below) is now re-raised with a diagnosable message naming the offending class and the actual unconsumed kwargs – not Object’s own pre-consumption view, which could wrongly flag a kwarg a downstream mixin in the MRO had already legitimately claimed (e.g. motion_status_interfaces). The precise leftover set is read from the deepest frame in the chained exception’s traceback, i.e. the cooperative call that actually reached object.__init__() and got rejected. Previously this raised only the generic, class-and-kwarg-less object.__init__() takes exactly one argument message.

v2.0.0.dev82 (2026-08-19)

• Object.__init__ now forwards leftover kwargs cooperatively via super().__init__(**kwargs) instead of silently absorbing them, giving mixins listed after Module in a subclass’s MRO (WeatherAwareMixin, MotionStatusMixin, PipelineMixin, etc.) a real chance to claim their own kwargs – prerequisite for eventually raising on genuinely unrecognized ones. Converted the 14 pyobs-core classes that composed mixins via explicit multi-call fan-out to a single cooperative call, reordering base classes in BaseRoof/BaseTelescope/DummyMode so Module runs first (every mixin that calls add_background_task/add_child_object or reads self.comm needs Object/Module already set up). FitsHeaderMixin’s cache-path attribute is now a lazily-computed property instead of eager at __init__ time, since it can now run before Module has set self._device_name. PipelineMixin’s archive default-injection now only applies to step classes that actually declare an archive parameter (checked via signature inspection), instead of injecting unconditionally and relying on silent-drop – exactly the behavior this change removes. See specs/plans/2026-08-18-cooperative-mixin-init.md.

• Fixed a regression from the change above: unconditionally dropping auto_update=False from Scheduler’s kwarg injection (justified at the time as “never declared, no effect”) silently re-enabled BackendObservationArchive’s 5-second polling loop, which does declare and gate on that kwarg – a poller deliberately disabled when the scheduler moved to event-driven on_tasks_changed updates, live on two fleet configs (MONET South, iagvtsrv). The kwarg-support check is now a general _class_accepts_param(), reused for both the observation_archive and auto_update injections.

• Fixed WindowingWidget.value_top (pyobs/utils/gui/camera/windowingwidget.py) returning the Left spinbox’s value instead of the Top spinbox’s (copy-paste bug) – windowing offsets set via the GUI applied the wrong Y coordinate.

v2.0.0.dev81 (2026-08-18)

• No pyobs-core code changes – version bump only.

v2.0.0.dev80 (2026-08-18)

• BaseVideo now declares itself a sender of NewImageEvent via comm.register_event() in open(), matching BaseCamera’s existing registration – without it, a peer’s disco#info-driven subscription logic never subscribed (since it skips event types a peer doesn’t advertise sending), so send_event() succeeded server-side but nothing ever received it: the GUI’s video/FITS grab silently never updated, with no errors on either side.

v2.0.0.dev79 (2026-08-18)

• Fixed a TOCTOU race in BaseVideo.activate_camera()/deactivate_camera(): two concurrent callers (e.g. an incoming video-stream request racing an RPC like set_exposure_time) could both observe self._active as unset and both proceed to open/close the camera, double-connecting to the same GigE device and tripping “Controller privilege required for streaming control” on devices that only grant one controller. The check-and-set is now serialized with a lock.

v2.0.0.dev78 (2026-08-18)

• BaseVideo gained a raw-frame streaming endpoint (/video.raw) alongside the existing MJPEG stream: a multipart, event-driven feed of a JSON FITS-keyed meta header plus raw little-endian frame bytes, with latest-frame-wins backpressure (a slow client drops stale frames instead of queuing). VideoCapabilities.video is renamed to mjpeg, and a new raw field (str | None) advertises the raw endpoint’s path; live_view collapses into video_path alongside the new raw_path. add_fits_headers() is split into add_local_fits_headers() (no VFS I/O, usable per-frame) plus the persistent FRAMENUM step. Also fixed video_handler’s hardcoded 1 fps interval and lowered the default sleep_time to 60s. A connected-but-idle raw client now keeps the camera awake correctly: raw_handler bounds its wait for a new frame with a timeout and re-touches the activity timestamp even when no frame arrives, instead of only doing so when a frame is actually sent. DATE-OBS is captured at acquisition time (in _set_image()) rather than at send time, avoiding drift under frame coalescing or scheduling delay. See specs/design/2026-08-11-basevideo-raw-frame-streaming.md. (#766, #770)

• Bounded add_requested_fits_headers()’s wait for peer-supplied FITS headers with a single asyncio.wait() deadline (new fits_header_timeout kwarg on FitsHeaderMixin, default 15s), instead of awaiting every header future with no timeout at all – a peer that never answers its IQ (e.g. a laptop put to sleep without closing its client) previously stalled frame finalization for the full ~120s XMPP IQ timeout. Fixed two bugs found while implementing this: asyncio.wait() raised ValueError on an empty futures dict (any module with no comm, or no peer implementing IFitsHeaderBefore/IFitsHeaderAfter), crashing every exposure; and BaseCamera/BaseVideo passed an explicit keyword list to ImageFitsHeaderMixin.__init__ instead of forwarding **kwargs, silently dropping a configured fits_header_timeout for exactly the modules (e.g. fli230) that motivated this fix. Fixes #764. (#765, #768)

v2.0.0.dev77 (2026-08-16)

• Every module now logs the versions of loaded pyobs-* packages at startup, alongside the existing IERS-cache priming. Guarded with the same try/except as that priming, so broken dist-info metadata on any installed package can’t abort startup; an editable install won’t show up in the output (packages_distributions() can’t derive its top-level-name mapping without RECORD file entries), which is fine since the only consumer only needs this in production. (#759)

• Fixed WeatherAwareMixin retrying park() on every 10s bad-weather check while a device was already in MotionStatus.ERROR, instead of only once: the error-state guard only suppressed the first post-error park() call, then fell through and called it again on every subsequent check. Devices whose park() raises on ERROR status (BrotRoof, BrotTelescope) flooded logs with a ParkError plus “Task exception was never retrieved” every cycle, since the call happens via a fire-and-forget background task.

v2.0.0.dev76 (2026-08-15)

• BaseCamera now publishes live IExposure state (progress/exposure_time_left) once a second while EXPOSING, via a new background task; previously it only published on status transitions, so both values stayed frozen at 0 for the whole exposure. exposure_time_left is clamped to >= 0.

v2.0.0.dev75 (2026-08-15)

• BaseVideo’s image-cache write now offloads image.to_bytes() to a worker thread (matching the existing create_jpeg() pattern), instead of serializing a ~5MB FITS frame directly on the event loop – same class of fix as Vfs.write_image()/write_fits() in dev53, applied to a different choke point.

v2.0.0.dev74 (2026-08-15)

• FocusModel’s coefficient fit (lmfit.minimize plus residual model evaluation) now runs on a worker thread instead of directly on the event loop.

v2.0.0.dev71 (2026-08-10)

• Added OBSNUM: Mastermind now assigns a compound <night>-<counter> observation number when a task starts running, persisted via a VFS-cached per-night counter and written to both Observation.obsnum and the OBSNUM FITS header – lets frames from possibly-multiple cameras be tied back to the observation that produced them. See specs/design/obsnum_fits_header.md. (#738)

v2.0.0.dev70 (2026-08-10)

• Added structured progress reporting to Reduction (a progress_callback reporting master-calib creation and per-frame science calibration results as they happen, with a cumulative whole-night frame total computed via a cheap pre-pass), so a caller (e.g. pyobs-pipeline’s web UI) can drive a live progress bar instead of only tailing free-text logs. Reduction is split into an abstract ReductionBase (archive/pipeline plumbing, master-calib cache/lookup, the progress-callback mechanism) and the concrete Reduction, giving a future second reduction strategy a base to build on.

• PipelineMixin now propagates a Pipeline’s own archive into any step that doesn’t specify its own – Reduction already passed archive=archive into get_object(pipeline, Pipeline, archive=archive), clearly intending it to reach steps like Calibration, but Pipeline.__init__ absorbed it into Object.__init__(**kwargs) and dropped it, so every Calibration step had to redundantly repeat the same archive config already set at the Reduction/site level. An explicit per-step archive still overrides it.

• Fixed Calibration aborting when a science frame arrived with a catalog already attached (e.g. quick-look photometry done at the telescope): it tripped Image.trim()’s stale-catalog guard, even though to_ccddata() never carries a catalog through anyway. The pre-existing catalog is now dropped before trimming.

• Made AperturePhotometry abstract instead of exporting it as if instantiable directly – it’s a base class for PhotUtilsPhotometry/SepPhotometry (each supplies its own calculator), and its __module__ override advertised a dotted path the package’s __init__.py never actually re-exported. Breaking change for any external code instantiating AperturePhotometry directly (subclasses that define their own __init__, as both existing ones already do, are unaffected).

• Removed the generated TypeScript interface exports (export/typescript/) – breaking change for any external tooling consuming them.

v2.0.0.dev68 (2026-08-09)

• Renamed pyobs.utils.pipeline.Night to Reduction (fits solar telescopes too, not just nighttime ones). Replaced store_local with a single output param (local path, dict, or Archive), so input and output can be different archives. Implemented LocalArchive.upload_frames (previously a silent no-op), including a ValueError on a missing FNAME header and a _update_root() refresh after write; the local output directory is now auto-created if missing. Calibration-frame creation failures are now isolated per instrument/binning/filter combination instead of aborting the whole run. Fixed a missing return None after the post-download frame-count check. Removed the dead worker_procs param and the **kwargs catch-all on Reduction.__init__, so a typo’d kwarg now raises TypeError instead of being silently swallowed. Breaking change for any external code using Night, store_local, or worker_procs directly.

v2.0.0.dev67 (2026-08-06)

• Re-landed the OnDemandScheduler event-loop offload (added in dev53, briefly reverted in dev66) unchanged, alongside a prefetch/freeze split for ObservationArchiveEvolution: a new prefetch() fetches every task’s observations plus the one real “current” night (night(start)) up front, then freeze() makes a subsequent task-id cache miss raise RuntimeError instead of silently falling back to a live archive lookup, while a miss on any other (necessarily later, necessarily empty) night seeds an empty result instead of fetching or raising. Groundwork for eventually moving that evaluation from a worker thread to a worker process (real GIL isolation); the thread-shared live archive connection was the blocker, since a process pool can’t share it. See specs/plans/scheduler-archive-prefetch-for-process-isolation.md.

v2.0.0.dev66 (2026-08-05)

• Briefly reverted the OnDemandScheduler event-loop offload added in dev53 while investigating the process-isolation groundwork above; re-landed unchanged in dev67, so only a dev66 build itself ran evaluation synchronously on the event loop again.

• Mastermind now logs a task failing with a domain PyobsError (e.g. AcquisitionError) at INFO instead of ERROR with a full traceback – these are expected outcomes, not bugs; only a genuinely unclassified exception still logs loudly.

v2.0.0.dev65 (2026-08-05)

• Fixed a scheduler event-loop stall traced live (py-spy dump) to the main thread, not the worker thread the dev53 offload targets: ObservationArchiveEvolution.evolve() called Time.now().night_obs() directly on the event loop for every scheduled task, redoing an uncached astropy sunset computation each time – a 52-block schedule triggered this dozens of times in a few seconds. evolve() now takes the night as a parameter (data.night(task.start), reusing DataProvider’s existing cache) instead of computing it from Time.now() internally. Also a correctness fix: keying off “now” filed a task scheduled hours ahead under the wrong night’s bucket, which a later FollowMerit/PerNightMerit check for that night would then miss. See specs/steering/scheduler-cpu-bound-merit-evaluation-stalls-event-loop.md.

• IERS auto-download being disabled (iers_offline) now logs at info instead of warning – it’s a deliberate config choice, not an anomaly.

v2.0.0.dev64 (2026-08-05)

• Fixed ArchiveFile._upload referencing self._auth, which was never set (leftover from before auth moved to self._headers/token-based auth) – any actual write to a pyobs-archive backend crashed with AttributeError before the request was even sent.

v2.0.0.dev63 (2026-08-05)

• Downgraded routine no-detection acquisition log lines (“no on-sky distance found”, a missed sun/star detection) from warning to info – neither is actionable on its own, since the acquisition loop just retries; warning level was noise for something expected to happen occasionally.

v2.0.0.dev62 (2026-08-05)

• ImageFitsHeaderMixin’s missing-WCS-header warnings (CDELT1/2, CRPIX1/2, the CD-matrix) now fire once per module lifetime instead of once per frame, for cameras missing the relevant config/header fields.

• BackendTaskArchive now follows pagination when fetching tasks/projects/observations from pyobs-robotic-backend (DRF PageNumberPagination, 100/page) via a new http_request_paginated() that follows the “next” link until exhausted – previously only the first page’s results were ever read, so tasks or projects past page 1 were silently invisible regardless of how long they’d existed.

• Mastermind’s get_next_observation/get_current_observation now skip and log an observation whose task failed to resolve (Observation.task left None – e.g. a deleted task, or the backend task-cache poll racing the observation-schedule poll) instead of returning it, which previously propagated a bare None into task_runner.can_run() and crashed Mastermind’s background thread with an AttributeError.

v2.0.0.dev61 (2026-08-04)

• Added CORS headers to HttpFileCache.download_handler (previously blocked browser fetch() clients like pyobs-web-client from cross-origin reads) and an opt-in token param, checked via constant-time compare, plus the required OPTIONS preflight handler – replacing HttpFile’s Basic Auth, which HttpFileCache was silently ignoring since the server never enforced any auth at all. (#725)

v2.0.0.dev60 (2026-08-04)

• Moved the iers_offline warning log after handler setup, so it actually reaches a configured log handler instead of firing before one exists.

v2.0.0.dev59 (2026-08-04)

• Added an iers_offline config option: the existing PYOBS_IERS_OFFLINE env-var check is now also reachable via Application.__init__(iers_offline=...) and wired through PyobsCLI.GLOBAL_CONFIG_KEYS, letting /etc/pyobs.yaml (or a --iers-offline flag) disable astropy’s IERS/leap-second auto-download regardless of how the module process was spawned (pyobsd vs. pyobs-web-admin), unlike an env var, which only propagates through whichever launcher’s own environment it happens to inherit.

• Fixed syslog being unsettable via pyobs.yaml: it was missing from PyobsCLI.GLOBAL_CONFIG_KEYS even though a --syslog flag exists, so a pyobs: section in the config file silently had no effect on it.

v2.0.0.dev58 (2026-08-04)

• pyobsd status’s CPU measurement is now opt-in (--cpu-interval SECONDS, sampling all modules over one shared sleep) instead of always-on – previously every “status” call paid a blocking 0.1s per running module, and that short a window is too close to /proc’s tick granularity to give a meaningful reading anyway. Uptime/RSS still report instantly.

v2.0.0.dev56 (2026-08-04)

• Also prime the leap-second table when warming astropy’s IERS cache at startup (see dev55): IERS_Auto.open() only covers UT1-UTC/polar motion, and the leap-second table is a separate auto-download that was still landing inside basetelescope.py’s celestial-header task and blocking the loop there even after dev55’s fix. Uses update_leap_seconds() (not the lower-level LeapSeconds.auto_open()), which both fetches it and installs it into erfa.

• Exported DummyMode from pyobs.modules.utils.

v2.0.0.dev55 (2026-08-04)

• Astropy’s IERS cache is now primed once at startup, off the event loop via an executor, before module/comm setup – left implicit, the first UT1-UTC lookup happened inside basetelescope.py’s periodic celestial-header task and triggered astropy’s auto_download synchronously, blocking the event loop for however long the download took at an unpredictable point during normal operation.

v2.0.0.dev54 (2026-08-04)

• XmppComm’s disco#info now tags each advertised <event> element with a role attribute ("send", "subscribe", or "send subscribe"), derived from splitting Comm._registered_events into _events_sent/_events_subscribed. Previously the two were merged into one undifferentiated set, so a client with no access to the pyobs.events catalog (e.g. pyobs-web-client) couldn’t tell which events a module actually publishes vs. which it only listens for. Comm.unregister_event() now also drops an event from _events_subscribed once its last handler is removed, so a torn-down subscription stops being advertised (previously it stayed advertised forever). See specs/plans/event-role-advertising.md.

v2.0.0.dev53 (2026-08-03)

• Offloaded OnDemandScheduler’s per-timestep constraint/merit evaluation (find_next_best_task/check_for_better_task/can_postpone_task) onto a dedicated single-worker ThreadPoolExecutor (new pyobs/robotic/scheduler/_executor.py’s run_cpu_bound), and cache target-independent sun/moon computations in DataProvider per timestep. These constraint/merit __call__ methods are declared async but never actually await – they run synchronous astropy/astroplan work directly on the event loop, blocking it for multiple seconds at real fleet scale; root cause of a 2026-07-30 “Event loop stalled” warning on the scheduler module. Same class of fix as Vfs.write_image()/write_fits() below. See specs/plans/scheduler-event-loop-blocking.md. (#712)

• ImageProcessor gained an on_error kwarg ("raise" (default) / "error" / "info" / "ignore") and a handle_error(image, error) override point, and PipelineMixin.run_pipeline() (pyobs/mixins/pipeline.py) now catches each step’s ImageError and dispatches per that step’s setting, instead of the whole chain aborting on the first exception from any step (previous behavior, equivalent to on_error="raise" everywhere). Non-ImageError exceptions still always propagate. Lets one step’s failure be non-fatal while another stays fatal within the same pipeline run – e.g. a calibration path that must have a valid WCS vs. a quick-look pipeline where a missing WCS is fine. Originally proposed in #328 as a nested ExceptionHandler wrapper processor; rejected there for adding config nesting for what should be a per-step toggle. AstrometryDotNet’s exceptions: bool kwarg is now deprecated in favor of on_error (exceptions=False == on_error="error"; on_error takes precedence if both are set) and its former internal try/except was removed –its on_error/handle_error dispatch now only happens when it runs as a pipeline step via run_pipeline(). Calling it (or any other migrated processor) directly, outside a pipeline, always raises on failure regardless of on_error/exceptions. See specs/plans/pipeline-step-error-control.md.

• Vfs.write_image()/write_fits() (pyobs/vfs/vfs.py) now serialize the FITS data (image.writeto()/hdulist.writeto(), including gzip compression for .fits.gz targets) via asyncio.to_thread() instead of directly on the event loop. That serialization is CPU-bound and can take multiple seconds for larger frames, and running it inline blocked every other coroutine in the module for the duration – including state-publishing background tasks and comm handling – observed as multi-second event-loop stalls in pyobs-iagvt’s SunCamera/FTS right at the vfs.write_image() call. Callers that were working around this themselves (serializing via asyncio.to_thread and uploading raw bytes with write_bytes() instead) no longer need to.

v2.0.0.dev52 (2026-07-29)

• Guarded optional WCS/image-type handling against missing capabilities: _ResponseImageWriter (astrometry.net response saving) unconditionally deleted PC/CDELT header keywords after a solve, raising KeyError when a header lacked one of them – now a no-op if the keyword is already absent. Acquisition/AutoFocusSeries required IImageType from the configured camera via a hard proxy, unlike AutoGuiding’s identical use case, which already tolerates a camera without it via safe_proxy – now consistent.

v2.0.0.dev51 (2026-07-28)

• No pyobs-core code changes – entirely root-causing and mitigating the ejabberd shaper/ xmpp_socket.erl reactivation bug behind the iag50srv capability-fetch timeouts (see specs/plans/ejabberd-throughput-benchmarking.md); the shaper fix itself (raising shaper.normal/shaper.fast) is an ejabberd-side configuration change, applied to iag50 production and baked into the new scripts/xmpp/install-ejabberd.sh for future hosts.

v2.0.0.dev50 (2026-07-28)

• Fixed the XMPP wire serializer silently corrupting two more shapes of value: Time-typed fields (e.g. WeatherState.time) had no encode/decode support at all and arrived client-side as a plain string; and a field declared dict[str, str]/str could arrive as a stray bool if the isinstance-based encode dispatch won over the field’s own declared type (seen as ModeState.modes containing bools instead of strings). Follow-up fixes for a not-yet-restarted sender still running the old serializer: Time decode no longer pins to isot format (a legacy sender’s plain str(Time) fallback used the space-separated iso format, which the strict parse rejected), and boolean/int/double values are now coerced to str on decode as well as encode.

v2.0.0.dev49 (2026-07-28)

• Fixed XmppComm.__init__ raising a bare, message-less ValueError on a malformed JID (found via a real login-window crash on a saved account with a trailing-slash, no-resource JID). The JID regex is now a module-level, properly end-anchored pattern (the previous re.match wasn’t end-anchored, so user@domain/res/extra matched as valid), and the new is_valid_jid() lets a caller taking raw user input (e.g. a login window) validate upfront instead of finding out via an exception.

• Fixed Application.run() crashing on a benign qasync shutdown race, where a Qt-backed event loop (pyobs-gui) can report itself stopped before the main task/cancellation gather has actually finished – now handled the same way as the existing CancelledError case.

• Application can now defer constructing its Module until the event loop is already running, via a new module_factory/loop_module_class alternative to config – letting the factory await something like a login dialog’s submit signal before the module and its comm connection exist at all. Existing config-file callers are unaffected.

v2.0.0.dev48 (2026-07-27)

• A module declaring a stateful interface but never publishing it (see the state-publishing enforcement below) now logs as an error, not a warning – it’s a standing per-module defect that recurs on every startup, not a transient condition.

v2.0.0.dev47 (2026-07-27)

• Every Module now runs a background watchdog that detects the event loop itself stalling: it times its own wakeups against how long it asked to sleep for, so a synchronous blocking call anywhere in the module (or a background task) shows up as a logged stall – once when it starts, once when it clears, with total duration – instead of only being visible indirectly as peers timing out trying to reach that module.

v2.0.0.dev46 (2026-07-27)

• The capability-fetch retry warning now includes the actual exception that caused the retry, instead of just noting that one occurred.

v2.0.0.dev45 (2026-07-27)

• Fixed FocusModel.open() skipping its state publish entirely when the weather module wasn’t reachable yet at startup (a common ordering race), which then always tripped the “declares state, but none published” warning; it now falls back to a placeholder focus=0.0 until the background update loop replaces it with a real value.

v2.0.0.dev44 (2026-07-27)

• DummyComm now takes a real name instead of a hardcoded "module" placeholder (Application fills it in from the config file’s stem for comm-less modules), fixing indistinguishable PYOBS_MODULE log tags and a guaranteed false positive against the config-stem-mismatch warning.

• Fixed Comm’s log-forwarding failure handler re-queuing its own failure log through the same queue it was draining, turning a sustained send outage into an unbounded stream of self-generated failure reports.

• pyobsd’s per-module file logging (--log-file) is now opt-in (--file-log/ --no-file-log, default off) instead of always passed to every started module regardless of the already-default-on journal logging; non-systemd deployments relying on the old always-on file logging need to pass --file-log explicitly.

• Removed IRunning.is_running() (and its ~12 implementers) now that RunningState (pushed via comm.set_state(IRunning, ...)) already covers the same information – both existed side by side for the same boolean.

• Extended the max_age freshness check (see below) to FollowMixin (default 3x its poll interval) and FocusModel’s temperature reads (default 2x), so a dead publisher’s frozen last value is treated as unavailable instead of trusted indefinitely.

v2.0.0.dev43 (2026-07-27)

• Added an opt-in max_age freshness check to Proxy.get_state()/wait_for_state(): a cached state value past its max_age is now treated the same as not-yet-published (None/timeout) instead of trusted forever, even if the publisher’s update loop has died or a stanza was shaper-delayed by minutes. WeatherAwareMixin passes weather_max_age (default 120s) so a dead Weather update loop degrades to bad-weather within a couple of check cycles.

• Comm now tracks which interfaces have actually had set_state() called, and Module.startup() warns when a module implements a stateful interface but never published it – caught Weather/DummyCamera both only ever publishing IWeather/ITemperatures from a background loop, never synchronously from open(); both now also publish a placeholder value from open().

• Fixed non-TLS XMPP connections failing entirely (“No appropriate login method”) against ejabberd 26.4.0: unencrypted_scram was already set for use_tls=False, but slixmpp also needs unencrypted_plain set explicitly before it will attempt PLAIN over a plaintext connection.

v2.0.0.dev42 (2026-07-26)

• Removed racy, non-gating has_proxy() startup checks in open() (FlatField, BasePointing/Acquisition, AutoFocusSeries, FlatFieldScheduler) – one-shot, no-retry, 5s presence checks that routinely false-positived under any slow/delayed presence propagation without actually gating anything (every real use of the peer already re-resolves a fresh proxy and raises its own well-attributed error if genuinely missing).

• LocalComm now broadcasts ModuleOpenedEvent on mark_ready(), matching XmppComm’s presence-driven behavior – previously a module joining the local network after a peer had already started stayed invisible to that peer’s ModuleOpenedEvent handlers until the peer restarted.

• BaseCamera no longer aborts closed-shutter (DARK/BIAS) exposures or sequences on bad weather, since the shutter being closed already means weather can’t affect them.

• Fixed the config-filename/module-name mismatch check (see below) firing a spurious warning against a legitimately-deactivated _test.yaml-style config, by stripping the leading underscore from the config stem before comparing.

v2.0.0.dev41 (2026-07-26)

• FlatField.open() now publishes IFilters state unconditionally, instead of only once set_filter() had been called via RPC – previously the pubsub state node never got created on a fresh start, leaving subscribers (e.g. GUI widgets) retrying forever.

• The config-filename/module-name mismatch check added below now logs a warning instead of raising, so a genuine mismatch no longer prevents a module from starting at all.

v2.0.0.dev40 (2026-07-25)

• A config file starting with an underscore (e.g. a deactivated _test.yaml) is now logged with its leading underscores stripped from the module-name tag, instead of tagging log lines with the raw filename stem.

v2.0.0.dev39 (2026-07-24)

• Application now refuses to start a non-MultiModule whose config filename doesn’t match its own comm-derived name, catching a class of drift where background tasks/RPC log lines use the real name but everything else falls back to the config file’s stem – silently splitting a module’s PYOBS_MODULE log tag in two whenever a config file is renamed independently of its comm configuration.

v2.0.0.dev38 (2026-07-21)

• Suppressed astropy’s VerifyWarning on FITS header comment truncation in Image.writeto(), the single choke point every FITS write funnels through – expected whenever a header value (aggregated from many modules) is long enough to leave no room for its comment within the 80-char card limit; the value itself is untouched, so the warning had nothing actionable to report.

v2.0.0.dev37 (2026-07-21)

• Comm.get_interfaces() now waits briefly for a not-yet-discovered client instead of raising KeyError instantly, tolerating a caller’s own startup racing slightly ahead of this module’s peer discovery (which only populates once _got_online has processed the peer’s presence). Also fixed a docstring/behavior mismatch: it promised IndexError on not-found, matching _get_client’s existing handling, but actually raised KeyError.

v2.0.0.dev36 (2026-07-21)

• Proxy.wait_for_state() now returns None on timeout instead of letting asyncio.wait_for’s TimeoutError propagate – every caller (waitformotion.py, weatheraware.py, pyobs-iagvt’s SunCamera, others) already checked for None immediately afterward, so a peer simply not having published yet was turning into an unhandled exception for all of them.

v2.0.0.dev35 (2026-07-21)

• Event handler exceptions are now logged properly (with handler and event context) via log.error() instead of surfacing only as asyncio’s generic “Task exception was never retrieved” warning at garbage-collection time, since _send_event_to_module dispatches every handler as a fire-and-forget background task with nothing checking the result.

v2.0.0.dev34 (2026-07-21)

• Added jitter to _safe_send’s retry wait (send_event/_set_state’s underlying path), matching the indefinite-retry jitter added for capability/state-subscribe fetches below –without it, many modules’ _safe_send calls retrying around the same moment (e.g. a fleet-wide restart) stayed in lockstep instead of spreading out.

v2.0.0.dev33 (2026-07-21)

• Capability-fetch and state-subscribe retries (_get_capabilities/_subscribe_with_retry) no longer give up permanently after a fixed attempt budget – they now retry indefinitely with capped exponential backoff and full jitter, stopping only when the peer actually goes offline or the last subscriber unsubscribes. A fixed budget could be exhausted across many module pairs at once by a simultaneous fleet-wide restart, leaving capabilities/state permanently missing until a manual restart.

v2.0.0.dev32 (2026-07-21)

• XmppComm’s XEP-0199 keepalive ping interval/timeout is now configurable, instead of a fixed 30s default that can be shorter than an ejabberd shaper’s IQ reply delay under load – letting a deployment tolerate a known-slow server instead of only a genuinely dead connection.

v2.0.0.dev31 (2026-07-21)

• Fixed pubsub subscribe retries only catching IqError, not the IqTimeout that _safe_send also raises once its own internal retries are exhausted – an uncaught IqTimeout killed the whole 30-attempt subscription retry loop after a single slow response, surfacing only as an unretrieved background-task exception instead of the intended “could not subscribe” warning, with the subscription then permanently abandoned.

v2.0.0.dev30 (2026-07-21)

• WeatherAwareMixin no longer parks on the first weather-check connection hiccup – a common right-after-restart race where XMPP roster/presence for the weather module hasn’t settled yet. A connection failure now only counts as bad weather after 3 consecutive failed checks.

v2.0.0.dev29 (2026-07-18)

• Modules no longer accept RPC calls until they’ve actually finished starting up. Some drivers (e.g. camera modules connecting to hardware) take a while inside open(), but were previously reachable over XMPP – and visible to peer discovery – the instant they connected to the server, long before open() returned. New ModuleState.STARTING, set at construction and cleared by the new Module.startup() once the full open() override chain (base setup plus every subclass’s own) has completed; Module.execute() now rejects any call outside a small introspection/recovery whitelist (get_permitted_methods, reset_error) with the new ModuleStartingError while a module is still STARTING. XmppComm/XmppClient also hold back the initial XMPP presence broadcast until a module reaches READY, closing a race where a peer reacting to _got_online could read capabilities (e.g. hardware-dependent ones like IWindow/IBinning) that hadn’t been published yet – scoped to comms with an actual starting Module attached, so bare/GUI-style XmppComm usage announces itself exactly as before. Application and MultiModule both call Module.startup() instead of open() directly now; any other code that opens a module standalone (a custom script, a test) needs to do the same, or call set_state(ModuleState.READY) itself after open() – see module-startup-gating.

v2.0.0.dev28 (2026-07-18)

• New InvalidArgumentError for bad-argument validation on RPC-exposed methods (unknown filter name, invalid focus value, invalid config parameter, out-of-range grab_sequence count/delay, …) – previously these stayed as plain ValueError, which works fine locally (LocalComm, direct calls, tests) but silently degrades to UnclassifiedError the moment the same call crosses XMPP, since ValueError is a builtin and never in the registry. That inconsistency is exactly the kind of bug that only shows up once code moves from a local test to a networked deployment. Fixed for IConfig.get_config_value/get_config_value_options/ set_config_value, IDataSequence.grab_sequence, ITrackingMode.set_tracking_mode, IFocuser.set_focus, IFilters.set_filter, IMode.set_mode, and IWeather.get_sensor_value (MockWeather’s implementation – the real Weather class’s own ValueError there is about a malformed station response, not a bad argument; see the next entry). Also reused the existing DeviceBusyError for FlatField.flat_field/ FlatFieldScheduler.run’s “already running” check, which was never actually about a bad argument either. Scoped out first (see DESIGN_exception_handling.md) before touching anything: real driver repos (e.g. pyobs-sbig) likely have the identical ValueError pattern for real hardware and would need the same companion fix, not reachable from this PR.

• New WeatherResponseError (pyobs.modules.weather.weather) for Weather.get_sensor_value getting back a station response missing its time/value fields – a different shape of problem from the bad-argument case above (an external dependency being flaky, not a caller mistake), same reasoning as BodyResolutionError: plausibly transient, worth retrying, not the caller’s fault. WeatherStatus.status’s similar-looking ValueError (weather_state.py) turned out to be unreachable from any RPC caller at all – it only fires inside a background polling loop that already catches it broadly and just logs a warning – so it’s left untouched.

• Fixed UnclassifiedError.original_type silently not surviving the wire: Module.execute() wraps a non-domain exception (IndexError, a vendor SDK exception, …) as UnclassifiedError before rpc.py ever sees it, but fault_to_xml was serializing the wrapper’s own class name instead of the original type it was tagged with – since UnclassifiedError itself is a registered type, the caller reconstructed a fresh one with original_type never set, making a remote IndexError and a remote ValueError indistinguishable. Now serializes the original type name instead, so the caller’s own registry lookup runs against it and correctly repopulates original_type on the (still-unresolvable) UnclassifiedError it falls back to. Also fixed in the same pass: every exception crossing the wire had its message doubled ("<ClassName> <ClassName> message") once displayed, because the message field serialized str(exception) (already "<ClassName> message") instead of the raw message, which reconstruction then fed back in as the new instance’s own message before formatting it again.

• Docstring sweep across every interface flagged by the exception-handling audit (16 of 27 documented interfaces had at least one mismatch) – Raises: clauses now match what’s actually raised: IFocuser/IPointingRaDec/IPointingAltAz/IPointingBody/ IPointingOrbitalElements/the Heliocentric*/Helioprojective family gain NotSupportedError/MissingObserverError/AltitudeLimitError/BodyResolutionError/ InvalidOrbitalElementsError (or a note that they propagate from the underlying RA/Dec move); IAutoFocus/IAcquisition now document the types their own @raises already declared instead of a stale ValueError; IData/IDataSequence document DeviceBusyError; IFocusModel documents its three new leaf types; IExposureTime/IFlatField/IWeather gain the clauses their implementations already needed. FlatField.set_filter’s docstring copy-paste bug (“If binning could not be set” on a filter setter) is fixed. The handful of interfaces with zero concrete implementers anywhere in this repo (ICalibrate, ISyncTarget, IMultiFiber.set_fiber, IPointingSeries, IRotation, IScriptRunner.run_script) gain a plausible Raises: clause so a future implementer has a contract to follow instead of guessing – the same guess that produced several of the mismatches this sweep fixes. Documentation only, no behavior changes. Note: several interface methods still document plain ValueError for bad-argument validation (as opposed to domain operation failures) – deliberately not promoted to typed exceptions in this sweep; see DESIGN_exception_handling.md’s “Still open” section for the tradeoff.

• Documented the AbortedError contract on every abort_event-taking hook in pyobs-core (BaseCamera/BaseSpectrograph._expose(), BaseTelescope._move_radec/ _move_altaz) – nothing had ever told driver authors which type to use for “this was cancelled, not a real failure,” so two of pyobs-core’s own in-tree implementations (DummyCamera, DummySpectrograph) had each independently guessed a different wrong type (InterruptedError, ValueError) for the exact same condition. Both now raise exc.AbortedError, as does _DummyTelescopeBase.set_focus’s equivalent abort check. (Two more instances of the same guessed-wrong-type pattern, in pyobs-sbig/pyobs-fli, are a companion fix in those repos, not something this PR can reach.)

• Documented the domain/transport split in pyobs/utils/exceptions.py as a deliberate axis: RemoteError and its subtree (RemoteTimeoutError, ForbiddenError) mean “the call itself didn’t reach/return,” which doesn’t benefit from the same fine-grained-per-reason treatment domain exceptions get – documentation only, no behavior change.

• First sweep of concrete exception-typing gaps (goal 5: specific types over generic ones/bare builtins). New cross-cutting types in pyobs.utils.exceptions: DeviceBusyError (“this device can’t service this request right now, back off and retry”) and NotSupportedError (“this module doesn’t implement this optional capability at all”). CameraException (BaseCamera) and AcquireLockFailed (LockWithAbort, a plain Exception that leaked out of move_radec/move_altaz/set_focus/stop_motion/roof init/park unconverted) are retired – both meant the same thing, “device busy,” now unified as DeviceBusyError. The telescope/roof init()/park() boundary specifically translates a lock-acquisition failure (or any other failure) into InitError/ParkError instead, per IMotion’s own documented contract, following BaseCamera.__expose()’s existing catch-and-translate pattern. Capability-check NotImplementedError sites (an alt/az-only telescope’s move_radec, ScienceFrameAutoGuiding.set_exposure_time, a dummy telescope’s set_focus_offset) now raise NotSupportedError instead. BaseTelescope gained MissingObserverError/AltitudeLimitError/BodyResolutionError/ InvalidOrbitalElementsError (all MotionError) for its previously-bare ValueError sites. FocusModel gained WeatherDataError/FocusTimeoutError/MissingSensorError (all FocusError). New ScriptError (pyobs.robotic.scripts) wraps whatever a script’s run() raises that isn’t already a domain exception, following the same pattern; AutoFocusScript.can_run() now checks for a target itself instead of only discovering its absence after run() has already started. BaseVideo/BaseSpectrograph.grab_data()’s “no image” ValueError sites now raise GrabImageError, matching BaseCamera. All new leaf types live next to the code that raises them (basetelescope.py, focusmodel.py, pyobs.robotic.scripts), not bolted onto exceptions.py, now that the registry from the previous step lets a domain exception survive the wire regardless of which module defines it. Fifth step of the exception-handling rollout in DESIGN_exception_handling.md (tracks #446); remaining items in that sweep (ScriptRunner’s per-script leaves if ever wanted, driver-repo AbortedError/NotImplementedError fixes) are out of scope for a pyobs-core PR alone.

• RPC calls over XMPP now carry a correlation id end to end: the origin-side log line for a domain exception (Module.execute()’s catch block) includes (call_id=...), and the same id is attached to the exception the caller receives as exception.call_id – reusing XEP-0009’s existing per-call iq["id"] rather than adding new plumbing. Lets an operator debugging a caller-side FocusError jump straight to the matching detailed log on the module that actually raised it, instead of neither side’s log line pointing at the other. Purely additive, no migration required; not set for LocalComm/MultiModule calls, which are already in the same log stream as the caller. Fourth step of the exception-handling rollout in DESIGN_exception_handling.md (tracks #446).

• Constructing a PyobsError is now side-effect-free, ordinary Python. raise exc.FocusError(...) always raises a FocusError – it no longer risks silently coming back as a SevereError instead, which could happen because the old severity-escalation metaclass intercepted construction, not raising or catching. SevereError is retired entirely: nothing in this repo or any sibling project ever caught it specifically, its only real consumer was register_exception’s callback (already used everywhere; production code never actually set throw=True), which already does the meaningful part itself (set_state(ModuleState.ERROR)). register_exception/handle_exception move from module-level free functions with process-global state to Module._register_exception()/an internal _record_exception(), called from Module.execute()’s catch block (the same chokepoint that already classifies and logs) – fixing a real cross-instance bug as a byproduct, where two Module instances in the same process (e.g. under MultiModule, or two instances watching the same remote module) used to share one counter. Ten in-tree call sites plus one in pyobs-alpaca need the mechanical exc.register_exception(...) -> self._register_exception(...) rename (the throw parameter is gone with the substitution it existed for). Also: any non-PyobsError exception escaping a module’s method body is now wrapped as UnclassifiedError right in execute(), not only on the XMPP fault path, so LocalComm/MultiModule get the same safety net as XMPP; RPC._on_jabber_rpc_method_call no longer logs domain exceptions itself since execute() already did (it still logs failures that never reach execute(), like malformed RPC parameters). Third step of the exception-handling rollout in DESIGN_exception_handling.md (tracks #446).

• A remote domain exception now arrives at the caller as its real type, catchable directly (e.g. except exc.FocusError: around a proxy call actually fires now) – previously every remote failure, transport or domain, arrived wrapped in InvocationError (now retired entirely), so a caller could only catch the broad wrapper and manually unwrap .exception. Exception classes are now resolved via a registry (PyobsError.resolve(), populated automatically via __init_subclass__) instead of a getattr lookup restricted to pyobs.utils.exceptions, so a domain exception can now live anywhere (a driver package, a pyobs-core submodule) and still survive the wire, keyed by fully-qualified name rather than bare class name. An exception that can’t be resolved (a raw builtin, a vendor SDK exception, or a domain type whose defining module was never imported in this process) arrives as the new UnclassifiedError instead of silently degrading to a generic RemoteError with only the message surviving – the original type’s qualified name is preserved as UnclassifiedError.original_type. UnclassifiedError joins ModuleError/SevereError as unsuppressible and always-loud. PyObsError is renamed to PyobsError (naming consistency with PyobsArchive, PyobsCLI, etc.) and its constructor is now PyobsError(message=None, **context), storing every keyword generically as an attribute – RemoteError/RemoteTimeoutError/ ForbiddenError no longer have their own constructors, so direct construction now takes module=/sender=/method= as keywords instead of fixed positional arguments. Breaking change for any external code constructing these directly, catching exc.PyObsError/ exc.InvocationError by name, or relying on a remote failure always arriving as some RemoteError subclass (pyobs-gui’s base.py/mainwindow.py reference PyObsError by name and need the rename applied). Second step of the exception-handling rollout in DESIGN_exception_handling.md (tracks #446).

• Every RPC-exposed method raising a domain PyobsError now logs a quiet INFO line locally by default, without a traceback – previously this only happened for methods explicitly decorated with @raises(...) (used on exactly two methods), and every other domain exception logged at ERROR with a full traceback despite the caller already receiving the same error. @raises no longer controls log level (documentation value only, for now); Module gained _disable_exception_logging(*exception_types) for a module to opt a high-frequency exception type out of even the quiet line entirely, since the caller already has it. ModuleError/SevereError are exempt from both the quiet default and the opt-out – they always log loudly, since both mean “this needs a human’s attention at the source,” not “an anticipated domain failure.” Part of the first step of the exception-handling rollout in DESIGN_exception_handling.md (tracks #446).

v2.0.0.dev27 (2026-07-16)

• Fixed PhotometryFocusSeries systematically underreporting its focus error by a factor of its own square root: fit_hyperbola returns (focus, variance), not (focus, error).

• curve_fit (pyobs.utils.curvefit.fit_hyperbola) now passes absolute_sigma=True, so the reported variance reflects the actual per-point measurement uncertainty passed in via y_err instead of being silently rescaled to force a reduced chi-square of 1; the weighted-mean focus calculation now uses inverse variances accordingly.

• Guiding-statistics FITS keys (GUIDING UPTIME/RMS/RMS1/RMS2) now get an explicit HIERARCH prefix, avoiding astropy’s VerifyWarning from auto-promoting them (they exceed the 8-char keyword limit).

v2.0.0.dev26 (2026-07-16)

• ICamera/ISpectrograph no longer inherit IExposure – they’re now pure IData identity interfaces (“this module produces images/spectra”), not “…and has an exposure clock.” BaseCamera and BaseSpectrograph (which push real ExposureState) now declare IExposure explicitly instead of getting it for free; PipelineCamera drops IExposure entirely instead of publishing a fabricated, never-updated ExposureState for a pipeline run that has no exposure to report progress on. Breaking change for any external module that subclasses ICamera/ISpectrograph directly (not via BaseCamera/BaseSpectrograph) and relied on inheriting IExposure implicitly – it needs to add IExposure to its own bases now if it actually wants to publish exposure state (e.g. pyobs_iagvt’s SunCamera, tracked separately, not fixed here). pyobs-gui’s CameraWidget needs no change: its exposure-progress panel was already conditional on has_proxy(..., IExposure). See DESIGN_ICamera_IExposure.md.

• Added Image.trim(), unifying three previously-independent implementations of TRIMSEC parsing (inline in ProjectedOffsets, fitssec(), and Pipeline.trim_ccddata()) into one method that also keeps mask/uncertainty aligned with data. Shifts CRPIX1/CRPIX2 to account for the new origin – a correctness fix, since none of the three prior implementations did this, silently leaving a stale WCS reference pixel after trimming. Raises ValueError if a catalog is already attached, since its pixel coordinates would otherwise silently go stale against the trimmed frame – run source detection after trimming, not before. Removed Pipeline.trim_ccddata(); its two call sites now trim the Image before converting to CCDData instead of after – breaking change for any external code calling it directly. fitssec()’s parser is now shared (pyobs.utils.fits.parse_section_bounds) and raises a well-defined ValueError for a malformed section keyword instead of an arbitrary IndexError/ValueError. See DESIGN_Image_trim.md.

• Added an optional delay parameter (seconds between grabs, default 0) to IDataSequence.grab_sequence(), for cadence control between grabs without resorting to dithering/offsets (a pointing-layer concern out of scope for this interface). Both abort_sequence() and abort() cut a pending delay short instead of idling out the full wait once the sequence has already been told to stop.

• Fixed Time.night_obs() crashing during polar day/night: it assumed observer.sun_set_time() always returns a valid Time, but astroplan returns a masked value when the sun never crosses the horizon in the search window; now falls back to the observer’s local calendar date.

• Fixed Scheduler reading the private _observer attribute instead of the public observer property in its altitude/airmass logging, which broke silently (AttributeError swallowed by the worker loop) on mocked scheduler instances in tests, aborting scheduling before add_observations was even called.

• Enriched existing scheduler/telescope log lines (per-block schedule summary, altitude-limit error) with the computed altitude/azimuth/airmass on each side, so a mismatch between what the scheduler validated and what the telescope sees at move time can be diagnosed without new per-candidate logging on hot paths.

• Fixed a missing OBJECT FITS header: Task/LcoTask/LcoScript.get_fits_headers() re-validated a brand-new Script instance from the stored config instead of reusing the instance run() actually executed, losing any state (like the target name) accumulated during the run; the actively-running script instance is now cached and reused for header lookups.

• Fixed CameraSettingsMixin still calling the synchronous, cache-only get_capabilities() after capability fetches became asynchronous background tasks (previous release), which made “Could not get full frame size.” fail deterministically for every IWindow camera; now awaits Proxy.wait_for_capabilities() for a Proxy camera.

• Fixed CircularMask destroying pixel data by zeroing it directly instead of flagging via image.mask – it now sets (OR’s) image.mask and leaves the underlying data untouched.

v2.0.0.dev24 (2026-07-15)

• Comm’s proxy capability fetch (previously a single blocking disco#info query with no retry) now runs as a background task retried up to 3 times, mirroring the existing state-pubsub race fix, plus a new Proxy.wait_for_capabilities() for callers that need a guaranteed value; a peer that hadn’t finished starting up no longer produces a permanent missing-capability warning.

v2.0.0.dev23 (2026-07-15)

• Added IDataSequence (grab_sequence()/abort_sequence(), pushed DataSequenceState) for taking a counted sequence of grabs in one call instead of driving a client-side loop of grab_data() calls. Implemented by BaseCamera. grab_sequence() returns immediately and runs the sequence in the background; abort_sequence() stops after the current grab finishes, while the existing abort() now also cancels the rest of a running sequence. See DESIGN_IDataSequence.md.

• Fixed a FocusModel startup race fetching module temperatures: it used get_state(), which only reflects whatever pub/sub state has arrived so far, crashing with a confusing “sensor not in data” error if the target module hadn’t published ITemperatures yet; now uses wait_for_state(), matching waitformotion.py/weatheraware.py.

• BaseCamera now aborts the current exposure/sequence on BadWeatherEvent (hard-stopping via abort(), clearing any running grab_sequence() count), matching the existing FlatField._abort_weather pattern – previously this only worked client-side, via pyobs-gui’s own sequence loop. (#547)

• Published previously-missing states for the dummy telescope/spectrograph modules: _DummyTelescopeBase (and its DummyAltAzTelescope/DummyRaDecTelescope subclasses) never published IPointingAltAz; DummySolarTelescope never published its Heliocentric/Heliographic/Helioprojective pointing states until a corresponding move_* call had already been made; BaseSpectrograph/DummySpectrograph never published IExposure. All three now publish at open() (and update live), mirroring BaseCamera.

• Published previously-missing initial/IReady states for several modules that formally require them via interface inheritance but never published them, leaving subscribers to retry indefinitely: AutoGuiding/ScienceFrameAutoGuiding’s IExposureTime, PipelineCamera’s state entirely (including the IExposure required by ICamera), FlatField’s IMotion, and MotionStatusMixin’s separately-enumerated IReady (which IMotion implies) – folding BaseRoof’s duplicate ready-derivation and DummyMode’s hardcoded publish into the mixin’s shared default.

• Added Comm.unregister_event(), the missing inverse of register_event() – previously a caller torn down while the app keeps running (e.g. a GUI widget discarded on client disconnect) could never stop receiving events, keeping it alive and firing on every future matching event. (#438)

• Fixed a ZeroDivisionError in FlatFielder._calc_new_exptime when a test flat-field’s median counts equal or fall below the bias level (e.g. sky still very dark); now clamps to the max exposure-time increase factor instead. (#481)

v2.0.0.dev21 (2026-07-15)

• Bounded XmppComm._safe_send with its own 15s asyncio.wait_for() timeout, instead of relying entirely on slixmpp’s internal IQ timeout (default 120s, which could fail to fire even on a healthy connection) – an unbounded hang here could freeze the calling module, including inside open(), which blocks the whole event loop. (#664)

v2.0.0.dev20 (2026-07-15)

• Internal only: fixed the remaining deprecated self._xmpp[...]-style slixmpp plugin-subscript access in XmppComm, following up on the earlier migration to client.plugin[...].

v2.0.0.dev19 (2026-07-15)

• Reintroduced the old IPointingHGS lon/lat contract as IPointingHeliographicStonyhurst, now a separate interface from IPointingHeliocentricPolar instead of a repurposing of it – drivers needing Heliographic Stonyhurst tracking (e.g. pyobs_iagvt’s SolarTelescope) should implement this one instead.

• Removed pyobs.modules.utils.AutonomousWarning (played warning sounds while an IAutonomous module was running). Found while writing tests for it: started_sound/stopped_sound were stored but never read anywhere, and _check_autonomous()’s sound selection looked inverted (it logged “Robotic systems started” but played stop_sound, apparently copy-pasted from _check_trigger()’s toggle logic without adjusting the polarity) – breaking change for anyone using it, no replacement provided.

• Fixed three BaseVideo bugs found while reviewing it: a request-list race where grab_data() removed its request from _image_requests without holding _image_request_lock, while _set_image() iterates/mutates the same list under that lock across an await (could skip another pending request); a fallback-filename bug where _finish_image()’s "image.fits" fallback (used when no filename pattern is configured) never reached image.header["FNAME"], the actual cache key, raising KeyError instead of degrading gracefully; and removed the dead _new_image_event (set every frame, never waited on anywhere – the real new-image notification is the separate NewImageEvent comm event).

• Split DummyTelescope into DummyRaDecTelescope (+IOffsetsRaDec), DummyAltAzTelescope (+IOffsetsAltAz), and DummySolarTelescope (+IPointingHeliocentricPolar, IPointingHeliographicStonyhurst, IPointingHelioprojective – always tracks the Sun via a dedicated background task, no compatibility alias for the old class name). See dummy-telescope-split-design.md.

• Renamed IPointingHGS to IPointingHeliocentricPolar and its fields from lon/lat to mu/psi, matching the existing HeliocentricPolarTarget – the old fields actually represented Heliographic Stonyhurst coordinates, a different frame; breaking change for any external driver implementing it (e.g. pyobs_iagvt, tracked separately).

• Object.location is now derived from Object.observer instead of being stored and propagated to child objects independently, removing a source of location/observer divergence. The location constructor argument is unchanged, but only affects the default observer built from it.

• Fixed Scheduler._on_task_finished silently ignoring TaskFailedEvent (registered as its handler alongside TaskFinishedEvent, but its own guard only accepted the latter – a sibling class, not a subclass), so a failed task never cleared _current_task_id or triggered a re-schedule even with trigger_on_task_finished enabled.

• Implemented non-sidereal tracking: new ITrackingMode/ITrackingRate interfaces (discrete vs. continuous tracking modes) and IPointingBody/IPointingOrbitalElements (track a named body or a set of orbital elements), plus ARCSEC_PER_SEC/AU units. BaseTelescope implements the continuous-tracking background task, propagating a tracked body or orbital elements (two-body Kepler solve, hand-rolled rather than via poliastro – ~6us/call for the propagation itself, dominated by astropy’s own frame-transform cost) into a live RA/Dec correction, with the refresh interval clamped against the mount’s own published min_update_interval (via the new Comm.get_own_capabilities(), mirroring get_own_state). DummyTelescope (see split above) is the reference implementation; BaseTelescope also now publishes AltAzState for IPointingAltAz telescopes. track_body() resolves via astropy.coordinates.get_body/JPL Horizons only – deliberately no automatic MPC/NEOCP-by-designation lookup, since MPC has no formal REST API and scraping would be riskiest for exactly the fresh-NEOCP-object case the feature exists for; a caller with elements in hand uses track_orbital_elements() directly. See tracking-mode-design.md.

• Fixed two crash paths in FitsHeaderMixin’s header-building: _fitsheadermixin_add_framenum() read hdr["DAY-OBS"] unconditionally, but that key is only set when DATE-OBS was present, crashing with KeyError instead of the warn-and-skip its own log message implied; and _fitsheadermixin_add_fits_headers() called date_obs.night_obs() (which needs a real Observer) whenever night_obs=True (the default) even with no observer configured, crashing with AttributeError instead of falling back to the plain calendar day like the location handling right above it already does.

• Fixed FlatFielder’s deviation check comparing a fraction against target_count (a raw count-rate) instead of the already-stored _allowed_offset_frac, so the “retry if the flat is way off target” check could essentially never trigger.

• Added an import-time registry for Interface subclasses (pyobs.interfaces.interface.get_registered_interface/registered_interfaces), and switched Module._get_interfaces_and_methods, XmppComm._get_interfaces, and Comm._interface_names_to_classes to resolve through it instead of hardcoded lookups against the pyobs.interfaces module namespace – previously a module implementing an externally-defined interface (e.g. a driver package’s own) was never discovered, advertised, or resolved by name at all.

• Fixed XMPP wire serialization corrupting/mishandling numpy scalar types: numpy>=2.0 changed np.float64’s repr() to "np.float64(x)", corrupting the <double> element text and aborting reconstruction of the whole enclosing dataclass (e.g. dropping IModule capabilities.version alongside a malformed location.longitude); np.integer types aren’t subclasses of int at all and silently fell through to the <string> fallback instead of <int>. Any np.generic scalar is now converted to its native Python type before serialization.

v2.0.0.dev18 (2026-07-12)

• Fixed DummyTelescope.set_focus_offset silently logging an error instead of raising, which made remote callers see a false success; its M1/M2 temperatures now drift like DummyCamera’s sensors instead of staying static after startup.

• pyobsd now defaults to sending module logs to the systemd journal (--syslog is on by default; pass --no-syslog to disable it).

• Added pyobsd logs [module] [journalctl arguments...], a thin passthrough to journalctl filtered to the module’s journal fields.

v2.0.0.dev17 (2026-07-11)

• Added a /ping health-check endpoint to HttpFileCache and BaseVideo, for verifying HTTP connectivity without touching the file/image cache.

v2.0.0.dev16 (2026-07-10)

• Added IStructuredConfig, letting a module push/apply its whole config dataclass as a unit (schema auto-derived via dataclass_to_schema), complementing IConfig’s per-field get/set.

• Added pydantic_to_schema, the Pydantic-model counterpart to dataclass_to_schema, for module configs (e.g. pyobs-iagvt’s FTSConfig) that need Pydantic’s own validation.

• Renamed HeliocentricPolar to HeliocentricPolarTarget for naming consistency.

• Replaced HelioprojectiveRadialTarget with HelioprojectiveTarget, using the Helioprojective frame’s Tx/Ty (arcsec) directly instead of a radial (psi/delta) representation.

v2.0.0.dev15 (2026-07-10)

• Fixed HttpFileCache rejecting uploads with “413 Request Entity Too Large” because client_max_size was never passed through to configure the upload limit.

v2.0.0.dev14 (2026-07-09)

• Fixed Scheduler, Trigger, Kiosk, PointingSeries, Weather, MockWeather, and Mastermind never publishing their advertised IRunning state, leaving subscribers retrying indefinitely.

• The LCO schedule backend now logs the next observation after a schedule update, matching BackendObservationArchive.

v2.0.0.dev13 (2026-07-09)

• Internal fixes only: resolved the remaining pyrefly type-check errors blocking CI.

v2.0.0.dev12 (2026-07-09)

• Added HelioprojectiveRadialTarget for solar coordinate scheduling.

• Added MockWeather, a deterministic in-memory IWeather implementation for tests and simulations.

• IAcquisition/IAutoGuiding now publish live state (AcquisitionState/AcquisitionAttempt, GuidingState); fleshed out their dummy modules to match.

• AcquisitionResult/AcquisitionAttempt now use a single offset_frame (RA_DEC/ALT_AZ) plus offset_lon/offset_lat instead of four separate off_ra/off_dec/off_alt/off_az fields, and attempts track the 2D offset per iteration.

• ApplyOffsets (and its RA/Dec and Alt/Az subclasses) now return an OffsetResult with the actually-applied correction instead of a bare bool.

• IMode.set_mode now takes a group name instead of a positional index.

• Added IRunning support to IAutoFocus/DummyAutoFocus.

• Implemented script dispatch for the SCRIPT config type (LcoScript), selecting a nested script via extra_params.script_name.

• Fixed context (comm/observer/vfs) not being propagated to Portal, LcoScheduleReader/LcoScheduleWriter, and LcoTask when constructed directly, which broke proxy lookups.

• Fixed a missing request field causing LCO script validation to fail.

• Fixed BaseVideo never publishing IImageType state (unlike BaseCamera).

• Made the telegram import lazy in the Telegram module, so it’s no longer a hard dependency for unrelated modules in pyobs.modules.utils.

• Added a “What’s New in pyobs 2.0” docs page tracking user-facing changes for the 2.0 release.

v2.0.0.dev11 (2026-07-05)

• Implemented access control (ACLs) for module RPC calls, via an acl: config block next to comm: (allow/deny policy, enforce/log mode, IModule.get_permitted_methods()).

• Completed the Unit annotation rollout across all applicable interface signatures.

• Interface and event schemas, including versioning, are now fully published via service discovery (disco#info), enabling a mixed-version-fleet diagnostic.

• IAcquisition.acquire_target() now returns a typed AcquisitionResult instead of dict[str, Any].

• Module’s constructor no longer takes a name parameter; a module’s name always tracks its comm object’s own identity (XMPP JID / LocalComm name) instead.

• Added --syslog to pyobsd, forwarding it to the pyobs processes it launches.

• A module now shuts down gracefully instead of endlessly reconnecting when kicked from XMPP due to a JID conflict.

• Fixed an XMPP reconnect storm after an ejabberd outage, and a module reconnect that could be silently dropped by a stale presence callback.

• Fixed DummyTelescope.park() not stopping an in-progress slew.

• Fixed a crash when a cooling setpoint or config name is None.

• Fixed CRITICAL log lines being journaled with the wrong priority under --syslog.

• Hardened RPC parameter parsing (xml_to_params) against malformed input.

• Fixed phantom XMPP state subscriptions for composite interfaces.

• A global pyobs.yaml config file (including under /opt/pyobs/storage/) is now looked up in addition to a module’s own config file.

v2.0.0.dev10 (2026-07-02)

• Interface features are now version-tagged in service discovery, giving a diagnostic for mixed-version fleets.

• IAutoFocus.auto_focus() now returns AutoFocusResult instead of a tuple; added live AutoFocusState; removed the old auto_focus_status().

• Added OptimalFocusState for structured focus-model state tracking.

• IWeather migrated to structured live state; dropped station from sensor readings.

• IFitsHeaderBefore/IFitsHeaderAfter now return dict[str, FitsHeaderEntry] instead of dict[str, tuple[Any, str]].

• Removed the old XML-RPC cast pipeline (pyobs.utils.types).

v2.0.0.dev9 (2026-07-01)

• Linked all pyobs-* module docs, now hosted on docs.pyobs.org.

v2.0.0.dev8 (2026-07-01)

• Documented optional extras and CLI tools in the README.

v2.0.0.dev7 (2026-07-01)

• Minor internal fixes (Dependabot configuration, import scoping in DummyCamera).

v2.0.0.dev6 (2026-06-30)

• Renamed the state/capabilities proxy accessor methods to get_state/get_capabilities for clarity; both are now also available directly on Interface.

• Added IERS offline mode support via the PYOBS_IERS_OFFLINE environment variable.

• The pyobs service now loads environment variables from /etc/default/pyobs.

• Added astropy-iers-data as a direct dependency.

v2.0.0.dev5 (2026-06-30)

• Refactored capabilities handling for consistency across interfaces.

v2.0.0.dev4 (2026-06-29)

• Interface State classes are now module-level dataclasses rather than nested classes.

v2.0.0.dev3 (2026-06-29)

• Removed pyobs.utils.simulation (SimWorld/SimTelescope/SimCamera) with no replacement.

• Added DummyVideo for simulated video streaming.

• IMode now uses capabilities/state dataclasses instead of separate mode-group methods.

• Added an optional sender attribute to LogEvent.

v2.0.0.dev2 (2026-06-29)

• Rolled out live state to (almost) all state-bearing interfaces; removed the corresponding get_*/is_* RPC methods project-wide in favor of subscribing to state.

• Added a shared XML serializer for both RPC and state payloads (pyobs.comm.xmpp.serializer); rewrote the XMPP RPC layer on top of it.

• Added Proxy.wait_for_state (returns cached state immediately, or waits for the first update).

• Added capabilities — fixed-for-lifetime values published via service discovery, alongside the new state mechanism.

• A module’s online/ready/error status is now tracked via XMPP presence rather than RPC; removed IModule.get_state()/get_error_string().

• LocalComm gained state, capabilities, and presence support to match XmppComm.

• Removed ILatLon and the never-implemented DbusComm backend.

v2.0.0.dev1 (2026-06-22)

• Added version (default 1) to Interface and Event.

• Proxy is now obtained via async with self.proxy(...) as x: only; the long-lived await self.proxy(...) pattern and the cache_proxies option are removed. Added has_proxy/safe_proxy.

• Added Unit annotations for physical quantities on interface signatures.

• Added the first live state implementation over XMPP PubSub, piloted on ICooling.

[1.53.0] - 2026-06-19

• Replaced the threading-based XMPP client internals with asyncio throughout.

• Added --syslog to log to the systemd journal, for both pyobs and pyobsd.

• pyobsd now names each module’s log file after its config file (e.g. camera.yaml → camera.log) instead of a fixed name, and takes its default module name from the config file.

• Added ObservationState.WINDOW_EXPIRED.

• Added HelioprojectiveRadialTarget.

• The scheduler now guarantees a minimum scheduling window length and excludes the currently-running task from “is there a better task” checks.

• Made XMPP message sending more robust: queues messages, skips repeated ones, and checks whether the user is actually logged in.

• Fixed object-ownership tracking so an ObservationArchive is only registered as a child once.

• Various performance improvements (lazy imports to reduce memory consumption, a fast path for transit-window calculation).

[1.52.0] - 2026-06-13

• Added TransitImagingScript for imaging around a target’s transit.

• estimate_duration now takes task/time parameters and can compute full-transit durations.

• Added ExposureTimeProvider (and StellarExposureTimeProvider) for dynamic per-exposure timing in ImagingScript.

• Rewrote pyobsd to no longer depend on start-stop-daemon; prints subprocess stdout/stderr if a module fails to start.

• Added a GitHub Actions integration-test run triggered on release.

• Removed the tornado-based logger handling in favor of slixmpp’s own logging.

• The scheduler now only fetches pending and in-progress observations rather than everything.

[1.51.0] - 2026-06-09

• Script.can_run can now report a reason via cant_run_reason, surfaced by the mastermind when skipping a task.

[1.50.0] - 2026-06-09

• Moved the robotic storage backends (filesystem, HTTP backend, LCO) into their own pyobs.robotic.storage subpackage.

• Added an in-memory TaskArchive/ObservationArchive implementation, useful for testing.

• Added Comm.has_module.

• Time is now imported from pyobs.utils.time instead of astropy.time throughout.

• Added Constraint.filter_skycoord for faster constraint filtering.

• Fixed a proxy deduplication bug.

[1.49.0] - 2026-06-08

• Added a time parameter to ObservationArchive.get_schedule and get_current_observation across all backends (filesystem, HTTP backend, LCO).

• Log timestamps are now formatted as ISO 8601.

[1.48.0] - 2026-06-07

• Switched linting from flake8 to ruff and fixed the resulting warnings project-wide.

• Replaced broad except Exception catches with narrower exception handling in several places.

• Changed log calls to use lazy (%-style) argument evaluation throughout.

• Renamed ensure_feature to create_task.

[1.47.0] - 2025-06-07

• Set minimum Python version to 3.11.

• Replaced old-style type hints (Optional, Union, List, etc.) with modern Python 3.11+ syntax throughout.

• Replaced deprecated asyncio.ensure_future with asyncio.create_task.

• Replaced deprecated asyncio.get_event_loop with asyncio.get_running_loop.

• Replaced str, Enum with StrEnum throughout.

• Replaced asyncio.gather with asyncio.TaskGroup where applicable.

• Added ruff linting to CI and pre-commit.

[1.46.0] - 2025-05-27

• Added DynamicTarget class for runtime target selection via a pluggable Picker interface.

• Added CsvPicker for selecting targets from a CSV catalogue, with constraint-aware filtering.

• OnDemandScheduler now resolves dynamic targets before evaluating constraints and merits.

• Added direction parameter to SolarElevationConstraint for filtering on rising or setting sun.

• Resolved target is now stored on Observation and restored by the mastermind via fetch_task.

• Target resolution result is cached per scheduling run.

• Added set_resolved_target and estimate_duration to Task.

• LcoTaskRunner now correctly injects script before run_task.

• Added missing abstract method implementations to LcoTaskArchive and LcoObservationArchive.

• Fixed inverted PENDING filter in LcoTaskArchive.get_schedulable_tasks.

• Fixed LcoConfiguration.state default to "PENDING".

• Portal now creates aiohttp session in open() instead of __init__().

• Portal now managed via add_child_object for correct lifecycle.

• Fixed SolarElevationConstraint direction logic using observer.midnight.

• Added AstroplanScheduler tests; fixed empty block list causing subprocess hang.

• Fixed LcoRequest.location field name conflict with BaseModel.location.

• Added ImagingScript for standard science exposures with acquisition and guiding support.

• Comprehensive test suite additions for LCO classes, scheduler, and scripts.

[1.45.0] - 2025-05-14

• Added unified get_observations interface to ObservationArchive with state and time filters.

• IntervalMerit and PerNightMerit now push filters directly to get_observations.

• Added ObservationArchiveEvolution.get_observations method.

• Added min_safety_time parameter to Mastermind.

• Added active field to Task.

• Fixed telescope not stopping after auto focus.

[1.44.0] - 2025-04-24

• Renamed SubClassBaseModel to PolymorphicBaseModel.

• Renamed MeritScheduler to OnDemandScheduler.

• Moved utils.archive and utils.skyflats into robotic.utils as pydantic models.

• Moved serialization.py from robotic.utils to utils.

• Converted SkyFlatsBasePointing, SkyflatPriorities, Archive, and PyobsArchive to PolymorphicBaseModel.

• Removed Object from Script base classes.

• Context injection now propagates to all child pydantic models.

• HTTP requests now use pagination; all list responses wrapped in {"results": [...]} format.

• Added create_script method to Task.

• Added http_request_with_retries with tenacity-based retry logic.

• BaseModel no longer inherits from Object.

• Added trigger_on_every_update parameter to Scheduler module.

• Added Observation priority field.

• Major documentation overhaul for the robotic subsystem.

[1.43.0] - 2025-04-20

• Added http_request_with_retries utility.

• Backend archives now use http_request_with_retries instead of direct HTTP calls.

• Added ignore_cert_errors parameter to backend archives.

• Added filesystem-based backend for robotic scheduling (YamlTaskArchive, YamlObservationArchive).

• Fixed inject_class_on_serialization for pydantic models.

[1.42.0] - 2025-03-08

• Backend archives migrated from requests to aiohttp.

• Observation fetching moved to a background task.

• Task changes now driven by on_tasks_changed callback.

• Refactored robotic subsystem to support pluggable backends.

[1.41.0] - 2025-02-20

• Added optional Qt widget dependencies for GUI support.

• Added general-purpose Qt widgets.

[1.40.0] - 2025-02-10

• XMPP communication now connects with TLS using slixmpp 1.4.1.

• Fixed PyPy compatibility issues.

• Added AvoidMoon constraint.

• Added FromList grid filter.

• PerNightMerit now uses last sunrise for night boundary.

• Auto focus now uses actual measured focus position.

[1.39.0] - 2024-12-10

• Major scheduler refactoring: introduced pluggable scheduler architecture.

• Added OnDemandScheduler (merit-based greedy scheduler) with DataProvider context.

• Added ABORTED, IN_PROGRESS, and FAILED observation states.

• Added ConfigurationSummary to LCO integration.

[1.38.0] - 2024-11-05

• Added config file support via --config flag.

• Added InfluxDB logging handler.

• Added CommLoggingHandler guard against duplicate logger registration.

• Added --verbose CLI switch.

[1.37.0] - 2024-10-20

• Added InfluxDB logging handler (InfluxLoggingHandler).

[1.36.0] - 2024-10-05

• Improved exception handling in background tasks; BackgroundTask now receives parent reference.

• Unknown remote exceptions now wrapped in RemoteError.

• Added SSL check option for XMPP.

• Fixed: don’t reconnect after intentional XMPP disconnect.

• Added example systemd service file.

[1.35.0] - 2024-09-01

• Added --verbose switch to CLI.

• Added CALIBRATING motion status.

• Refactored shell command handling into dedicated classes.

[1.34.0] - 2024-08-20

• Refactored shell command and response into dedicated ShellCommand classes.

[1.33.0] - 2024-08-05

• Added new grid creation for pointing series.

• Configurable wait times in pointing module.

[1.32.0] - 2024-07-20

• Moved exceptions to a dedicated module.

• Added AcquisitionError for failed acquisitions.

• Changed GeneralError to FocusError for focus-related errors.

• Added 60s timeout to autoguider stop method.

• LCO integration: use ConfigDB to fetch instrument; support multiple configurations.

[1.31.0] - 2024-05-25

• Added documentation for image processors.

• Registered pyobs exceptions now logged as INFO without stack trace.

• Added overwrite parameter to image writing.

• Added Matrix chat client module.

[1.30.0] - 2024-05-01

• Added Matrix chat client module.

• Fixed context propagation for non-dict objects.

• Added reset method to BrightestStarGuiding.

• Added oneshot parameter to ScriptRunner.

• Added logging when a script cannot run.

[1.29.0] - 2024-03-15

• Added PipelineCamera for running image pipeline on camera images.

• Added Flip image processor.

• Added TypeScript interface export.

• Added HttpServer to serve images via HTTP.

• Added SolarHelioprojective image processor.

• Added Pipeline module.

[1.28.0] - 2024-02-20

• Added offset parameter to IGain.

• Reduced httpx logging verbosity.

[1.27.0] - 2024-02-01

• FilenameFormatter now supports formatting with FITS headers via GetFitsHeaders.

[1.26.0] - 2024-01-15

• Added comprehensive typing throughout using TypedDict and npt.NDArray.

• Improved get_object overloads.

[1.25.0] - 2024-01-01

• Migrated build system to uv.

• Pointing module updated to use new grid classes.

[1.24.0] - 2023-12-30

• Added new spherical grid classes (RegularSphericalGrid, filters, pipeline) with tests.

[1.23.0] - 2023-12-29

• Improved spilled light guiding sigmoid function for more accurate relative offset calculation.

• Added weather station workaround.

[1.22.0] - 2023-12-29

• Added auto_focus integration to relevant modules.

[1.21.0] - 2023-12-28

• Added init_offset_to_zero functionality.

• Spilled light guiding: added binning correction, image trimming, and improved calculations.

[1.20.0] - 2023-12-28

• VFS not created if none is configured.

• Added pixel offset image processor for simplified acquisition workflow.

• get_object no longer overwrites existing parameters.

[1.19.0] - 2023-12-28

• Added spilled light guiding processor.

• Added IMultiFiber interface.

[1.18.0] - 2023-12-28

• ITelescope no longer directly implements IPointingRaDec/IPointingAltAz; BaseTelescope handles dispatch.

• Added bright star guiding.

• Restore initial focus if focus series fails.

[1.17.0] - 2023-12-28

• Added DummyMode module and ModeChangedEvent.

• Added ResolvableErrorLogger utility.

• Reactivated publisher module.

• Added XMPP auto-reconnect on connection loss.

[1.16.0] - 2023-12-28

• New pyobsd daemon with improved module management and config testing.

• Added several default scripts.

• Added group support for module configuration.

[1.15.0] - 2023-12-28

• Added group support for module configuration.

[1.14.0] - 2023-12-28

• Stop telescope after autofocus completes.

• Updated to Python 3.12; replaced deprecated datetime.utcnow().

• Warn when no RA/Dec given for pointing.

[1.13.0] - 2023-12-28

• Added acquisition support for bright central star.

• Added derotator offset handling.

• Added BackgroundTask class to simplify background task management in Object.

• Added unit tests for BackgroundTask and Object.

[1.12.0] - 2023-12-28

• Added list command for pyobsd, which outputs all configurations.

• Added bash auto-complete script pyobsd.

• Added timeouts (to be defined in the config) for ScriptRunner modules.

[1.11.0] - 2023-12-25

• Acquisition and AutoFocus both got a broadcast option to disable broadcast of images.

• AutoFocus got a final_image parameter to take a final image at optimal focus.

[1.10.0] - 2023-12-24

• Added CallModule script.

• Changed ScriptRunner module so that it can run a script multiple times.

[1.9.0] - 2023-12-23

• Added getters and safe getters for Image class.

[1.3.0] - 2023-02-04

• Adopted LCO default task to new LCO portal.

[1.2.0] - 2022-10-06

• Added AltAzOffsets and RaDecOffsets and (partly) implemented them in the ApplyOffsets classes.

[1.1.0] - 2022-09-20

• Changed signature of pyobs.robotic.TaskSchedule.get_schedule to have no parameters.

[1.0.0] - 2022-09-13

[0.22.0] - 2022-08-25

• Removed comm.sleexxmpp implementation.

• Renamed comm.slixmpp to comm.xmpp.

[0.21.0] - 2022-08-25

• Some pipeline stuff.

• Added DbusComm for communicating via Dbus.

• Cleaned up parameter casting for communication.

[0.20.0] - 2022-06-22

• Some fixes with asyncio and the GUI.

• Handle JID conflicts in XMPP.

[0.19.0] - 2022-05-17

• Getter/setter methods in Module must be async.

• get_task() in TaskScheduler is now async.

• Lots of bug fixes.

[0.18.0] - 2022-03-13

• New IGain interface.

[0.17.0] - 2022-02-14

• Restructuring robotic system.

[0.16.0] - 2022-01-14

• Added new exceptions.

• Use those new exceptions to keep track of errors over time and raise SevereErrors.

• Add new state to module, so that a severe error can put a module into an error state.

• Added get_state() and get_error_string() methods to modules.

[0.15.0] - 2021-12-29

• Added Comm implementation for SliXMPP (which should now be default) and moved old comm.xmpp to comm.sleekcmpp.

• Using asyncio throughout the project, all method and event handlers are async now, as well as open/close methods.

• Got rid of multi-threading as best as possible.

• VFS now also uses asyncio.

[0.14.2]

• Fixed a bug with Poetry

[0.14.1]

• Added possibility to use class hierarchy for events, i.e. subscribe to a class and receive all derived events.

• Change to Poetry as build system

[0.14] - 2021-11-03

• Guiding modules accept a pipeline now, so more image processors than just Offsets can run.

• Renamed ICameraBinning, ICameraExposureTime and ICameraWindow and removed the “Camera” part.

• Added meta attribute (temporary storage, not I/O persistent) to Image.

• Extracted IImageGrabber from ICamera and renamed expose() to grab_image().

• Added new IVideo interface and a corresponding BaseVideo module.

• Raising exception, if XmppComm cannot connect to server, allowing for graceful exit.

• On shutdown, wait for hanging threads, and kill them after 30 seconds.

• Multi-processing for the pipeline, using ccdproc now.

• New interface IPointingSeries, giving access to methods at the telescope that support pointing series.

• Send logs in thread.

• Added concept of image processors that take an Image as parameter and return it after some processing.

• Added new NStarOffsets image processor (T. Masur).

• Improved scheduler.

• Added pipelines that take a list of image processors (see Pipeline mixin).

• Re-organized all get_object methods.

• Improved type hints throughout the code.

• Renamed all coordinated interfaces (IRaDec, etc) to IPointing*, i.e. IPointingRaDec.

• Renamed all offset interfaces to IOffsets*, i.e. IOffsetsRaDec.

• Renamed IFitsHeaderProvider to IFitsHeaderBefore and also renamed its only method.

• Added IFitsHeaderAfter to fetch FITS headers after an exposure as well.

• Moved functionality from Module to Object.

• New meta data system for images.

• Renamed IStoppable to IStartStop.

• Added new proxy interfaces in interfaces.proxies. All proxies now derive from these interfaces instead of the original ones.

• And a lot more cleanup and re-organization.

[0.13] - 2021-04-30

• Added a Telegram bot module.

• Added a module for a Kiosk mode, in which pictures are published on a webpage.

• Added new IImageFormats interface for cameras that support multiple ones (e.g. grayscale and color).

• Moved more enums into utils.enums, like WeatherSensors and MotionStatus.

• Added list_binnings() to IBinning interface and (temporary) default implementation in BaseCamera.

• Restructured image processors into pyobs.image.processors.

• Split photometry into separate SourceDetection and Photometry interfaces, added DaophotSourceDetection, and PhotUtilsPhotometry.

• Sending events non-blocking, which might solve some problems with disappeared XMPP clients.

• Added lots of documentation, which included setting __module__ for many classes.

[0.12] - 2021-01-01

• Changed PyObsModule to Module.

• Removed possibility for network configs.

• Added MultiModule, which allows for multiple modules in one process.

• Flat scheduler: add options for readout times.

• New OnlineReduction module for reduction during the night.

• Fixed bug that sometimes appears in the interface caching for Comm.

• LcoTaskArchive: added MoonSeparationConstraint, fixed AirmassConstraint.

• Optimized Scheduler by only scheduling blocks that actually have a window in the given range.

• Added module Seeing that extracts FWHMs from the catalogs in reduced images and calculated a median seeing.

• Introduced concept of Publishers, which can be used to publish data to log, CSV, and hopefully later, database, web, etc.

• Created new Object class that handles most of what Module did before so that Module only adds module specific stuff.

• Added some convenience methods for reading/writing files to VFS.

• Added new IConfig interface which is implemented in every module and allows remote access to config parameters (if getter/setters are implemented).

• Removed count parameter from ICamera.expose().

• Removed exposure_time parameter from ICamera.expose() and introduced IExposureTime interface.

• Removed image_type parameter from ICamera.expose() and introduced IImageType.

• Moved ImageType enumerator from ICamera to utils.enums.

[0.11] - 2020-10-18

• Major changes to robotic system based on LCO portal.

• Setting filter/window/binning in acquisition.

• Added WaitForMotion and Follow mixins.

• Added support for flats that don’t directly scale with binning.

• New module for acoustic warning when autonomous modules are running.

• Improved SepPhotometry by calculating columns used also by LCO.

• New interface for Lat/Lon telescopes, e.g. solar telescopes.

[0.10] - 2020-05-05

• Re-factored acquisition modules and added one based on astrometry.

• Added combine_binnings parameter to FlatFielder, which triggers, whether to use one function for all binnings or not

• Added get_current_weather() to IWeather

• New FlatFieldPointing module that can move telescope to a flatfield pointing

• Changed requirements in setup.py and put packages that are only required by a server module into [full]

• Removed HTTP proxy classes

• Some new mixins

[0.9] - 2020-03-06

• working on robotic system based on LCO portal

[0.8] - 2019-11-17

• Added module for bright star acquisition.

• Added and changed some FITS header keywords.

• Added module for flat-fielding.

• Changed some interfaces.

• Added basic pipeline.

• Started with code that will be used for a full robotic mode.

• Re-organized auto-guiding modules.

• and many more…