CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

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

[4.0.1]

Added

• Alyx-API-Notice header can be used to attach information for ONE users

Changed

• Detail key no longer included in pagination response if user likely on unsupported ONE version

Fixed

• Breeding pair parent no longer cleared when saving Line admin changeview

[4.0.0]

Added

• Public databases can offer self-registration, enabled with the PUBLIC_DATABASE lab setting. Adds a sign-up page at /signup, email confirmation of new accounts, and a password reset flow so that self-registered accounts are recoverable without an administrator
• set_public_permissions management command, creating a 'Public users' group carrying view permissions only, and never over user accounts or the permission structure itself
• ec2_modify_groups.sh script now included in container to remove EC2 firewall before letsencrypt validation

Changed

• Public users are no longer shown other people's accounts. Over REST, /users returns redacted users and the requester only, without email addresses; in the admin, the user and group models are hidden and user filter dropdowns no longer enumerate every account
• set_user_permissions skips public users, which it previously added to the lab members group and marked active - granting members of the public write access and activating accounts that had never confirmed their email address

Removed

• Core-api support for the /docs endpoint. SpectacularRedocViewCoreAPIDeprecation and data/coreapi.json are gone; /docs now serves SpectacularRedocView directly regardless of the request's Accept header. #1055

Fixed

• Subjects in different labs may share a nickname. REST endpoints taking a subject nickname returned a 500 status when it matched more than one subject; the lab is now used to resolve duplicates - the lab field of the request data, ?lab= on /subjects/<nickname> and /water-requirement/<nickname>, or labs and the repository's labs for /register-file. An unresolved duplicate returns a 400 status, or 409 for a lookup, naming the labs. /subjects/<id> also accepts the subject UUID.
• Open water restrictions are ended when a subject already has a death date, e.g. when a cull is added with the same date, not only when the death date is first set. The subject's protocol number is now updated when its water restrictions are ended.
• Relative paths correctly rendered in cache table using pandas 3.
• The data notice admin change page no longer renders the datasets as a multi-select widget: every attached dataset had to be fetched and rendered as a selected option, which did not scale past a few hundred, and saving posted one form field per dataset, which exceeded DATA_UPLOAD_MAX_NUMBER_FIELDS and failed with a 400 status. The datasets are now a read-only, scrollable list of the first 100, with the total count and a link to them in the dataset list. They can still be attached when creating a notice, or through the REST API.

[3.6.4]

Fixed

• Sessions and probe insertions tag REST filters no longer join the datasets table into the main query, which made them 3-174x faster.
• tag filter on the fields-of-view REST endpoint. FOVFilter had a filter_tag method but never declared the filter that routes to it, so /fields-of-view?tag= was silently ignored.
• dataset_qc_lte REST filter returned sessions and insertions once per matching dataset, so the rows repeated across the paginated response and the count was the number of (row, dataset) pairs: /sessions?dataset_qc_lte=WARNING reported 3,234,065 of 94,594 sessions.
• atlas_name, atlas_acronym and atlas_id REST filters returned a session once per insertion recording the region, and likewise an imaging stack once per slice.
• The dataset_types, datasets, dataset_qc_lte and atlas_* REST filters now count and match in a subquery instead of joining and grouping over the full row, making them up to 3.5x faster (/sessions?datasets= 55 s to 16 s, /chronic-insertions?atlas_acronym= 0.7 s to 0.01 s). /insertions?tag=&dataset_qc_lte= previously timed out, as the two joins multiplied.
• The datasets REST filter counted the matching datasets rather than the distinct names on the insertions and fields-of-view endpoints, so a row holding two datasets of one requested name passed as having two different ones: /insertions?datasets= now agrees with /sessions?datasets=. One name matches several datasets across collections and revisions for 142,624 (session, name) and 50,078 (insertion, name) pairs, so this over-returned by ~4x on affected queries.
• A dataset name repeated in a datasets REST filter, e.g. ?datasets=obj.attr.npy,obj.attr.npy, returned nothing at all rather than the rows having that dataset.
• The chronic insertions REST list eagerly loaded its nested probe insertions while serialising each row, which discarded the prefetched result and cost two queries per chronic insertion: /chronic-insertions issued 329 queries for 176 rows, and now issues 4 regardless of the number of rows (1.7x faster overall, 2.3x less time in the database).
• The projects of a session, and the probe insertions of a chronic insertion, are now serialised in a defined order. Project has no Meta.ordering and insertions of one chronic insertion often share a session start time and name, so the database was free to return either in any order.
• The sessions, insertions, fields-of-view and chronic insertions REST lists are now totally ordered, so a page is the same from one request to the next. 916 sessions share a start time with another and 120 insertions share a session start time and name, and rows tied that way came back in a different order each request; /chronic-insertions had no ordering at all. The keys are now (-start_time, number, subject, pk) for sessions and (-session start_time, name, subject, session number, pk) for insertions and fields of view, the primary key being what makes them total: Session and ChronicInsertion declare no uniqueness, and the insertion constraint is on (session, name), which does not separate two sessions sharing a start time.
• /imaging-stacks returned a stack once per slice, serving 250 rows covering 162 of the 174 stacks while reporting 174, as the list was ordered by slices__name and so joined the slices into the query. The stacks are ordered by their own fields now and the slices within each stack by the prefetch.
• Admin no longer fails to start with djangorestframework >= 3.18

Added

• dataset_qc_lte filter on the fields-of-view REST endpoint, matching the sessions and probe insertions endpoints. The datasets filter there now applies the QC bound too, as it does on those endpoints, rather than ignoring it.
• Index on data_dataset.name, which the ?datasets= REST filters look up and which had no index (23 MB; deduplicated, as a few hundred names repeat across the table). Built concurrently, as the table is continuously written. /insertions?datasets= is 7x faster, but note /sessions?datasets= with two or more names is ~1.3x slower, as the planner switches from a parallel sequential scan to a bitmap scan past roughly 5% of the table.
• A single dataset name in the sessions datasets REST filter is now matched with one Exists semi-join anchored on the session, rather than gathering every dataset of that name in the table to group them: /sessions?datasets= with one name is 1.8x faster (9.9 s to 5.4 s), and 8x faster than before this release. Several names, or a dataset_qc_lte below WARNING, keep the grouped count, both being cases where the semi-join measured no better or worse. The insertions and fields-of-view filters are not special-cased, measuring the same either way.

Changed

• The custom admin site is now installed through alyx.apps.AlyxAdminConfig instead of by reassigning django.contrib.admin.site. Deployments must replace 'django.contrib.admin' with 'alyx.apps.AlyxAdminConfig' in their INSTALLED_APPS, otherwise the stock Django admin index is served in place of the Alyx one. alyx.base.mysite is gone; use django.contrib.admin.site, which now refers to the site Alyx serves.
• Groups are administered through django.contrib.auth's GroupAdmin rather than a bare ModelAdmin, so group permissions get the two-pane selector

[3.6.3]

Fixed

• subjects.default_responsible selects the primary key rather than the whole user record, so that migrating from scratch is not broken by columns added to LabMember in later migrations
• No longer required to quote special characters in database connection settings

[3.6.2]

Fixed

• Scoped delete_zygosity_rule post delete hook, improving performance on delete

Changed

• Removed unused dj_database_url dependency
• Omit WSGI from CI code coverage
• Changed default docker project name
• Use commit message for release description on squash commits to master
• Pin ruff and remove flake8

Added

• DOI badge in README
• Release steps to README

[3.6.1]

Fixed

• Correct relative paths from aggregate datasets with revisions, registered via REST
• Added missing migration file

Changed

• Workflow tests now run in parallel and asserts no new migrations

[3.6.0]

Added

• DataNotice model to attach information/notices to datasets (#1007)
• Registered-file validation (#1014)
• CI/CD release pipeline: bumping __version__ on master cuts a git tag + GitHub release and builds/pushes the alyx_apache[_base] docker images, gated behind an ansible smoke test (#954)

Changed

• The production docker image and compose are now built from this repository (deploy/app/) as the single source of truth; iblalyx is no longer baked into the image (bind-mounted at deploy time), and deploy orchestration (ansible, per-server overrides) lives in iblsre (#1017)
• Restructured deploy/ into app/ (production) and editable/ (postgres-only for editable installs)
• Bump Django 5.2.14 → 5.2.15 (#1012) and Pillow 12.2.0 → 12.3.0 (#1019)

Fixed

• Remove unknown dataset type from the test dump fixture (#1021)
• Flaky task-cleanup test (save within the datetime mock context)

[3.5.1]

Fixed

• to be culled filter in cull subjects admin

[3.5.0]

Modified

• support registration of datasets unassociated to a session
• sort insertions by name
• use ONE Globus class for transfers

Added

• user REST request rate limits

[3.4.2]

Fixed

• error in register-files endpoint when labs is a list

[3.4.1]

Modified

• added actual severity column to SubjectCullAdmin
• renamed 'Mice' to 'Subject' in views

Fixed

• all alive filters depend on death date instead of cull
• django 5.1 deprecation: CheckConstraint check -> condition
• various filter typos, e.g. 'To be reduced' filter now works
• improvements to notifications performance
• fix log typo in delete_expired_notifications management command

[3.4.0]

Modified

• moved prune_cortexlab.py to iblalyx repository

Fixed

• removed test for removed subject death save logic
• fixed command for dumping test database fixtures

Added

• in alyx.misc the one_cache command module contains utils to generate cache dataframes from sessions and datasets querysets

[3.3.3] 2025-12-04

Fixed

• default implant weight is 0 and not none to allow plotting of water curves.
• water restriction admin form makes sure the current subject is selectable even if it is not alive anymore.

[3.3.2] 2025-11-26

Fixed

• water history plots: the weight thresholds is (w - iw) / (ew - iw), where w is the measured weight, iw the implant weight and ew the expected weight. Expected weight is (ew = (rw * a + zw *b) + iw (weighted sum of reference weight and zscore weight + implant weight). The display was computing thresholds according to w/ew , not taking into account the implant weight and is now fixed.

[3.3.1] 2025-11-06

Changed

• The documentation endpoint /docs is only for the browser and uses openapiv3. The database schemes are accessed through /api/schema. For compatibility, if the headers require coreapi, the endpoint returns a frozen set of endpoint to the client. #929
• narrative template is now available on the base action instead of only for surgeries #938

Removed

• coreapi dependency is removed