v1.22.3

What's Changed

• handle deprecations by @Clare72 in #53
• Merge synonyms by (scope, label) with combined refs by @Robbie1977 in #54

Full Changelog: v1.22.2...v1.22.3

v1.22.1 - Fix duplicated term-info cache writes

Fix duplicated term-info cache work

get_term_info (and get_template_roi_tree) were double-wrapped with
@with_solr_cache: the *_cached wrappers carried the decorator while also
delegating to an original that is itself decorated with the same bucket. Every
request therefore ran the SOLR cache layer twice — two reads and two writes
for the same key (the duplicated Cached term_info … log lines and doubled Solr
writes after v1.22.0).

These were the only two double-wraps (every other _cached wrapper relies on its
decorated original). Removed the redundant decorators; for template_roi_tree
also forwarded force_refresh, which the outer decorator previously consumed.

No behaviour change beyond halving cache work and Solr writes per term-info
request.

Full changelog: v1.22.0...v1.22.1

v1.22.0 - Term-info parity, coverage & targeting queries

Term-info parity, full field coverage, and targeting queries

Moves the v2 term-info side panel onto VFBquery's get_term_info with display
parity against the legacy SOLR-field path, completes field coverage against the
VFB_json_schema_indexer spec, and adds two live query types.

Restored term-info fields (parity with the legacy panel)

• Class definition references (def_pubs) — inline microref links on the
  description (was dropped).
• Individual-term synonyms (pub_syn) — synonym block was Class-only.
• Publication external content (pub_specific_content) — title / PubMed / DOI /
  FlyBase; the block was gated on Publication but the SOLR marker is pub.

Complete field coverage vs the indexer

• External DB cross-references → Xrefs (e.g. Insect Brain DB, gene FlyBase).
• related_individuals / term_replaced_by → Meta.RelatedIndividuals.
• DataSet and License homepage link + logo → Meta.Link / Meta.Logo.

New live query types (preview + count badge)

• SplitsTargeting — splits that target a neuron class.
• TargetNeurons — neurons targeted by a split class.
• Cypher mirrors the indexer neuron_split / split_neuron clauses, served from
  pdb and cached; surfaced as queries rather than static fields.

Robustness

• Term-info no longer 5xx/stalls on cold-miss terms (e.g. a License individual):
  the SOLR cache write defaults to a soft commit so a wedged IndexWriter cannot
  stall the request. Override with VFBQUERY_SOLR_WRITE_COMMIT=true.
• fill_query_results skips the limit=-1 re-run when a preview is not
  saturated, cutting cold latency on query-heavy SuperTypes.

Tests

• src/test/test_term_info_parity.py (14) covers every field gap, the License
  smoke test, and the targeting queries.

After release, warm the v3-cached layer:
owlery-cache-reload --only "V3 term info" --force-refresh

Full changelog: v1.21.1...v1.22.0

v1.21.1

What's changed

• Gzip-compress Solr result-cache payloads with a 100 MB cap on compressed size, so oversized entries (e.g. AllAlignedImages, ~383 MB raw) can now be cached; adds defensive cache decode and self-healing of corrupt cache entries (#49)
• Fix graph_builder's single-query-term expectation; increase AllAlignedImages timeout (#48)
• CI: restrict push trigger to main/dev to avoid duplicate runs on PRs

Full changelog: v1.21.0...v1.21.1

v1.21.0

VFBquery v1.21.0

✨ Highlights

Class-level connectivity queries now return a per-subclass breakdown and use
VFB_connect-consistent normalization. Plus a round of caching/versioning
hardening so cache behaviour is correct across code versions and CI never
pollutes the production cache.

🔌 Connectivity by class (get_upstream_class_connectivity /
get_downstream_class_connectivity)

• Per-subclass results. Results now cover the queried class and each of its
  subclasses that has connectivity instances — the queried term's rows first,
  then one block per subclass. The queried (sub)class for each row fills the
  upstream_class (downstream query) or downstream_class (upstream query) slot,
  so every row carries both upstream_class and downstream_class (queried class +
  partner), matching the v2 table layout. A query_id field is included for
  grouping rows by queried (sub)class.
• Presynaptic normalization (matches VFB_connect). total_n / connected_n /
  percent_connected now describe the presynaptic (source) side of each
  connection — the queried class for downstream queries, the partner class for
  upstream queries — consistent with VFB_connect.get_connected_neurons_by_type.
  Validated against VFB_connect: identical partner set and matching
  pairwise_connections/total_weight (~99%; residual differences are
  data-snapshot drift).
• Column names are unchanged (total_n, connected_n, percent_connected,
  pairwise_connections, total_weight, avg_weight), so the v2 frontend renders
  without changes. Subclass enumeration reuses Owlery's cached subclass closure
  (no extra Neo4j round-trip).

▎ ⚠️ Behaviour change: these queries now return more rows (subclass blocks),
▎ populate both class columns, and total_n/connected_n are
▎ presynaptic-normalized. Consumers that assumed a single block per query or
▎ queried-class-normalized counts should review.

🗄️ Caching & versioning

• Cache invalidation is now monotonic. A cached entry is only invalidated when
  it was written by an older version; a newer entry is treated as a miss but
  never deleted. This fixes cross-version cache thrashing and stops an
  older/stale client (or a downgrade) from purging fresher entries.
• No cache writes from PR checks. VFBQUERY_CACHE_ENABLED=false is now a true
  full bypass (no read/write/purge/cache-server contact), and a new
  VFBQUERY_CACHE_READONLY=true allows reads while suppressing all writes/purges.
  CI uses these so PRs validate live queries and never modify the shared
  production cache.
• Single source of version truth. The package version now lives only in
  src/vfbquery/_version.py; setup.py, vfbquery.version, and the cache's
  version stamp all derive from it, and releases bump just that file (and commit
  it back to main). Fixes the previous drift between setup.py and init.py.

🐛 Fixes

• Term-info downloads links now use the absolute
  https://virtualflybrain.org/data/... host (consistent with filemeta); removed
  a dead duplicate get_secure_data_url.
• Docker image version-verify now reads the version from _version.py (the
  single source), fixing a false "version mismatch" on build.

🔧 CI / internal

• Performance-test job timeout raised and thresholds adjusted for the live,
  uncached path; cache is writable on push-to-main/scheduled runs (read-only
  only on PRs) so those runs warm the cache.
• New docs: RELEASING.md, expanded CACHING.md (bypass/read-only flags +
  version invalidation).

v1.20.0 - TermsForPub expression refs + Reference type column

1.20.0 — TermsForPub: expression-cited terms + Reference type column

TermsForPub only followed has_reference edges, so a publication cited solely
via the overlaps/part_of relationship pub property (the AnatomyExpressedIn /
TransgeneExpressionHere model) returned nothing — e.g. FBrf0232433 (VT-GAL4
expression data) is cited on ~63k expression edges covering ~5.9k patterns but
has zero has_reference edges.

TermsForPub now lists terms from both mechanisms, deduplicated per term, and
adds a Reference type column: Reference for has_reference sources,
Expression for overlaps/part_of expression patterns (joined if a term is
cited both ways).

The minor bump from 1.19 rotates the major.minor solr_result_cache key.

Validated read-only against prod pdb: FBrf0232433 -> expression patterns (was
empty); FBrf0242477 -> Reference datasets + Expression GAL4 lines.

v1.19.0 - cache all handler-reachable queries

1.19.0 - cache all handler-reachable queries

Completes the persistent SOLR result-cache coverage: every query-result
function reachable from the HA API handlers is now cached, instead of only a
subset. Adds composite-key persistence for the cross-dataset
query_connectivity, and moves the cache to the dedicated query-cache Solr
(vfbquerycache.virtualflybrain.org:80) with a VFBQUERY_SOLR_URL override.

Coverage sweep: 12 uncached handler entry points -> 0. Cache-hit proofs show
46x-2790x cold->warm speed-ups across the newly-cached query types. The minor
bump rotates the major.minor cache key.

v1.18.0 - clusterExpression Function column

1.18.0 — clusterExpression Function column

The migrated "Genes expressed in " (clusterExpression) query did not
surface the gene Function column that v2 prod shows. Gene molecular-function
categories (e.g. Enzyme, Small_molecule_binding) are Neo4j labels on the Gene
node, so Function is now derived from labels(g), excluding the structural and
availability labels (Entity, Class, Individual, Gene, Feature and any has*
facet) and sorted. Added alongside the existing columns; the reverse
expressionCluster query is unchanged.

The minor bump from 1.17 rotates the major.minor solr_result_cache key,
flushing cached clusterExpression results so the Function column appears.

Validated read-only against prod pdb: 128up -> "Enzyme; Small_molecule_binding".

v1.17.0 - carousel up to 5 example images

1.17.0 — carousel up to 5 example images for class/dataset rows

AlignedDatasets, AllDatasets, AnatomyExpressedIn and TransgeneExpressionHere
each bound a single representative example image (LIMIT 1) into the Images
cell, so a class/dataset row showed only one thumbnail even when it has many
examples. These four queries aggregate one-to-many example individuals into a
single row, so they now collect up to 5 examples and emit them as a
' | '-joined [![alt](url 'title')](ref) list -- the carousel shape the V2
imageMarkdownToVariableJson splits into a SlideshowImageComponent.

Per-individual queries (instances, templates, NBLAST similarity, connectivity
partners, dataset image listings, terms-for-pub) are one-image-per-row by
design and are unchanged.

The minor bump from 1.16 rotates the major.minor solr_result_cache key,
flushing the single-image cells cached under 1.16.0.

Validated read-only against prod pdb: AlignedDatasets=5, medulla=4, EP=2.

v1.16.0 - rotate cache version to flush stale entries

1.16.0 — rotate cache version to flush stale entries

Minor-version bump from 1.15.x. The solr_result_cache key is keyed on the
major.minor package version, so moving 1.15 -> 1.16 invalidates every cached
query result. This forces fresh output that includes the accumulated 1.15.x
query-output fixes:

• Dataset Reference column linked to publications (FBrf / DOI)
• Term labels for all link columns (template keeps symbol)
• Per-class Images carousel (all example images)
• /reports permalinks, Template anatomy fix, Type-column links, Gene names

No code change beyond the version.