chore: migrate to new column-access mechanism (SDK 1.81.x)

[AStaroverov]

What

Update @platforma-sdk to 1.80.x and migrate the block model off the removed column-access APIs onto the new host-driven ColumnsCollection / ColumnRecipe mechanism.

Why

ColumnCollectionBuilder / ColumnMatch / ArrayColumnProvider were removed in the SDK. The new API discovers and filters columns host-side and fetches specs on demand (getSpec()) instead of pulling every upstream spec eagerly into the 8 MB model sandbox.

Key changes (model)

• ColumnCollectionBuilder + findColumns → ColumnsCollection(["result_pool"]).discover(...); ColumnMatch → ColumnRecipe (.column.spec/.column.id → .getSpec()/.id).
• resultPool.getPColumnSpecByRef(ref) → Column(ref).getSpec().
• Filtering pushed host-side into include/exclude selectors wherever expressible (linkers, label, cluster-id names, per-sample axis, type: "String", hideDataFromUi/Graphs); only non-expressible predicates (File type, "produced-by-lead-selection" last-trace-step) remain as JS post-filters.
• Main table createPlDataTableV3 display rules moved from (spec) => boolean lambdas to host-side ColumnSelectors.
• Annotation-key string literals replaced with Annotation.* constants where available.
• Catalog @milaboratories/helpers bumped 1.14.2 → 1.14.4 to match the SDK (fixes a TS2742 declaration-portability error).

No user-facing behavior change intended.

Validation

• type-check, lint, format, build green across model / ui / workflow / test (no unit/integration tests exist in the block).
• Needs a live-backend integration pass before merge: main table row/column set + rank column, workflow bundleBuilder.addSingle id compatibility (incl. persisted filter/ranking ids), and umap/pf pframe assembly.

Greptile Summary

Migrates the block model from the removed ColumnCollectionBuilder / ColumnMatch / ArrayColumnProvider APIs to the new host-driven ColumnsCollection + ColumnRecipe mechanism introduced in SDK 1.81.x, with the main table's (spec) => boolean display lambdas moved to host-side RelaxedColumnSelectors. No user-facing behavior change is intended.

Key touched terms:

Term	Definition	Change
ColumnsCollection	Host-side factory that discovers/filters columns from a result pool or accessor(s)	Replaces ColumnCollectionBuilder + AnchoredColumnCollection; now called with ["result_pool"] or a TreeNodeAccessor[] instead of constructing a builder
ColumnRecipe	Tagged union of column handle types returned by .discover().getColumns(); accessed via .getSpec(), .getData(), .id	Replaces ColumnMatch; .column.spec / .column.id access replaced by .getSpec() / .id
DataColumn	Subtype of ColumnRecipe for bare leaf storage columns; also provides DataColumn.getStatusByPlRef()	New — used with isDataColumn() to filter recipes for createPFrameForGraphs; status check replaces try/catch on ColumnAbsentError
ScopedColumnId	Wire format persisted in BlockArgs for filter/ranking selections	anchorName: string removed (was always "main", never read by the workflow); column type stays PObjectId via extractPObjectId
ColumnsMeta.allMatches / scores	Metadata arrays for discovered columns	Type updated ColumnMatch[] → ColumnRecipe[]
getSpecByRef	New helper wrapping Column(ref)?.getSpec() with an absent status guard	Replaces scattered ctx.resultPool.getPColumnSpecByRef(ref) calls across the model
discoveryExcludeSelectors	New function returning the host-side exclude selectors for filter/rank discovery	Supersedes commonExcludeSelectors; adds label, cluster-id, and per-sample axis excludes pushed host-side
dedupByLeafId	New function collapsing multiple recipe variants for the same leaf column to one entry	Restores pre-migration dedup that findColumns() did internally; required because discover().getColumns() returns one recipe per reachability variant
buildCollection	Shared collection builder for filter, rank, MSA, and table outputs	Signature drops ctx: RenderCtx argument (relies on SDK ambient context); now exposes anchorSpec directly; returns ColumnsCollection(["result_pool"]) instead of an AnchoredColumnCollection

• BlockArgs shape change: ScopedColumnId.anchorName dropped → existing projects re-run the workflow once on next open; stored BlockData keeps the field as an inert leftover.
• Two concerns remain open (flagged in earlier review threads): the ColumnsCollection() no-arg scope used for poolDiscovered in the main table output vs ColumnsCollection(["result_pool"]) everywhere else, and the missing File value-type guard in isSelectableMatch.
• The changeset notes that a live-backend integration pass is needed before merge (main table row/column set, bundleBuilder.addSingle id compatibility, umap/pf pframe assembly).

Confidence Score: 5/5

• This PR is a mechanical API migration with well-documented intent and no user-facing behavior change; the core logic paths are preserved and the author has verified the build and type-check pass.
• The migration faithfully replaces each removed SDK call with its documented successor. Wire formats (ScopedColumnId.column stays a PObjectId leaf, anchorRef preserved) are carefully maintained for workflow and UI compatibility. The two concerns noted in earlier threads are acknowledged in the changeset and PR description as known follow-ups requiring a live-backend pass before merge. The one remaining raw annotation string at line 449 is a style inconsistency with no runtime impact. No logic regressions are introduced by the changed files.
• model/src/index.ts and model/src/util.ts warrant a live-backend integration pass (main table column set, filter/ranking dropdowns, umap pframe assembly) before merge, as noted in the PR description.

Important Files Changed

Filename	Overview
model/src/index.ts	Core model output definitions migrated from old ColumnCollectionBuilder/findColumns/ArrayColumnProvider to ColumnsCollection/discover/ColumnRecipe. Display rules for the main table moved from JS lambdas to host-side RelaxedColumnSelectors. Two open concerns flagged in prior threads: ColumnsCollection() (no-arg) scope for the main table's poolDiscovered vs ColumnsCollection(["result_pool"]) used elsewhere, and the missing File-type guard in isSelectableMatch.
model/src/util.ts	Utility layer migrated: ColumnCollectionBuilder removed, getSpecByRef added as a safe PlRef→PColumnSpec wrapper, discoveryExcludeSelectors replaces commonExcludeSelectors, dedupByLeafId restores pre-migration dedup semantics, isSelectableMatch simplified to only the JS-only predicates. One raw annotation string remains at line 449 (Annotation.Score.RankingOrder missed).
model/src/types.ts	ScopedColumnId.anchorName removed (was always "main", workflow never read it). ColumnsMeta.allMatches/scores type updated ColumnMatch[]→ColumnRecipe[]. Dead Column/ScopedColumn types removed. Clean migration with clear JSDoc explanations.
ui/src/composables/useAnchorSyncedDefaults.ts	No logic changes; ScopedColumnId.anchorRef is still present so anchor-staleness detection in configAnchorKey is unaffected by the anchorName removal.
.github/workflows/build.yaml	Adds require-package-path-bump: true to enforce changeset bumps on PRs. Straightforward workflow hardening.
pnpm-lock.yaml	SDK bumped 1.79.x → 1.81.x across model/ui/workflow/test; helpers 1.14.2 → 1.14.5; ts-builder 1.6.0 → 1.6.2; ts-configs 1.3.0 → 1.4.0; block-tools 2.12.7 → 2.12.13. Lock file updated consistently.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[RenderCtx output evaluation] --> B[buildCollection]
    B --> C["ColumnsCollection(['result_pool'])"]
    C --> D["discover({ anchors, mode: 'related', maxHops: 2 })"]
    D --> E[dedupByLeafId]
    E --> F[isSelectableMatch]
    F --> G[allMatches / scores]
    G --> H[computeDefaultFilters]
    G --> I[computePresets]

    A --> J[filterConfig output]
    J --> K["result.collection.discover({ anchors, exclude: discoveryExcludeSelectors })"]
    K --> L[dedupByLeafId]
    L --> M[isSelectableMatch]
    M --> N[filterableMatches → options]

    A --> O[rankingConfig output]
    O --> P["result.collection.discover({ anchors, exclude: [..., type:'String'] })"]
    P --> Q[dedupByLeafId]
    Q --> R[isSelectableMatch]
    R --> S[rankableMatches → options]

    A --> T[table output]
    T --> U["ColumnsCollection() — no-arg"]
    U --> V["discover({ anchors: leadSelectionSpec })"]
    V --> W[filter isProducedByLeadSelection]
    W --> X[primaryColumns / secondaryColumns]
    X --> Y[createPlDataTableV3]

    A --> Z[umapPf / umapPcols]
    Z --> AA["ColumnsCollection(['result_pool']).discover(umap name pattern)"]
    AA --> AB[toGraphColumns → createPFrameForGraphs]

Loading

_{Reviews (2): Last reviewed commit: "refactor: guard on column status instead..." | Re-trigger Greptile}

Context used:

• Context used - Terms is a types in codebase. Provide the list of ... (source)

[gemini-code-assist]

Code Review

This pull request updates @platforma-sdk to 1.80.x and migrates the block model to the new host-driven column-access mechanism, replacing the old ColumnCollectionBuilder and ColumnMatch APIs with ColumnsCollection and ColumnRecipe. The review feedback correctly identifies a critical bug where ColumnsCollection is instantiated without sources, which would break main table rendering, and notes the accidental removal of the assemblingKabatPf resolver. Additionally, a defensive programming improvement was suggested to safely access axesSpec[0]?.name to prevent potential TypeErrors.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[AStaroverov]

@greptileai