Skip to content

Rebuild grid extraction using shared UI and canonical helpers - #2536

Open
michielbdejong wants to merge 5 commits into
codex/shared-ui-extractionfrom
codex/tablepage-extraction-only
Open

michielbdejong wants to merge 5 commits into
codex/shared-ui-extractionfrom
codex/tablepage-extraction-only

Conversation

@michielbdejong

@michielbdejong michielbdejong commented Nov 16, 2026 •

Copy link
Copy Markdown
Contributor

Ready for trekmeester's review as a bounded extraction foundation. Full Main CI is still a merge prerequisite. The fresh manually dispatched run is https://github.com/ontola/atomic-server/actions/runs/44264955689; runner selection passed, but the self-hosted Main and downstream jobs are queued. Dedicated checks were green at code head cbaa6c1; the latest 221d161 commit updates the review plan only, with no runtime changes. Combined code verification is recorded in batch bd4fa0e, whose Table and Shared UI lanes pass; full Main CI remains pending. The incomplete whole-TablePage port is separate follow-up scope, not a claim of this PR. No merge will be performed by this worker.

The original grid extraction copied UI helpers and was followed by a mixed stack of extraction and functionality changes. Rebuild its code-move foundation on #2535, preserving existing Atomic behavior and using the same @tomic/ui implementations as the rest of the Data Browser.

Move the existing grid, selection/keyboard/clipboard code and primitive cell editors into private @tomic/table; leave compatibility exports and the existing Atomic display/host adapters. Move EventManager and civil-date helpers into one dependency-free private build input, @tomic/utils. SDK and UI distributions include the compiled helpers, so SDK consumers gain no React dependency or private runtime dependency. Consume table/UI from the pinned submodule; no npm publication is needed.

This adds 304 net frontend production lines relative to #2535 (513 across both clean foundation PRs), excluding tests, scripts, declarations and configuration. Existing keyboard/editor state ownership is retained. Optional host-owned raw-text drafts, saved-view controllers, write queues/receipts, retries and changed validation are deferred. The complete existing TableViewTabs stays in the host until it moves with the complete page.

This is preparation, not the completed TablePage port. The page orchestration, built-in alternate renderers and remaining shared controls still need extraction. The ELFA submodule pin and native app are unchanged; record-editor/resource binding design is a separate discussion. This branch replaces the grid foundation in #2434 without rewriting that branch or its descendants. All nine legacy mixed PRs now have structural move/improvement splits; the originals are closed with history preserved. Those replacement pairs still retain legacy ancestry. Move-only restacking onto this clean foundation is now underway; improvements remain separate and require their own subsequent restacking. The current inventory is in planning/tablepage-extraction-reset.md.

Validation:

  • 14 table tests and 9 pure-helper tests pass; 580 affected host tests passed, followed by 23 focused editor/cell regressions after removing the optional draft improvement.
  • Table and Data Browser types, staged-snapshot workspace lint, and clean React compiler checks for the changed primitive editors; root Clippy and formatting checked during foundation validation.
  • Packed table/UI install, render and typecheck in a clean consumer, including shared CSS-variable identity. Packed SDK works in ESM/CJS and typed consumers without @tomic/utils installed.
  • Fresh real WASM and production app build; the actual Flutter embed also builds after deleting helper dist outputs and passes its size budget (981.1 kB gzip first-use / 1000 kB budget). Four Dagger CI regression checks pass. Fixed-list embed, Tauri, Dagger/static, cloud/demo and SDK-release builds now include workspace dependency graphs.
  • real browser grid/context menu/column-edit readback, mobile dialog and Escape dismissal pass without page errors. Captures are outside Git at /workspace/artifacts/table-extraction-only; desktop/mobile screenshots were shown in the task chat.
  • SDK publint passes. Its attw check crashes identically on the unchanged SDK baseline; that existing tooling failure is not claimed green.

Add an independent Table package CI lane with packed-consumer validation; retain full app integration in Main CI. Full CI remains outstanding; remaining whole-page extraction is tracked separately.

Tracks #2484 and the agreed response to Joep.

This was referenced Nov 16, 2026

Copy link
Copy Markdown
Contributor Author

The nine legacy mixed PRs have been structurally separated and their move-only parents are now ready for trekmeester’s content review.

Original mixed PR Code-move review Separate improvement review
#2500 schema choices / float editor #2539 #2540
#2506 stored-view codecs #2541 #2542
#2508 query composition #2543 #2544
#2511 filter chips #2545 #2546
#2519 precision / date formats #2547 #2548
#2527 quick-add controls #2549 #2550
#2528 action configuration fields #2551 #2552
#2531 create-button configuration #2553 #2554
#2533 row-selection controls #2556 #2557

For every pair, the improvement child includes the move parent in its ancestry and its complete Git tree was verified identical to the original mixed head. The original mixed PRs are closed as superseded, with original branches and review history retained. All 18 replacement heads had green dedicated Table package and Flutter checks at this audit; individual descriptions record the prior local checks.

Reviewable does not mean merge-ready. Each pair deliberately retains its legacy base so the move-versus-improvement boundary is reviewable. The move parents still inherit older state-management work and must be restacked onto the clean extraction and pass full Main integration before merging. They are not prerequisites of #2535/#2536 in their current form. Improvement children are separate follow-ups; their review and validation proceeds independently.

No merges or speculative reviewer requests were made. Whole-TablePage orchestration/renderers remain unextracted, ELFA’s submodule pin remains unchanged, and persistence/binding semantics are tracked separately in #2538.

Copy link
Copy Markdown
Contributor Author

Clean-base restack review map (18 November):

Move PR Review base Head
#2539 format controls / Float #2536 e48821d
#2541 configuration codecs #2536 5c35267
#2545 filter chip #2541 9ca3698
#2547 precision / date format #2536 912992d
#2549 quick-add controls #2536 273c6f5
#2551 action fields #2536 9c9b5e9
#2553 create-button fields #2536 b25afcf
#2556 row-selection controls #2536 5136f63

The independent CI policy #2497 is also restacked on #2536 (faa6ba0), with 14 existing tests and its mandatory hook passing. The eight component/code moves above are combined at bd4fa0e on codex/tablepage-clean-restack-validation for full integration CI. Export-list/type-wrapper/coverage conflicts are resolved there without adding behavior. Full application types, UI/table builds and table types pass, as do 39 package tests and 107 host regressions under UTC, including actual WASM formula/aggregate tests. Real local-only TablePage SDK/OPFS desktop and touch-mobile checks pass with row selection and no page errors; actual captures were shown in chat and remain outside Git at /workspace/artifacts/tablepage-clean-restack-combined.

Each original move head was published under codex/legacy-split-<original-number>-move BEFORE rewriting its review branch, and its improvement child was retargeted to that identical preserved base. Improvement code and review diffs remain separate; they still need later restacks. #2497 likewise preserves its original base for #2498. Nothing has merged. #2434 and #2485 are closed as superseded, with branches and history retained.

Remaining blockers: full Main CI must complete; #2543 still depends on missing query/configuration APIs and is excluded. The earlier nine-pair inventory was not exhaustive: some other legacy controls also mix recovery/validation/lifecycle changes, so they need baseline audits and separation rather than a blind rebase. The query/result-boundary work and whole-TablePage extraction remain paused; ELFA's pin is unchanged. The detailed inventory, preservation rules and LOC scope are in the batch plan. Trekmeester owns review/merging.

Copy link
Copy Markdown
Contributor Author

ELFA pin compatibility is now documented in database-apps #121, with a reproducible source-import gate and migration checklist. At ELFA main d11dd2c, the current Atomic pin afea6cc supplies all 72 imported table API names; clean validation batch bd4fa0e lacks 45 (46 including test helpers). Even some present names have different editor/quick-add contracts. Shared delivery-app view navigation is among the consumers, so this is not only an experimental-grid issue.

The audit is preparation only: it changes no ELFA application code, package dependency, lockfile or submodule pointer. We must not treat the clean package's green CI as proof that it is a drop-in replacement for ELFA's current pin, nor copy the whole deferred state framework back just to satisfy the import list. Actual adaptation, fresh UI/file-package setup and NextGraph type/build/browser parity gates remain open.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant