Catalog: widget parent/depth, class/style and primary action (mendixlabs/mxcli#1268)

[ako]

Implements mendixlabs#1268 per the agreed design in the maintainer comment there.

What changed

Catalog (widgets_data, schema 19): the walker now passes the parent through. New columns:

• ParentWidgetId is the nearest indexed ancestor. Skipped wrappers (conditionalVisibilityWidget…), layout grid rows and columns, tab pages, and a pluggable widget's properties and object-list items are transparent, so a widget in a LayoutGrid column or a DataGrid2 column is parented to the grid.
• Depth is 0 at the page or snippet root. A list view template is indexed (Catalog: activities inside loops and their properties; CONV013/CONV014 read the action's error handling #940), so it counts as a level: list view d, template d+1, template content d+2. Depth does not cross a SnippetCallWidget (the snippet's widgets are its own rows, starting at 0).
• Class, Style and DynamicClasses come from Forms$Appearance.
• ActionType is the raw $Type of Action, else OnClickAction, else ClickAction.
• HasConfirmation is set when there is a ConfirmationInfo: MicroflowSettings.ConfirmationInfo for a microflow call, or directly on a nanoflow or workflow call. Forms$DeleteClientAction has no confirmation property, and a test pins this against the generated metamodel.

Starlark widgets() gains parent_widget_id, depth, class_name (class is a keyword), style, dynamic_classes, action_type, has_confirmation and page_ref.

Docs:

• The widget table moved to write-lint-rules/catalog-tables.md, because SKILL.md would have gone over the 700-line bound. It now includes an example rule: inline style, class allow-list, and delete button → call a microflow with a confirmation.
• The SKILL.md note that "action_type is the SDK name" now names widget.action_type as the exception: it is the raw storage type.
• docs-site tools/catalog-tables.md gets a new CATALOG.WIDGETS section.
• docs-site internals/catalog-schema.md had a stale four-column WIDGETS block; it is replaced with the real schema.
• CHANGELOG updated.

Vocabulary test: docRowValues returned the first action_type row in the file. The widget row would have shadowed the activity row and silently retargeted TestSkillDocumentsRealActionTypes, so a section-scoped docSectionRowValues was added. Documented widget action types must be Forms$…Action names that the codec registry knows. The issue's own "DeleteAction" would fail this check.

Values measured on TestApp: Forms$NoAction, MicroflowAction (21 with confirmation), CallNanoflowClientAction, FormAction, CancelChangesClientAction, ClosePageClientAction, SaveChangesClientAction, DeleteClientAction, OpenUserTaskClientAction, OpenLinkClientAction, SetTaskOutcomeClientAction and CreateObjectClientAction.

Test plan

• make build, make lint, make check-conformance, make check-findings
• go test ./mdl/catalog/... ./mdl/linter/... ./mdl/executor/ ./cmd/mxcli/
• Walker unit tests (builder_pages_tree_test.go) cover: parent through a skipped wrapper, LayoutGrid + tab container, ListView template (a level of its own), DataGrid2 column and child slot (parented to the grid), snippet roots, appearance, and the primary-action priority and confirmation per action kind.
• Studio Pro-authored TestApp (builder_pages_tree_testapp_test.go):
  • No dangling parents, child depth = parent + 1, roots at 0, across all 1615 widgets (with a control).
  • A 9-level ancestor chain through layoutGrid → dataView → … → tabContainer → dataGrid2 column → button.
  • Administration.Account_Edit.label4 has the right class and style.
  • Account_Overview.actionButton4 is Forms$DeleteClientAction with no confirmation.
  • A Studio Pro microflow button with a confirmation has HasConfirmation = 1.
• Starlark projection test (widgets_projection_test.go).
• End-to-end mxcli lint on a copy of testdata/pedapp. I added a page through MDL with a styled container, a disallowed class, a delete button and a microflow button with Confirmation:. A .claude/lint-rules/widget_hygiene.star rule implementing the issue's three checks plus a depth check reported 7 issues, all expected. The skill's example rule, extracted verbatim, ran clean and reported 4 issues. The confirmed microflow button was not flagged as a delete.
• Revert checks (each confirmed failing, then restored):
  1. Children not re-parented: tree tests fail (parent "" want dataView1, wrong depths).
  2. Skipped wrapper made non-transparent: textBox1 parent id "wrap", depth 2.
  3. Property bag loses its parent: pb1/emptyText1 parent "", and the TestApp chain collapses to [container5 actionButton5].
  4. MicroflowSettings.ConfirmationInfo not read: got (Forms$MicroflowAction, false).
  5. Appearance not read: unit test and TestApp label4 fail.
  6. has_confirmation dropped from the Starlark struct: projection test fails ("struct has no .has_confirmation").
  7. Skill doc changed to "DeleteAction": vocabulary test fails.
  8. E2E: an mxcli binary built with action and parent extraction reverted reports only the 3 style/class issues. The DELETE and DEPTH findings disappear.

Left open (follow-up)

• A pluggable widget's actions live in its property bag (WidgetValue.Action, e.g. a DataGrid2 or Gallery on-click), and a widget can have several, so there is no single "primary" one. action_type is empty for those. This needs a design decision.
• No finding was recorded: this is a feature, not a bug fix.

🤖 Generated with Claude Code