Group E vscode halves: Signal dropdowns, Alignment editing, mpt paste and shared-name ids, on core v1.40.0

[ww-mw]

The vscode halves of the group E ruling of 2026-10-16. The maintainer's ruling: "Agree your suggestions ... Add unit tests to lock down the fix. I pre-approve PR and merge as long as it passes all tests." Each item takes its recommendation in discuss.md.

Core pin: #v1.40.0. The last commit moves package.json from #v1.39.0 to #v1.40.0 (mathworks/data-explorer-core#129). That tag carries the core halves, mathworks/data-explorer-core#124, #125 and #127. It also carries #128 (MCOS numeric classes and per-element object classes), which the four item commits were not gated against.

• package-lock.json resolves data-explorer-core to f9acdf939aa726cb9ddd3162a6cac19fc4f4a5ca, version 1.40.0, the commit tag v1.40.0 points at.
• Relative to main, the lockfile changes only data-explorer-core's two entries: the root spec and the node_modules entry. check:leak is OK.
• CHANGELOG.md is unchanged. It is added at release.

Results at the head, after npm ci with v1.40.0:

• npm run verify: 175 files, 3422 of 3422 tests pass, and check:leak is OK.
• npm run test:integration: 221 passing, 1 pending, 0 failing, on the first run. The pending test is cutPasteUndo's "a same-document move ...". It skips itself where undo does not apply to an editor.
• These are the same counts as with core main 3decebe installed --no-save, so #128 adds no failure.

Against v1.39.0, the earlier pin:

• tsc fails once: src/host/parseWarnings.ts:131, TS2367, because 'duplicate-id' is not yet in ParseWarningCode.
• 12 unit tests and 10 integration tests fail. These are the new locks, listed below. (The at-pin log shows 13 integration failures: the extra 3 are the known typeLinkBinary fixed-wait timing flake, not locks.)

C-030: a Signal's Complexity and DimensionsMode are dropdowns

Ruling: "(a). Confirm Signal's enum sets ... in MATLAB first, then reuse the BusElement and ValueType mechanism with Signal's own options. Dimensions stays a label."
Pairs with mathworks/data-explorer-core#124 (cca1e6e).

• What changed: no behaviour change. The generic select cell already renders and edits the new cell. The column-group comment in rowBuilder.ts now names ValueType and Signal.
• Changed existing test: test/dimensionsMode.test.ts, "the Signal surfaces an editable dimensionsMode column". It pinned the read-only label that the ruling removes. It now expects {text:'Fixed', editable:true, editor:'select', options:['auto','Fixed','Variable']}.
  • Fail-before: against v1.39.0 it fails with expected 'label' to be 'select'.
  • A stale comment in test/elementProps.test.ts is updated. That is a comment-only change.

C-086: Alignment can be edited in the table, by MATLAB's rule

Ruling: "(b): the validation rule is already recorded, but the written form and defaults-stripping are exactly what the plan deferred. One MATLAB check turns this into a bounded core and vscode change."
Pairs with mathworks/data-explorer-core#124 (71b3446).

• What changed: no behaviour change. A comment in rowBuilder.ts now lists alignment among the columns that can be edited.
• New test: test-integration/suite/alignmentEdit.test.ts, with 6 tests through the real SlddTextEditorProvider. They run over core's MATLAB-written classes_r2027a_text.sldd, which is copied byte for byte. The tests check that:
  • the cell is editable text under ExportedGlobal, and plain text under Auto;
  • an edit of 8 saves pAl8's CoderInfo;
  • -1 removes the key;
  • 3 is refused in MATLAB's words, and the document does not change;
  • under Auto, an edit is refused;
  • setting StorageClass to Auto resets Alignment.
  • Fail-before: against v1.39.0 all 6 fail, because the cell is plain text and the edit is lost.
• Not changed: jsonHostEditRepaint's capitalised COLUMNS. None of the fixtures that sweep reads has a non-Auto Parameter or Signal, so it would never reach an alignment cell.

C-180: a pasted or dropped mpt entry lands in Design Data

Ruling: "(a). The paste copies an mpt object MATLAB already wrote, rather than writing one from scratch, so the doc's reason for refusing creation does not cover it."
Pairs with mathworks/data-explorer-core#125 (7f04c0b).

• What changed: src/host/sectionRules.ts now posts getAdmittedTypes(). On a core without that method it falls back to getAllowedTypes().
  • Before this, the drop cursor and the Paste item refused an mpt entry that the host would accept.
  • The Add gallery keeps the creation list.
• New tests:
  • sectionRules (+4): the posted rules agree with allowsType for every class, and Design Data lists mpt.
  • dropDecision (+2) and menuItems (+2): built from the rules a real model posts.
  • structuralEdit and xmlStructuralEdit (+4 together): both paste paths keep the class.
  • addCatalog (+1): no tile is an mpt class.
  • test-integration/suite/mptPasteDrop.test.ts: the paste and the drop through the real JSON and binary providers.
• Fail-before:
  • With sectionRules.ts reverted, on core 3decebe, 5 unit tests and both integration tests fail. The integration message is "Design Data's posted rule admits mpt.Parameter".
  • The paste and addCatalog tests are locks on the core change. Against v1.39.0 they fail with A "mpt.Parameter" entry is not allowed in Design Data.
  • The "still refuses ..." tests and the fallback test are controls. The fallback test fails if the fallback is removed.
• Changed existing tests: none have changed assertions. There is one import line and one comment.

F1-50: an edit on one of two same-named entries no longer lands on the first

Ruling: "(a). It stops the wrong-data delete without changing any id, and it suits a state that only malformed input reaches."
Pairs with mathworks/data-explorer-core#127 (98f8f7c).

• What changed:
  • SlddModel.findNode now treats the session's null as final when the session holds the source. Before, its cache fallback returned the first twin.
  • It still falls back to the cache for a core without findNodeById, and for a source the session no longer holds.
  • warningBanner does not count duplicate-id as a part that could not be read. When shared names are the only warning, it shows a headline of its own.
• New tests:
  • test/findNodeSharedName.test.ts (3 tests): two K Parameters and two S structs.
  • test/parseWarnings.test.ts (+3).
  • test-integration/suite/sharedNameEntries.test.ts (3 tests): the banner, a refused edit on a twin, and a control edit on another entry.
• Fail-before, with the src reverted on core 3decebe:
  • findNode: expected 'the first twin' to be null.
  • Banner: "2 parts of this file could not be read".
  • Integration: the edit set the first gravity to 1.
  • The fallback test is a control. It fails if the fallback is dropped.
• Changed existing test: test/binaryEntryScopedEdit.test.ts, "does NOT undo correctly with the inverses in forward order". It pinned the wrong-twin remove that the ruling removes.
  • It now expects No entry is indexed.
  • It also checks that both entries are left standing.