[slisson]

What

Node attributes (jetbrains.mps.lang.core.structure.NodeAttribute) are now rendered and editable in the projectional editor.

• Rendering as layers. A node's cell is wrapped by its node attributes, the first one outermost as in MPS. An attribute's notation says where the attributed node goes with the new attributed cell (AttributedCellTemplate, DSL attributed { attributedConcept(...); multiple(...) }).
• No hidden context. Each layer is an ordinary CellCreationCall (AttributedNodeCellCreationCall). The attributed cell works out the next layer inward from the model alone (the attribute's parent and the next attribute after it), the way a child cell would. ChildSpecReference now holds a CellCreationCall, so adding or removing an attribute only rebuilds the layers next to it.
• Fallbacks.
  • An attribute whose notation has no attributed cell is not rendered, so it can't hide the node, and its check messages stay on the attributed node.
  • An attribute without any notation uses the reflective editor, with an attributed node: line.
  • The reflective editor no longer lists smodelAttribute as a child role.
• Adding and removing without intentions.
  • Typing the first constant before attributed in front of a node adds the attribute; typing the first constant after it behind the node does the same.
  • The caret position decides how the new attribute nests among existing ones.
  • The attributed concepts and multiple from @attribute info are respected.
  • Where several nodes start at the same position, the attribute goes on the outermost one.
  • Deleting the attribute's own text removes it and keeps the node.
• Notation language. New AttributedCell concept with behavior, a canBeChild constraint (node-attribute notations only), an editor and a generator rule that reads @attribute info. ReviewMigration in the jetbrains.mps.lang.core notation now uses it (@Review(reason) above the node).

Separate fix in the first commit

updateCodeCompletionActions rebuilt the menu entries from the anchor cell's substitute actions. When the menu had been opened as a side transform, this dropped all of its entries on the second keystroke, so no side transform longer than one character could be completed by typing it. It now reuses the menu that was opened, as hasCodeCompletionActions already did. This affects all multi-character side transforms, not only attributes.

Notes for reviewers

• IConceptEditorRegistry gained getAllConceptEditors(), implemented in LanguageAspectsFromMPSModules; the editor uses it to find attribute notations.
• Attributes are read through allChildren rather than the smodelAttribute link, because a model may store that role by name or by ID, and reading it one way registers an incremental dependency that misses changes stored the other way.
• CodeCompletionTest's list of known duplicates lost its <smodelAttribute> tokens because of the reflective-editor change.

Tests

• NodeAttributeTest (kernelf, 14 tests): rendering and nesting order, both fallbacks, adding by typing before, after, nested and key by key, the multiple and concept restrictions, and delete to remove.
• BaseLanguageTests (MPS-backed): @Review renders above a method, typing it adds it, deleting it removes it. All 31 tests in that module pass.
• The JVM suites of projectional-editor and kernelf-editor pass; the JS targets compile; the headless MPS build (:mps:assembleMpsModules) generates and compiles the notation modules.

🤖 Generated with Claude Code

[github-actions]

Test Results

623 tests   610 ✅  10m 57s ⏱️
 95 suites   13 💤
 95 files      0 ❌

Results for commit 743131a.

♻️ This comment has been updated with latest results.