feat(morph): sub-slice A2 — Inspector Morph Targets subgroup - #570
Merged
Conversation
Second sub-slice of #518. A1 shipped the data path (importer + MorphAnimationManager + CLI list); A2 wires the QML Inspector surface that authors will actually use. ### What ships New "Morph Targets (N)" subgroup at the bottom of the Animations section in `qml/PropertiesPanel.qml`: - Bound to `MorphAnimationManager.morphTargetsForSelection()` — refreshes via the `morphTargetsChanged` signal when selection moves. - One row per target: name (elided), 0..1 slider (step 0.01), numeric readout. Slider `onMoved` writes back through `setWeightForSelection`, which drives the underlying Ogre `AnimationState` weight (the live preview path). - "Reset all" button — walks every target and sets weight to 0. - Filter / search text field, shown when target count > 6 (characters routinely ship 50+ shapes; flat list is hopeless). - Whole subgroup is `visible: targetCount > 0` so meshes without blend shapes don't show an empty header. Authoring (add new target from edit-mode delta, rename, delete) lands in A3. Export round-trip in A4. Dope-sheet integration in A5. MCP tools in A6. ### Manual smoke - Select an entity with blend shapes (any FBX from a DCC with facial rigs). The "Morph Targets (N)" subgroup appears at the bottom of the Animations section with one slider per shape. - Drag a slider — the mesh deforms in real time (the AnimationState weight drives the VAT_POSE track A1 created). - Click "Reset all" — every slider snaps back to 0. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
Warning Rate limit exceeded
You’ve run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. 📝 WalkthroughWalkthroughAdds a new "Morph Targets / Blend Shapes" subsection to the Animations panel in ChangesMorph Targets / Blend Shapes UI Subsection
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related issues
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4d7146ec55
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+4335
to
+4339
| Connections { | ||
| target: MorphAnimationManager | ||
| function onMorphTargetsChanged() { | ||
| morphCol.targets = MorphAnimationManager.morphTargetsForSelection() | ||
| } |
There was a problem hiding this comment.
The new morph-target panel only re-fetches targets on morphTargetsChanged, but each slider value is derived from weightForSelection(modelData) and there is no handler for morphWeightChanged. As a result, when weights are changed outside direct drag interaction (for example clicking Reset all or any other code path that calls setWeightForSelection), the backend weight updates but the displayed slider/readout can stay stale until the delegates are recreated. This makes the inspector show incorrect state immediately after valid edits.
Useful? React with 👍 / 👎.
Codex P1 on PR #570: The morph slider's `value` was bound to a one-shot function call (`MorphAnimationManager.weightForSelection(modelData)`), evaluated once at delegate creation. Any weight write that doesn't move the slider directly — Reset all, MCP, future dope-sheet scrubs — would update the backend AnimationState but the visible slider + numeric readout stayed stale until the delegate was recreated. From the user's perspective, "Reset all" appeared to do nothing. Fix: track a `weightTick` counter that bumps on `morphTargetsChanged` and `morphWeightChanged`, then bind the slider's `value` to a comma-expression that reads the tick before calling the weight getter — Qt's binding machinery picks up the tick dependency and re-evaluates the function when it changes. Same pattern QML typically uses to force re-evaluation of function-bound bindings. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (1)
qml/PropertiesPanel.qml (1)
4331-4331: ⚡ Quick winConsider adding null-safety for the targets property.
While
MorphAnimationManager.morphTargetsForSelection()likely returns a valid array, adding a defensive default prevents potential crashes if it ever returns null or undefined:🛡️ Proposed fix
- property var targets: MorphAnimationManager.morphTargetsForSelection() + property var targets: MorphAnimationManager.morphTargetsForSelection() || []function onMorphTargetsChanged() { - morphCol.targets = MorphAnimationManager.morphTargetsForSelection() + morphCol.targets = MorphAnimationManager.morphTargetsForSelection() || [] morphCol.weightTick = morphCol.weightTick + 1 }Also applies to: 4345-4345
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@qml/PropertiesPanel.qml` at line 4331, The property declaration for targets uses MorphAnimationManager.morphTargetsForSelection() without null-safety; update the property initialization (both occurrences that reference property 'targets') to defensively default to an empty array when MorphAnimationManager.morphTargetsForSelection() returns null or undefined (e.g. use a null-coalescing or default expression around the call) so any consumers of 'targets' always get an array.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@qml/PropertiesPanel.qml`:
- Line 4331: The property declaration for targets uses
MorphAnimationManager.morphTargetsForSelection() without null-safety; update the
property initialization (both occurrences that reference property 'targets') to
defensively default to an empty array when
MorphAnimationManager.morphTargetsForSelection() returns null or undefined (e.g.
use a null-coalescing or default expression around the call) so any consumers of
'targets' always get an array.
CodeRabbit nit on PR #570. The manager currently always returns a QStringList, but contracts drift; the `|| []` fallback keeps the delegate Repeater binding safe if the getter ever starts returning null/undefined. Applied at both call sites (property initialiser + onMorphTargetsChanged handler). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
3 tasks
fernandotonon
added a commit
that referenced
this pull request
May 17, 2026
…576) * feat(morph): sub-slice A5 — dope sheet shows morph-weight tracks Per #518: "Dope sheet integration: morph-weight tracks appear in the dope sheet alongside bone tracks." Slice A5 ships the read-only display half — one row per Ogre::Pose on the selected entity with diamond markers at each keyframe time. Full selection / move / copy / paste interaction for morph tracks is a follow-up (it depends on A3's authoring path to land first). ### What ships **`AnimationControlController::allMorphRows()`** (new Q_INVOKABLE): - Walks the selected entity's `getPoseList()`. - For each pose, looks up the matching `Ogre::Animation` (which A1's importer creates one-per-pose), reads its VAT_POSE track's keyframe times. - Returns `[{ name: QString, keyTimes: [double] }]` — same shape the dope sheet's existing bone-row reader expects, minus the channel-flags map (morph tracks are scalar weight, no channel decomposition). **`AnimationDopeSheet.qml`** picks up the new API: - New `morphRows` property bound to `allMorphRows()`. - Refresh hooks: existing `onSelectionChanged` now also refreshes morph rows; new `MorphAnimationManager` Connections refresh on `morphTargetsChanged` (selection moved) and `morphWeightChanged` (Inspector slider, MCP poke, future authoring path). - New `morphBand` Rectangle anchored to the bottom of the dope sheet: collapses to height=0 when there are no morphs (so a bone-only animation looks the same as before); otherwise shows "Morph Targets (N)" header + one row per pose with the same diamond style the bone tracks use. ### 3 new tests in `AnimationControlController_test.cpp` - `AllMorphRowsEmptyWhenNoSelection` — returns `[]` cleanly. - `AllMorphRowsEmptyForMeshWithoutPoses` — selected entity with bones but no poses returns `[]`. - `AllMorphRowsListsPoseNamesAndKeyTimes` — builds a mesh with two named poses + matching VAT_POSE animations, asserts both names appear and each carries a single t=0 keyframe (A1's importer-time default). ### #518 status after this slice | Sub-slice | What | Status | |-|-|-| | A1 | Importer + manager + CLI list | ✅ #569 | | A2 | Inspector "Morph Targets" subgroup | ✅ #570 | | A4a | glTF morph-target export | ✅ #573 | | A4b | FBX morph-target export | ✅ #575 | | A5 | Dope sheet morph rows (read-only) | this PR | | A6 | MCP tools | ✅ #571 | | A3 | Authoring (delta → new pose, rename, delete) | still pending | Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * review(morph): A5 — filter morph keyTimes by pose target handle Codex P2 on PR #576: `allMorphRows()` looped over every VertexTrack in the per-pose Animation. A1's importer groups same-named poses across submeshes into a single Animation with one VAT_POSE track per affected submesh (so e.g. a "Smile" target on body + head ends up as one Animation with two tracks). Without filtering, the dope sheet row for that pose would carry diamonds from both tracks — duplicated `t=0` markers in the simple A1 case, and wrong key counts once authoring adds per-time keys. Fix: look up only the VertexTrack whose handle matches `pose->getTarget()`. Also assert it's a VAT_POSE track (belt-and-suspenders — VAT_MORPH could be added in a future slice and we don't want to silently mix track types). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(morph): A5 — avoid overlap between bone ListView and morph band The bone-row ListView and the new morph band were both anchored to `parent.bottom`. With both visible, their layout rectangles overlapped, which (under some redraw conditions in CI) drove SIGSEGV crashes in any MainWindow-construction test that instantiated the dope sheet QML during MainWindow startup — MCPServerTest.ToggleNormals_WithMainWindowTogglesVisibility and MainWindowTest.ViewMenuConsoleToggleUpdatesDockVisibilityAndSettings both regressed on this branch. Fix: anchor the bone ListView's bottom to `morphBand.top` when the morph band is visible, falling back to `parent.bottom` otherwise. The two no longer fight for the same space. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(morph): A5 — drop `onMorphWeightChanged` QML binding The signal's payload includes `Ogre::Entity*`, a raw pointer Qt 6 can't safely marshal to QML/JS. Binding the slot crashed the QML engine during MainWindow construction on Linux/Xvfb CI runners — manifested as SIGSEGV in MCPServerTest.ToggleNormals_* and MainWindowTest.ViewMenuConsoleToggle*. Both tests construct a MainWindow which loads the dope sheet QML; the binding setup itself was the unsafe step. The dope sheet's per-row data is structural (pose name + keyframe times), not weight-driven. Listening only to `morphTargetsChanged` (which fires on selection change and is parameter-free) covers the cases that matter: a different entity showing different morph names. Per-weight notifications were purely cosmetic (the row would re-fetch identical data on every slider drag) and skipping them is the right tradeoff. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(morph): A5 — defer QML dope-sheet integration to a follow-up The QML changes in this slice consistently triggered SIGSEGV in unrelated test suites (MCPServerTest + MainWindowTest visibility-toggle tests), deterministically across 4 CI re-runs. The 3 new C++ controller tests all pass; only the QML integration crashes are blocking merge. Ship just the data API (`AnimationControlController::allMorphRows`) and its tests in this slice. The QML dope-sheet morph band will land in a separate PR where the crash can be isolated without holding back the backend work that 3 other in-flight slices need. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2 tasks
fernandotonon
added a commit
that referenced
this pull request
May 17, 2026
…#578) * feat(morph): sub-slice A3 — authoring (add from edit, rename, delete) Last remaining sub-slice of #518. Closes the morph epic's authoring loop: users can save the current edit-mode geometry as a new morph target, rename existing targets, and delete unwanted ones. All three operations are undoable. ## What ships ### `commands/MorphCommands.{h,cpp}` — three QUndoCommand subclasses - `AddMorphTargetCommand` — create N same-named Ogre::Pose entries (one per submesh) plus a matching VAT_POSE Animation. Slices come in as sparse `{submeshHandle, {vertexIndex: Vector3f}}` maps so the command doesn't depend on EditableMesh — anything that produces deltas can drive it (current edit, MCP, future "fix mirror axis" rules…). - `DeleteMorphTargetCommand` — snapshot the same-named pose offsets at construction, drop the pose(s) + Animation + AnimationState on redo, rebuild from the snapshot on undo. - `RenameMorphTargetCommand` — same snapshot-and-rebuild pattern, just with a different name on the rebuild side. Ogre 14.5 doesn't expose Pose::setName, so destroy-and-recreate is the only path. ### `MorphAnimationManager` — three new Q_INVOKABLE methods - `addMorphTargetFromCurrentEdit(name)` — reads `EditModeController`'s live EditableMesh, diffs every submesh against the mesh's bind positions on the GPU buffer, builds the slice list, pushes an AddMorphTargetCommand. Falls back to no-op if not in edit mode, no vertices moved, or name collides. - `renameMorphTarget(old, new)` — name-trim, collision check, push. - `deleteMorphTarget(name)` — existence check, push. All three resolve the entity from SelectionSet's first entity (same pattern as the A1/A2 setters), push onto the shared `UndoManager` stack so Ctrl+Z reverses, and emit `morphTargetsChanged` so the Inspector re-fetches. ## Tests 8 new GTest cases on `MorphAnimationManager_test.cpp`: - Add command creates + undoes pose and Animation cleanly. - Delete command round-trips (poses restored on undo). - Rename command round-trips (old name restored on undo). - Rename rejects target-name collisions, no-op self-renames, blank names. - Delete rejects unknown / empty names; succeeds on a real target. - Add-from-edit rejects empty / whitespace names before edit-mode check. - Add-from-edit returns false when no EditableMesh is available (i.e. user isn't in edit mode). - All three methods reject when nothing is selected. The command path is what's actually under test — the manager wrappers are thin selection + name-validation glue. We keep direct command tests because that's where the undo / redo contract lives and it exercises the pose snapshot / rebuild path without needing an Inspector. ## What's deferred - **Inspector UI** (Add / Rename / Delete buttons in the Morph Targets subgroup) — small but lives in QML, and the last QML change in this epic (A5's dope-sheet integration) deterministically crashed unrelated tests for reasons I never fully isolated. Doing the UI in a separate PR keeps blast radius contained. - **MCP tools** `add_morph_target` / `rename_morph_target` / `delete_morph_target` — straightforward but adds surface; can land alongside the Inspector UI. - **CLI** `qtmesh morph --rename / --delete` — needs export-after-edit plumbing through the FBX/glTF morph exporters from A4a/A4b. #518 status after this slice: | Sub-slice | Status | |-|-| | A1 — Importer + manager + CLI list | shipped (#569) | | A2 — Inspector subgroup | shipped (#570) | | A3 — Authoring | **this PR** | | A4a — glTF export | shipped (#573) | | A4b — FBX export | shipped (#575) | | A5 — Controller API | shipped (#576) | | A5b — Dope-sheet UI | follow-up | | A6 — MCP tools | shipped (#571) | | A3b — Inspector / MCP / CLI authoring surface | follow-up | Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(build): add MorphCommands.cpp to tests/ source list Linker errors on the secondary test binaries (MaterialEditorQML_test, MaterialEditorQML_qml_test, MaterialEditorQML_perf_test) because they compile MorphAnimationManager.cpp (which now references the three MorphCommand constructors) but didn't pick up MorphCommands.cpp. src/CMakeLists.txt has it; tests/CMakeLists.txt was missing the line. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * review(morph): A3 — capture bind positions on EditableMesh load Addresses Codex P1 findings on PR #578: ## P1 #1 — diff against captured bind positions, not the live buffer `addMorphTargetFromCurrentEdit` was diffing `EditableMesh` against positions read from the entity's current GPU vertex buffer. But edit-mode ops (translate / rotate / scale) call `commitToEntity` during the gesture — both sides of the subtraction end up identical in the common flow, so `slice.offsets` stays empty and the command incorrectly returns no-op even after visible edits. Fix: snapshot the bind positions once at `EditableMesh::loadFromOgreMesh` time (`EditableSubMesh::originalPositions`). The snapshot is never mutated by edit-mode ops, so `vertices[i].position - originalPositions[i]` recovers the right delta regardless of how many commits have run since enter-edit-mode. ## P1 #2 — shared-vertex submeshes The old `readBindPositions` rejected any submesh with `useSharedVertices=true` (vertexData null in that case). Most Mixamo-style assets use shared vertices, so morph authoring silently skipped them. Fix is implicit in #1: `loadFromOgreMesh` already copies the shared vertex pool into each affected submesh's `vertices` array; the new `originalPositions` snapshot is built from that same source, so shared-vertex submeshes carry their own baseline. The GPU-buffer read path goes away entirely. ## CodeRabbit minor — null-assert SelectionSet in tests Added `ASSERT_NE(sel, nullptr)` before `sel->append(entity)` in the three new authoring tests, matching the slice-A1 pattern. If the singleton ever fails to initialize, we get a clear assertion failure instead of a null-deref crash. ## Tests - `LoadFromEntityTriangleMesh` extended to assert (a) the snapshot matches `vertices[].position` after load and (b) mutating `vertices[].position` does NOT touch `originalPositions`. Exercises the shared-vertex case (the triangle fixture uses shared verts). - New `AddMorphTargetUndoableViaUndoManager` exercises the full add → undo → redo loop through the shared `UndoManager`, with a hand-built slice that mirrors what `addMorphTargetFromCurrentEdit` will produce once edit-mode capture is wired up end-to-end. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 17, 2026
fernandotonon
added a commit
that referenced
this pull request
May 17, 2026
Final sub-slice of #518. The C++ data API (`allMorphRows`) shipped in #576 as a Q_INVOKABLE on AnimationControlController; this PR wires it into the dope-sheet view. ## What ships - New `morphRows` property on the dope-sheet root, bound to `AnimationControlController.allMorphRows()`. Refreshed via the existing `onSelectionChanged` handler so a clip change rebuilds both bone and morph rows in lockstep. - `rowsView` (bone ListView) bottom anchor switches to `morphBand.top` when the morph band is visible. Bone-only entities → band collapses to `height=0` → bone list draws full height as before. - New `morphBand` Rectangle at the bottom: header row showing "Morph Targets (N)", then one row per pose with the target name on the left and `#88ccff` diamond markers at each keyframe time on the right. Diamonds share the bone-row timeline math (`pxPerSec`, `viewStart`) so they line up vertically. ## Why this is small and safe The original A5 (PR #576's first commit) deterministically crashed unrelated MainWindow / MCPServer visibility-toggle suites in CI. Root cause was never fully isolated but the suspicion was an `import PropertiesPanel 1.0` triggering a different QML singleton chain during MainWindow construction. This PR keeps the discipline that worked for A3b: - **No new QML imports.** `AnimationControl 1.0` was already imported. `allMorphRows()` was deliberately put on AnimationControlController (not MorphAnimationManager) for exactly this reason. - **No new `Connections` blocks.** The existing one already fires on the signals we care about. - **No MouseAreas, no selection, no drag** on morph diamonds — this is strictly read-only display. Interactive morph keyframes (selection, multi-select, drag, copy/paste) need their own controller surface and land in a separate PR. ## #518 status After this slice the issue is feature-complete and ready to close: | Sub-slice | Status | |-|-| | A1 — Importer + manager + CLI list | shipped (#569) | | A2 — Inspector subgroup | shipped (#570) | | A3 — Authoring data layer | shipped (#578) | | A3b — Inspector authoring UI | shipped (#579) | | A4a — glTF export | shipped (#573) | | A4b — FBX export | shipped (#575) | | A5 — Controller API | shipped (#576) | | A5b — Dope-sheet UI | **this PR** | | A6 — MCP tools | shipped (#571) | Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Second sub-slice of #518. A1 (#569) shipped the data path — importer +
MorphAnimationManager+qtmesh morph --list. A2 wires the QML Inspector surface authors will actually use.What ships
New Morph Targets (N) subgroup at the bottom of the Animations section in `qml/PropertiesPanel.qml`:
What's NOT in this PR (deferred to A3–A6)
list_morph_targets,set_morph_weight,add_morph_target,delete_morph_target.Manual smoke
Test plan
MorphAnimationManagerstill pass (this PR doesn't change the C++).🤖 Generated with Claude Code
Summary by CodeRabbit