WIP: UI Fix: Interactive Docking modal operator re-run glitch #126380

Closed
Jonas Holzman wants to merge 2 commits from Brainzman/blender:fix-bad-docking-modal into main

When changing the target branch, be careful to rebase the branch in your fork to match. See documentation.
Member

The following is a weird bug, with an even weirder and (temporary) fix.

Bug Description

The main interactive docking modal operator (SCREEN_OT_area_join) can sometimes be wrongly re-run by the Window Manager modal handler after finishing when opening/closing windows. This is due to the modal handler interpreting changes in the current context window as the operator being freed. (see the condition in wm_event_system.cc). This causes the operator to get in a broken/in-between state, where the cursor randomly changes and different areas get random docking highlights and hover behavior.

Demo

Capture_2024-08-15_22.21.08.gif


Fix

As the check happens in the modal handler, my original fix idea was to try to fake it, by re-overriding the original window back into context, see this diff:

diff --git a/source/blender/editors/screen/screen_ops.cc b/source/blender/editors/screen/screen_ops.cc
--- a/source/blender/editors/screen/screen_ops.cc
+++ b/source/blender/editors/screen/screen_ops.cc
@@ -4021,8 +4028,10 @@
+  wmWindow *modal_win = CTX_wm_window(C);
+
   /* execute the events */
   switch (event->type) {
 
     case MOUSEMOVE: {
       area_join_update_data(C, jd, event);
       area_join_dock_cb_window(jd, op);
       WM_cursor_set(jd->win1, area_join_cursor(jd, event));
       WM_event_add_notifier(C, NC_WINDOW, nullptr);
@@ -4133,8 +4141,21 @@
 
         const bool do_close_win = jd->close_win;
         wmWindow *close_win = jd->win1;
         area_join_exit(C, op);
         if (do_close_win) {
           wm_window_close(C, CTX_wm_manager(C), close_win);
         }
 
+        if (modal_win != close_win) {
+          /* This modal operator may change the active context window by creating new ones.
+           * However, the Window Manager modal handler excepts the active context window to
+           * not change. If it does, the handler will assume the operator has been freed,
+           * preventing its function from properly finishing, and calling its invoke function
+           * again.
+           * To circumvent this issue, set the active context window back to its original state
+           * if it wasn't closed.
+           *
+           * See wm_handler_operator_call() in wm_event_system.cc for more details.
+           */
+          CTX_wm_window_set(C, modal_win);
+        }

This however, does not work when windows are being closed for their areas to be merged into other screens, as seen towards the end of the demo, as they have been freed and naturally cause null derefs in window_set.

I thus went for the alternative (and very hacky approach) of tagging the operator via a fake op->customdata address, acting as a value, a hack possible since the modal handler doesn't clear it between runs. This let the modal function check if it's being wrongly run again, and finish gracefully. See the content of this PR.

This took me quite some time to debug and figure out. And this is certainly not a proper fix. It (maybe?) could be merged in to let user better test the experimental feature. But a proper fix would be to either :

  • Find a better (less hacky) workaround
  • Not opening/closing window in the operator modal function.
  • Improving the check on the Window Manager modal handler side, looking at the git history, this used to check if the entire window_manager had changed, instead of just the window (link), which would work in our case, but is no longer valid. Perhaps we could find a better, sort of in-between/less invasive check without breaking operators/addons that rely on this behavior? (cc @ideasman42)
The following is a weird bug, with an even weirder and (temporary) fix. ### Bug Description The main interactive docking modal operator (`SCREEN_OT_area_join`) can sometimes be wrongly re-run by the Window Manager modal handler after finishing when opening/closing windows. This is due to the modal handler interpreting changes in the current context window as the operator being freed. (see [the condition in wm_event_system.cc](https://projects.blender.org/blender/blender/src/commit/f6141b3fa7ebcb1ba9fe1cbe345f1f995ac691de/source/blender/windowmanager/intern/wm_event_system.cc#L2549)). This causes the operator to get in a broken/in-between state, where the cursor randomly changes and different areas get random docking highlights and hover behavior. #### Demo ![Capture_2024-08-15_22.21.08.gif](/attachments/c0175d1b-58ef-485b-957d-e8962221a335) --- ### Fix As the check happens in the modal handler, my original fix idea was to try to fake it, by re-overriding the original window back into context, see this diff: ``` diff --git a/source/blender/editors/screen/screen_ops.cc b/source/blender/editors/screen/screen_ops.cc --- a/source/blender/editors/screen/screen_ops.cc +++ b/source/blender/editors/screen/screen_ops.cc @@ -4021,8 +4028,10 @@ + wmWindow *modal_win = CTX_wm_window(C); + /* execute the events */ switch (event->type) { case MOUSEMOVE: { area_join_update_data(C, jd, event); area_join_dock_cb_window(jd, op); WM_cursor_set(jd->win1, area_join_cursor(jd, event)); WM_event_add_notifier(C, NC_WINDOW, nullptr); @@ -4133,8 +4141,21 @@ const bool do_close_win = jd->close_win; wmWindow *close_win = jd->win1; area_join_exit(C, op); if (do_close_win) { wm_window_close(C, CTX_wm_manager(C), close_win); } + if (modal_win != close_win) { + /* This modal operator may change the active context window by creating new ones. + * However, the Window Manager modal handler excepts the active context window to + * not change. If it does, the handler will assume the operator has been freed, + * preventing its function from properly finishing, and calling its invoke function + * again. + * To circumvent this issue, set the active context window back to its original state + * if it wasn't closed. + * + * See wm_handler_operator_call() in wm_event_system.cc for more details. + */ + CTX_wm_window_set(C, modal_win); + } ``` This however, does not work when windows are being closed for their areas to be merged into other screens, as seen towards the end of the demo, as they have been freed and naturally cause null derefs in `window_set`. I thus went for the alternative (and very hacky approach) of tagging the operator via a fake `op->customdata` address, acting as a value, a hack possible since the modal handler doesn't clear it between runs. This let the modal function check if it's being wrongly run again, and finish gracefully. See the content of this PR. This took me quite some time to debug and figure out. And this is certainly not a proper fix. It (maybe?) could be merged in to let user better test the experimental feature. But a proper fix would be to either : - Find a better (less hacky) workaround - Not opening/closing window in the operator modal function. - Improving the check on the Window Manager modal handler side, looking at the git history, this used to check if the entire window_manager had changed, instead of just the window ([link](https://projects.blender.org/blender/blender/commit/2939251a05c29b6462065394714a79195d067029)), which would work in our case, but is no longer valid. Perhaps we could find a better, sort of in-between/less invasive check without breaking operators/addons that rely on this behavior? (cc @ideasman42)
Jonas Holzman force-pushed fix-bad-docking-modal from 0c3abf0a9f to 84cda54113 2024-08-15 22:52:42 +02:00 Compare
Jonas Holzman requested review from Harley Acheson 2024-08-15 22:53:18 +02:00
Jonas Holzman changed title from WIP Fix: Interactive Docking modal operator re-run glitch to WIP: UI Fix: Interactive Docking modal operator re-run glitch 2024-08-15 22:57:07 +02:00
Jonas Holzman added 1 commit 2024-08-15 22:58:41 +02:00
Author
Member

Somehow, a Gitea bug seems to have duplicated this PR (seems to have happen on rename), closing in favor of the original: #126379

Somehow, a Gitea bug seems to have duplicated this PR (seems to have happen on rename), closing in favor of the original: #126379
Jonas Holzman closed this pull request 2024-08-16 00:55:27 +02:00

Pull request closed

Sign in to join this conversation.
No reviewers
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset System
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Code Documentation
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Viewport & EEVEE
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Asset Browser Project
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Module
Viewport & EEVEE
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Severity
High
Severity
Low
Severity
Normal
Severity
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
1 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#126380
No description provided.