Merge pull request #5861 from RosettaCommons/roccomoretti/interactive_bind_soft_crash
[Foldit] Don't (immediately) crash if callback objects are past their lifetime.
Continuing on from my examination of the crash reports from the intro tutorials, I noticed there's a fair amount of crashes related to callbacks, specifically failing the indirection_shim object lifetime check from the checked_fxn_wrapper.
It looks like this mostly coming from the EventHandler::on_lose_highlight() callback, and from what I can tell that's pretty much all from the Panel and StaticImage callbacks set in src/game/application/PuzzleDialog.cc. Unfortunately, I don't see anything obviously wrong with the lifetime handling which we can fix.
With a fair amount of history with the checked_fxn_wrapper, I think it's now okay-ish to make failing the object check to simply be a no-op. (That is, if the object we're running the member function on has ended its lifetime and no longer exists, don't crash, but simply skip running the function.) -- For many things, such as the display updates, this is fine: we don't need to update the display of an object which no longer exists. For other things, this may be slightly dangerous, as it may result in inconsistent state or a return value which is nonsense. But if it is an issue, chances are that it would crash later, and if it doesn't, then we're doing better by the user.
But to help with debugging, I've added a soft-crash report when it happens, such that we can get informed if that's a major source of issues. But to be useful, we need to have a backtrace with the soft crash. I already had a branch with this enabled, which also includes some use cases with the ligand entries.
So from an end user perspective, what should happen is things which were resulting in a hard crash aren't (necessarily) a crash anymore. That may result in a more obtuse error result later, but hopefully combined with the soft crash info, it should be interpretable.
RELEASE NOTES: Address some crashes related to switching puzzles.
Merge pull request #5858 from RosettaCommons/roccomoretti/interactive_hotkey_on_start
[Foldit] Avoid a crash with hotkeys on intro puzzles.
If you press a hotkey when the StartPuzzleTool is running, it can result in a crash. Don't run actions (e.g. from a hotkey) from the SelectionTool unless the SelectionTool is active.
RELEASE NOTES: Fix crashes with hotkeys during intro puzzle loading.
Merge pull request #5859 from RosettaCommons/roccomoretti/interactive_ideality_composite
[Foldit] Fix crashes related to IdealityComposite.
There were a fair number of crashes in the intro levels because of IdealityComposite::update. It looks like the issue is stale pose sizes. We can fix that by calling for an update if we're stale, and also being cautious about other accesses.
RELEASE NOTE: Fix crashes in intro puzzle loading.
Merge pull request #5857 from RosettaCommons/roccomoretti/interactive_move_count_win_issue
[Foldit] Reset MoveCountTracker for new puzzle.
The MoveCountTracker was carrying the last action into subsequent puzzles. This means that if you play "Swing It Around" directly after "One Small Clash", the ActionPullSidechain (which is a "First use only" move cost) would be counted toward the "One Small Clash" move count, but not for the "Swing It Around". This would then cause the "One Small Clash" to fail the move count check that was added with PR #5813
The solution is to reset the MoveCountTracker for new puzzle loading, as the last-used action shouldn't carry over between puzzles.
RELEASE NOTES: Fix issue where intro puzzle win conditions weren't registering.
Merge pull request #5828 from RosettaCommons/jflat06/fix-directory-spam
Fix an issue where the sequence data store was creating extraneous directories.
Merge pull request #5829 from RosettaCommons/jflat06/fix-phantom-foldtree-roots
Fix an issue with custom fold tree suggestions accumulating redundant suggestions on cut close.
Merge pull request #5777 from RosettaCommons/koepnick/foldit_hbnet_highlight
[Foldit] HBondNetworkFilterGeom only show BUNS when highlighting network
Release Notes:
- HBond Network Objective only highlights BUNS when the cursor hovers over a network
Merge pull request #5784 from RosettaCommons/roccomoretti/interactive_fix_shrink_residue_bug
[Foldit] Fix bug with shrink_marked_residues_with_animation.
In ligand design puzzles, I'm seeing an odd crash related to shrink_marked_residues_with_animation(), which is being called by the tool activation in setup_for_puzzle().
Best I can tell, this is related to having the display pose being out of step with the Geometry. (The actual crash happens in Conformation::residue(), which is being called from GUIPoseProxy::has_mainchain().
This is a somewhat quick fix, in that we should be able to handle the shrink_marked_residues_with_animation() prior to the tool activation/switch.
RELEASE NOTES: Fix crash on puzzle loading.
Merge pull request #5783 from RosettaCommons/roccomoretti/interactive_fix_hbond_crash
[Foldit] Fix issue with crashes in hbond scoring.
One of the more regular crashes we're getting now in ligand design puzzles is a crash in ligand scoring, related to vector normalization.
Tracing through where it's used downstream, I think we should be okay with returning a zero vector in these rare cases.
RELEASE NOTES: Fix some crashes in scoring related to hydrogen bonds
Merge pull request #5795 from RosettaCommons/roccomoretti/interactive_linux_close
[Foldit] Fix how Linux reacts to the window close button.
Right now, the Foldit linux client is unceremoniously killed when you click the main (OS-generated) "window close" button. (Skipping any "save options" code, and often resulting in a segfault on exit.) The reason for this is that the "official" GLUT specification (apparently) does not have any mechanism for detecting this event and responding properly with cleanup. However, most Linuxes use OpenGLUT or FreeGLUT, which have an extension which is able to handle this (same API for both).
This PR adjusts the GLUTWrapper implementation to recognize when we're compiling with OpenGLUT/FreeGLUT, and use that mechanism. This requires a bit of a rework for the on_close functionality, because unlike the Windows or Mac equivalents, the program is getting closed when that box is triggered on Linux, with no mechanism to cancel it.
I also adjusted things such that the on_close functionality goes through the same handle_exit() mechanism the rest of the exit button in-game go through. The one exception is when the button is clicked prior to the login screen coming up, where handle_exit() causes crashes due to uninitialized variables, in which case we use the current approach. (You can still get crashes when closing prior to the login screen coming up due to multithreading issues, but those are crashes we already have in the released client.)
RELEASE NOTES: Fix how we handle the close window button for Linux clients.
Merge pull request #5808 from RosettaCommons/roccomoretti/fix_collagen_disulfide
[Foldit] Fix Collagen EduMode puzzle for disulfides.
Apparently you should be able to make disulfides in the Collagen puzzle. (Re)enable that.
RELEASE NOTES: Fix disulfide display in the Collagen Education Mode puzzle.
Merge pull request #5815 from RosettaCommons/roccomoretti/interactive_menu_to_front
[Foldit] Bring panels to top and on screen when toggled.
Add `toggle_to_top()` functionality for panels which will bring the specific panel to the top (over other items) and will make sure the panel is on screen.
There's a slight rework of Widget::ensure_on_screen(), to make sure we don't cut off the top of panels (and thus remove the title bar handle) when the screen is too short.
RELEASE NOTES: Make sure panels are on-screen and visible when they appear.