OrbWeaver

Fixed in Git revision a1f0c789f34b4bb11a132aaeb03a380da18fe6b5. I tested switching light textures in the Light Inspector and switching brush textures on the Media tab, and everything still looks fine so hopefully nothing is broken as a result of switching MaterialSelector to use a sigc::signal instead of an explicit selectionChanged callback.

The actual crash occurs on this line of LightInspector.cpp: 85 void LightInspector::shaderSelectionChanged() 86 { 87 // Get the selected shader -> 88 auto ishader = _texSelector->getSelectedShader(); The _texSelector member is still NULL here, so this is a segfault. The initialisation of _texSelector happens in setupTextureWidgets(), which is called unconditionally in the LightInspector constructor, so a missing method call is not the issue. However, the value of _texSelector is a newly-constructed MaterialSelector object, and this object takes a function pointer to the shaderSelectionChanged() in its constructor: _texSelector = new MaterialSelector(parent, std::bind(&LightInspector::shaderSelectionChanged, this), MaterialSelector::TextureFilter::Lights); It appears that this is creating a race condition: the constructor of MaterialSelector is able to call the shaderSelectionChanged callback (presumably as a result of selection changes during its Populate() method), but it isn't actually safe to call shaderSelectionChanged() until after the MaterialSelector is fully constructed and assigned to the _texSelector member. I think my preferred solution would be to remove the "selectionChanged" function parameter to the MaterialSelector constructor entirely, and replace it with a public signal_selectionChanged() which client code could connect to. This would have two advantages: No possibility of race condition, since you would have to construct the MaterialSelector before connecting to the signal. Signals are more powerful than manual std::function callbacks (e.g. they can auto-disconnect if the target object is destroyed) and are widely used throughout the codebase. However, if anything currently relies on receiving selectionChanged callbacks during the MaterialSelector construction, this would no longer work. @greebo what do you think?

I can reproduce it on Linux in a local source build. Logged as issue 6216.

I would like this option as well. I can't be bothered to hold down Ctrl and a movement key (especially when you're moving slower, so you have to hold it for ages), so I just use toggle crouch as my "creep" replacement. But being able to toggle creep would be a better option, not least because it doesn't change the camera perspective as crouch does.

I've often felt that door opening sounds were much too loud, so this sounds like a potential improvement. One problem is that certain cupboards with two doors actually play the sound twice at the same time (once for each door), making it double the expected volume. But even regular single doors often sound too loud to me.

Looks like I jumped the gun slightly. It doesn't seem to have been completely removed, but has been deprecated for several versions. Knowing Apple, I will be very surprised if they don't eventually remove it altogether in a few versions time.

I don't think you can use OpenGL at all on M1 macs (natively). Anyone who wants to make that port is going to need to write a complete Metal replacement for the renderer.

This does seem to be the cause of the problems. Replacing all calls to FindPage() with GetPageIndex() fixes the issue for me.

I've done more debugging, and it seems like it is FindPage() that is failing for reasons which are unclear. It even fails immediately after the page is added in addControl(), i.e. bool succ = AddPage(content, control->getDisplayName(), false, tabIconIndex); wxASSERT(succ); // succeeds wxASSERT(FindPage(content) != wxNOT_FOUND); // FAILS I guess this means one of two things: A bug in Ubuntu's wxWidgets 3.0.x package (it wouldn't be the first time), or the GTK wxWidgets build more generally. A false assumption in our code about when/how wxAuiNotebook::FindPage() is able to work. For example, perhaps it only works once the AUI interface has actually been displayed on the screen, rather than during the setup process. EDIT: I wonder if it is this bug: https://github.com/wxWidgets/wxWidgets/issues/15932 I can armour restoreState() against passing an invalid index to GetPage() and friends, which avoids the assertion, but then I have multiple copies of every tab in the group widget. On the other hand, I can completely disable the contents of restoreState() and everything looks fine with all tabs present and correct (although I presume any changes to tab layout wouldn't be preserved), so I wonder if it's even necessary for restoreState() to add new tabs at all?

It looks like GetPageCount() and GetPage() are working with size_t indices which (on Linux) are equivalent to unsigned int, so I guess this is wrap-around behaviour. Yes, it's very strange that the index should be -1 immediately after the page was added — I'll do more debugging and see if I can find out which function call isn't working as expected.

I'm also seeing an assertion failure, although it doesn't seem to have anything in common with what other people have posted. This appears at startup before the main window is shown, and appears to be fatal (attempting to Continue just produces the same assertion again). ASSERT INFO: ../src/aui/auibook.cpp(2270): assert "page_idx < m_tabs.GetPageCount()" failed in GetPage(). BACKTRACE: [1] wxAuiNotebook::GetPage(unsigned long) const [2] sigc::internal::signal_emit0<void, sigc::nil>::emit(sigc::internal::signal_impl*) [3] sigc::signal0<void, sigc::nil>::emit() const [4] module::ModuleRegistry::loadAndInitialiseModules() [5] radiant::Radiant::startup() [6] wxEvtHandler::ProcessEventIfMatchesId(wxEventTableEntryBase const&, wxEvtHandler*, wxEvent&) [7] wxEvtHandler::SearchDynamicEventTable(wxEvent&) [8] wxEvtHandler::TryHereOnly(wxEvent&) [9] wxEvtHandler::ProcessEventLocally(wxEvent&) [10] wxEvtHandler::ProcessEvent(wxEvent&) [11] wxEvtHandler::ProcessPendingEvents() [12] wxApp::DoIdle() [13] g_main_context_dispatch [14] g_main_loop_run [15] gtk_main [16] wxGUIEventLoop::DoRun() [17] wxEventLoopBase::Run() [18] wxAppConsoleBase::MainLoop() [19] wxEntry(int&, wchar_t**) [20] __libc_start_main More info from LLDB (since the built-in stacktrace seems to miss out a lot of function calls): frame #10: 0x00007ffff6a9eeca libwx_gtk3u_aui-3.0.so.0`wxAuiNotebook::GetPage(unsigned long) const + 106 frame #11: 0x0000555555d4f768 darkradiant`ui::PropertyNotebook::restoreState(this=0x000055555611c490) at PropertyNotebook.cpp:264:34 frame #12: 0x0000555555d3f7b6 darkradiant`ui::AuiLayout::restoreStateFromRegistry(this=0x0000555557179c10) at AuiLayout.cpp:654:36 frame #13: 0x0000555555d46817 darkradiant`ui::MainFrame::construct(this=0x00005555560c53a0) at MainFrame.cpp:271:38 frame #14: 0x0000555555d46680 darkradiant`ui::MainFrame::postModuleInitialisation(this=0x00005555560c53a0) at MainFrame.cpp:211:11 frame #15: 0x0000555555d4c024 darkradiant`sigc::bound_mem_functor0<void, ui::MainFrame>::operator(this=0x0000555556cd1258)() const at mem_fun.h:1991:48 frame #16: 0x0000555555d4ba80 darkradiant`sigc::adaptor_functor<sigc::bound_mem_functor0<void, ui::MainFrame> >::operator(this=0x0000555556cd1250)() const at adaptor_trait.h:256:20 frame #17: 0x0000555555d4b22c darkradiant`sigc::internal::slot_call<sigc::bound_mem_functor0<void, ui::MainFrame>, void>::call_it(rep=0x0000555556cd1220) at slot.h:483:35 frame #18: 0x0000555555aec9e2 darkradiant`sigc::internal::signal_emit0<void, sigc::nil>::emit(impl=0x0000555556a31300) at signal.h:798:79 frame #19: 0x0000555555aee4a8 darkradiant`sigc::signal0<void, sigc::nil>::emit(this=0x0000555556163398) const at signal.h:2804:32 frame #20: 0x00007fffeba2acfb libradiantcore.so`module::ModuleRegistry::loadAndInitialiseModules(this=0x0000555556163320) at ModuleRegistry.cpp:192:32 frame #21: 0x00007fffeba8bc87 libradiantcore.so`radiant::Radiant::startup(this=0x00005555562af760) at Radiant.cpp:97:58 frame #22: 0x0000555555be315f darkradiant`RadiantApp::onStartupEvent(this=0x00005555560a4660, ev=0x00005555562dc8d0) at RadiantApp.cpp:230:30 This is the actual line of code which fails in PropertyNotebook.cpp. It seems that existingIndex is actually -1 here.

I don't see how DoF can possibly work in a game which isn't tracking your eyes. How does the game know what you are looking at? Your eyes could be focussed anywhere on screen, which might show objects at a variety of distances. If the depth is taken from the central aim point, that means that instead of just moving your eyes to look at something on screen, you'd also need to move the mouse to remove the artificial blur. DoF looks nice in photos, and as stgatilov says it might work in a VR simulation with eye tracking, but for a 2D image on a standard screen I can't see it adding any value.

I'm a big fan of Rust. It's an well-designed language with excellent tooling. But it doesn't have great GUI support yet (although there are a few toolkits available), and integrating it into a larger C++ application is non-trivial (although possible). Designing a new loader from scratch in Rust with an Electron-based front-end — that might actually be possible, although it would probably need somebody with more than my beginner-level Rust knowledge.

In addition to what Greebo says about the center view, note that you can freely dock floating widgets on the right as well as the left (or the top or the bottom). Just drag them into position and you will get a visual cue when they are ready to be docked.

An excellent step forward. After the recent discussion in one of the newbie threads about limitations of the dockable layout, I was actually wondering whether and when should look into this, but you're way ahead of me as usual.

What do you use to drag out a new brush?

I actually had a go at this myself once, but found it far too difficult. Personally I too hate the default input style, particularly the requirement to constantly hold down modifiers just to select things, but changing the defaults just breaks too much stuff because the selection system isn't designed to work that way. In order to allow "normal" selection rules, we would also need: A different method to create a new brush (since left-drag is now a selection or manipulation operation), for example something like Blender's Shift+A shortcut. A different way to resize brushes. Currently dragging anywhere outside the brush performs a resize, but if clicking/dragging on empty space is supposed to define a new selection, this will no longer work. So we would need something like traditional resize handles instead of the planar dragging approach. The selection system to decide what operation is being performed after the initial click or drag starts (as Greebo already mentioned). [Optional but almost certainly necessary] A system for setting up alternative keyboard profiles so people could choose between classic Radiant selection or the new system, since not everybody would agree with the changes.

The odd thing is that GTK itself can clearly distinguish the two cases, because I see different "drop destination lines" when dragging to the top of the view (one line above) versus dragging onto the first layer (one line above and one line below). So wxWidgets itself is actually swallowing up the event, even though it would actually be useful and consistent with the Windows behaviour, and the purpose of wxWidgets is for creating cross-platform UIs? I can't really see the logic of that, but I guess it was probably introduced to fix some bug without really considering the consequences. My suggested UI would be to add a couple of buttons (which are more discoverable than a context menu), perhaps next to the existing rename and delete buttons since the New button already has more than enough width. One button would move the selected layer up by 1 in the hierarchy, and the other would move it all the way to the top.

To be honest I can't remember ever using TAB to choose between items in a group. I certainly wasn't aware of it when writing the user guide, which is why the section on converting to func_static mentions using TAB, but the section on grouping doesn't. But it would certainly be useful and consistent behaviour.

I can reproduce this. It doesn't surprise me much since these legacy GtkRadiant layouts are hardly maintained or tested these days. I recommend using the Dockable layout instead and dragging the dock windows to the positions you want. I also see the checkboxes not doing anything. Drag and drop partially works for me, although I can only create a parent relationship. I cannot find a way to unparent a layer and bring it back to the top level. I think the Dockable layout would avoid that problem too (because the media browser would no longer be a top-level window).

I'm afraid you may be on your own with this one. Using dodgy unofficial compiler patches to hack in a new way of passing exceptions between DLLs (which is generally risky and prone to silent failure even when using a sane compiler setup), and then using this to target a very non-standard platform/package combination (GTK on MSYS/Windows), sounds like such a horrific maintenance nightmare that I wouldn't personally touch it with a ten foot pole.

These are problems in the source code and/or libraries used during building (e.g. libgit-dev). They have nothing to do with the Git version or command you used to check out the DarkRadiant git repository itself. Git is used internally to build the version control functionality inside DarkRadiant, and it is this version control plugin which is failing to build.

I don't even understand what this code is about. What is the "TDM based GCC compiler"? The TDM team doesn't write C compilers, does it? Why would we need to deal with such low-level mutex and shared memory code?