Addcallback, assert Size > 0 ?

[voidware]

I'm having a problem with AddCallback. Not sure if i'm doing something wrong;

Works;

  if (ImGui::Begin("Dear ImGui", 0, 0))
  {
        ImGui::GetWindowDrawList()->AddCallback(draw_scene_1, 0);
  }
  ImGui::End();

Fails:

 if (ImGui::Begin("Dear ImGui", 0, 0)) {

        if (ImGui::BeginChild("sokol-gfx", ImVec2(360, 360), true))
        {
            ImGui::GetWindowDrawList()->AddCallback(draw_scene_1, 0);
        }
        ImGui::EndChild();
 ...
 }

Works;

   if (ImGui::Begin("Dear ImGui", 0, 0)) {

        if (ImGui::BeginChild("sokol-gfx", ImVec2(360, 360), true))
        {
            ImGui::Text("hello!");
            ImGui::GetWindowDrawList()->AddCallback(draw_scene_1, 0);
        }
        ImGui::EndChild();
   ...

When it fails, i get assert Size > 0 from imgui.h ImVector::front.

Adding the bogus "hello!" text fixes it.

Is this something to do with the child draw list being otherwise empty, should it be doing this, or should i be doing something different?

Thanks for your help.

[PathogenDavid]

I'm not getting a crash with your provided repro. Can you provide a call stack of the crash?

Also what version of Dear ImGui are you using?

[voidware]

I'm using the github master branch. It's not a crash, imgui asserts Size > 0 in ImVector. I think it's because i otherwise have nothing to render. I'll see if i can get a stack. It might also be worth noting the callback (draw_scene_1 in this case) does not get called.

[voidware]

Hang on, i just realised it might not be imgui. I'm using sokol to render the backend and it could be that.

[voidware]

#1 0x00000000004d08a1 in ImVector::front (this=0x27bf768) at v:/sw/imgui/imgui.h:1727
#2 0x00000000004ac9e7 in simgui_render () at ../../sokol/util/sokol_imgui.h:1930
#3 0x00000000004ae19a in frame () at imgui-tex.cpp:211
#4 0x000000000040165a in _sapp_call_frame () at ../../sokol/sokol_app.h:2333
#5 0x0000000000401e7d in _sapp_frame () at ../../sokol/sokol_app.h:2512
#6 0x0000000000405e38 in _sapp_win32_run (desc=0x81fca0) at ../../sokol/sokol_app.h:6592
#7 0x00000000004060d2 in WinMain (hInstance=0x400000, hPrevInstance=0x0, lpCmdLine=0xa13ebf "", nCmdShow=10) at ../../sokol/sokol_app.h:6674
#8 0x00000000004013c7 in __tmainCRTStartup ()
#9 0x00000000004014fb in mainCRTStartup ()

[voidware]

Yep, sorry my bad. this is a sokol problem.
Here's the fix for anyone interested;

sokol_imgui.h::simgui_render

change front to begin

       const size_t vtx_size = cl->VtxBuffer.size() * sizeof(ImDrawVert);
            const size_t idx_size = cl->IdxBuffer.size() * sizeof(ImDrawIdx);
            
            //const ImDrawVert* vtx_ptr = &cl->VtxBuffer.front();
            //const ImDrawIdx* idx_ptr = &cl->IdxBuffer.front();

            const ImDrawVert* vtx_ptr = cl->VtxBuffer.begin();
            const ImDrawIdx* idx_ptr = cl->IdxBuffer.begin();

[PathogenDavid]

Edit: Looks like you beat me to it.

Yeah, this is a bug with the Sokol backend:

https://github.com/floooh/sokol/blob/667df71044ecd19c218c93c5cbdb188c9ed6eeb8/util/sokol_imgui.h#L1930

This is unconditionally calling front() regardless of whether there's actually any vertices to render. Contrast this with the official backends, which just access the Data pointer directly:

imgui/backends/imgui_impl_opengl3.cpp

Line 427 in 17a7084

glBufferData(GL_ARRAY_BUFFER, (GLsizeiptr)cmd_list->VtxBuffer.Size * (int)sizeof(ImDrawVert), (const GLvoid*)cmd_list->VtxBuffer.Data, GL_STREAM_DRAW);

[voidware]

Yes indeed. Thanks for helping on this. I've posted by changes to the sokol repo in case they want them. It shouldnt call front regardless. If you look at the "C" version, it does the right thing.

[PathogenDavid]

No problem!

I've posted by changes to the sokol repo in case they want them.

In the future you should really provide context on why the changes are necessary. It's not fair to floooh to expect them to read your mind on why begin() is more appropriate than front().

(Also you can probably safely mark this issue as closed.)

[floooh]

My bad, this is now fixed in sokol_imgui.h. Many thanks for investigating the problem!