Crash on OSX about pyimgui

[swistakm]

Yeah, imgui requires proper initialization and careful new_frame() / render() functions handling. We should definitely add some guards to protect from crashes like this one. Anyway, such usage will still raise exceptions. Did you try to run examples from doc/examples? You should take a look at them.

[jeremyherbert]

Yep, examples work fine. I was more reporting this because the error message could perhaps be better.

[swistakm]

Yeah of course this should be fixed. Your lldb traceback shows already that there is some python assert exception raised from C++ code. This will require updating some function signatures in cython sources.

[swistakm]

If we update all Cython signatures as following:

diff --git a/imgui/cimgui.pxd b/imgui/cimgui.pxd
index c2877ff..7ced93c 100644
--- a/imgui/cimgui.pxd
+++ b/imgui/cimgui.pxd
@@ -283,7 +283,7 @@ cdef extern from "imgui.h" namespace "ImGui":

     # ====
     # Window
-    bool Begin(const char*, bool*, ImGuiWindowFlags)  # ✓
+    bool Begin(const char*, bool*, ImGuiWindowFlags) except + # ✓
     # note: following API was deprecated
     # bool Begin(const char*, bool*, const ImVec2&, float, ImGuiWindowFlags)
     void End()  # ✓

Then running functions without Imgui context being properly initialized will result in imgui.core.ImGuiError exception. For instance, the Python traceback for imgui.begin() function can be following:

Traceback (most recent call last):
  File "issue.py", line 4, in <module>
    imgui.new_frame()
  File "imgui\core.pyx", line 993, in imgui.core.new_frame (imgui/core.cpp:15179)
    cimgui.NewFrame()
imgui.core.ImGuiError: ImGui assertion error (g.IO.DisplaySize.x >= 0.0f && g.IO.DisplaySize.y >= 0.0f) at imgui-cpp/imgui.cpp:2046

@jeremyherbert is such traceback more helpful?

[swistakm]

I have the made required changes on fix/full-exception-coverage branch. I need to test it extensively but it should fix this issue.

[swistakm]

Fix merged. Now every C++ assert should be treated as imgui exception.