[danoli3]

Summary

finishRender() cleared the framebuffer stack at the end of every frame. The next ofFbo::end() then logged:

[ error ] unbalanced fbo bind/unbind binding default framebuffer

even when the window was already bound. A real forgotten end() was only reported on the following frame. The same path is in ofGLRenderer and ofGLProgrammableRenderer. bind() also returned early when an ofFbo was bound onto itself, so the matching end() under-popped the stack.

Both renderers now share one stack in ofFramebufferBinding.h:

• A matching begin() / end() inside the frame is silent. The window entry pushed by startRender() stays until finishRender().
• A forgotten end() warns once when the frame ends, with the framebuffer id that was still bound, and binds the window again. A later end() stays quiet.
• An extra end() while only that window entry is left warns once and does not pop it.
• Binding an ofFbo onto itself still warns, and still pushes, so its end() pops the same entry.
• startRender() binds the window framebuffer, so the GL binding matches the id the stack records.

The check at the end of the frame uses the bound id, not stack.back(). After begin() from the window the stack is [window, window] while the ofFbo is current, so back() == window would miss a forgotten end().

Fixes #8564

Test

tests/gl/fboBindStack runs the stack with no GL context (balanced pair, forgotten end(), extra end(), begin/end outside a frame, self-bind, nested ofFbos).

Local macOS makefile build of that test:

[ notice ] framebuffer bind stack passed
[ notice ] 1/1 tests passed

[danoli3]

This is an interesting way to handle FBO issues if detected by using fbo binding deque.

I had applied similar fixes for fbo textures binding in unity many years ago after leaks were detected in production game.

Code assisted with Grok for this one with my code examples linking to initial reported issue