Windowresized and Surface

Do I need a SDL_DestroySurface when I get a new Surface?
In my example, when you change the size of the window.

                case SDL_EVENT_WINDOW_RESIZED:
                    if (winSurface) {
                      SDL_DestroySurface(winSurface); //  necessary ?
                    winSurface = SDL_GetWindowSurface(win);
                    SDL_FillSurfaceRect(winSurface, nullptr, 0);

A new surface will be created with the optimal format for the window, if necessary. This surface will be freed when the window is destroyed. Do not free this surface.

1 Like

I should have mentioned that I mean SDL3.
But according to the wiki there doesn’t seem to be any difference, except that SDL2 mentions that you should re-get WindowsSurface after every UpdateWindowSurface.

It’s a shame that the examples in the wiki don’t show how to do it with SDL_PollEvent.

What makes the use of this function inside a SDL_PollEvent loop special that it would motivate a special explanation and/or example?

Both the SDL2 and SDL3 wiki page clearly says “Do not free this surface”. This is always the case.

I would like to clarify for anyone who’s mind jumped to “but wouldn’t that be a memory leak?”:
If it detects that the said window surface is no longer valid (such as when the screen size has changed) the function SDL_GetWindowSurface calls SDL_DestroyWindowSurface to destroy the internal surface pointer held by the window .

SDL_Surface *SDL_GetWindowSurface(SDL_Window *window)

    if (!window->surface_valid) {
        window->surface = SDL_CreateWindowFramebuffer(window);
        if (window->surface) {
            window->surface_valid = SDL_TRUE;
            window->surface->flags |= SDL_DONTFREE;
    return window->surface;

No leak, the memory block is recovered by SDL_GetWindowSurface() before the new surface framebuffer is created and returned.

Edit: Also, the SDL_DONTFREE flag ensures that if you do try to destroy the surface before calling SDL_GetWindowSurface, the function SDL_DestroySurface will ignore the call, so it won’t crash your program due to a double deallocation. So you could do it either way, but doing it the “right way” will likely save about one or two dozen CPU cycles each time the event fires.

1 Like