Emulator

On the one hand, r7607 seems to have taken care of the two error cases described above.

On the other hand, however, there's still another way to create "renderer window leaks", involving the fact that the "Play" button isn't disabled/blocked ASAP upon initializing emulation. Two examples:

Example A¶

Steps 1. 2. 3. 4. and 5. are the same as before, but don't close both error messages yet;
6. In this state, press the "Play" button once again. Another renderer window opens, then we see "Emu Thread already running" and "Could not init the core."
7. Close these two error dialogs, which will close the second renderer window;
8. Now close the other error dialogs from the first renderer window. Notice how it won't close properly, unlike the second one.
9. Now, any time you press the "Play" button, you'll get "Emu Thread already running" and a window that appears and close properly. Thus, this way will allow the "renderer window leak" to occur only once.

Example B¶

Steps 1. 2. 3. 4. 5. and 6. are the same as in example A;
7. Close the error dialogs from the first renderer window, the one with the DSP-LLE errors. This one will close properly;
8. Do the same for the second renderer window. That window will be leaked.
9. Now, notice how pressing "Play" to create another renderer window won't result in "Emu Thread already running" but instead the DSP-LLE initialization errors.
10. This means that by repeating steps 5 through 9, it is possible to leak more windows... This process will stop once you do the mistake of closing out the "Emu Thread already running" errors before the DSP-LLE initialization errors.

Dolphin dislikes having more than one rendering frame created at once, so preventing that from happening should be good. One or two flaws may need to be fixed, the more important one being the one with the "Play" button (the other one being the difference between examples A and B).

Alright, I admit I wrote too much reproduction details about these "possibly windows-specific, trivial, extremely rare cases", which made the "conclusion" I made rather unclear... (I wrote quite a lot of posts during these days, including bug reports; should probably begin revising for my upcoming (written) French exam)

I guess I'll conclude by saying that this extra problem (and possibly a few other problems due to being able to manage only one pointer/handle) may be avoided, by properly controlling the action of the "Play" button so it doesn't give the possibility of spawning more than one "rendering frame".

This shouldn't be too complex to do, nor windows-specific (easier said than done, however); but yeah, that issue is quite different from the one here, not to mention that this issue mainly focuses on a particular regression... I suppose I'll mark this as fixed, then. (Though I am not sure I am willing to spend some time writing a report for this issue, at least not now...)