See bug 665832 comment 4: On Linux, when we "optimize" a decoded image, we upload it to the X server. We then free the memory inside Firefox, so the decoded image's size is not accounted for anywhere in about:memory. This memory is still being used, so I think we should account for it in about:memory.

So this would be a separate non-heap reporter of some sort?

(In reply to comment #1) > So this would be a separate non-heap reporter of some sort? Correct. AFAICT the decoded image is not only outside the heap, it's not in our process's virtual memory, so it would be a completely separate reporter.

It'd have to be a MR_OTHER reporter, because it's neither MR_HEAP nor MR_MAPPED.

Alternatively, we could report on the X server process separately. A bit like bug 648415. That might be difficult, though.

(In reply to comment #4) > Alternatively, we could report on the X server process separately. A bit > like bug 648415. That might be difficult, though. FWIW, the xrestop program is able to give some interesting information on the X server memory usage, pixmap counts, etc.

My patch for bug 670967 is going to fix this.

Whoops; wrong bug. I meant bug 664659.