On Thu, May 16, 2002, David Brownell <[EMAIL PROTECTED]> wrote: > > I think you still misunderstand the guarantees the reference counting > > code provide. > > Maybe you should just stop saying "misunderstand X" and > just start saying "disagree with JE about Y". It seems like > it's gotten to be a reflex with you. > > That'd really help improve the tone of discussion, since > I really do understand "X" in this case (and others). > Despite your constantly repeating otherwise.
You keep saying you understand it (X), but then you keep doing things that show you don't understand. Take for instance your implementation. It's plainly obvious that you didn't understand the design because you implemented it with the assumption that it guarantees being called from the thread context. There was never any such guarantee stated nor does the code imply that. Then take your most recent statement: "... in the face of buggy drivers (which we know happen with unfortunate inevitability :). Yes." I've proved that is not true in my previous email, which once again shows that you don't understand it. As a result, the fact I'm constantly repeating it is because you are constantly making statements that aren't true. > > The fact that it mostly happens in a thread context is to be considered > > dumb luck. > > No, the fact that it (bus->deallocate) doesn't _always_ happen in a > thread context is to be considered a bug ... :) Nope, it's not a bug because you don't understand it. The design, nor the implementation, give that guarantee, so the bug is clearly in usb-ohci and ohci-hcd. And the fact the it's called from a thread context now, with your recent patch, is just convenient since the only requirement for a thread calling context is simply your implementation and not anything else fundamental. > That's the point of disagreement. I wish I understood just why > you've gotten in such high dudgeon about a bug that doesn't > affect you. It's really annoying, to me and I'm certain everyone > else on the list. An on a cooperativity scale of 1-10 it's a zero. Listen, I really don't like to get into the kind of heated discussions we've had lately, but it has gotten to the point where you are just getting arrogant about it. You made an assumption you shouldn't have and now are claiming that it should have always been the way you mistakenly interpreted it. I can say you are mistaken because I'm the one who wrote it, so by definition I'm the authority on the semantics of the design and the implementation. I can also say you are mistaken because reading the implementation proves that you made incorrect assumptions. This is not a case of some subtle knowledge not being passed on, not like that would make your assumptions any more excusable. This could be a whole lot easier for the both of us. You could have said "I originally misunderstood the guarantees that the reference counting code provide, but I now see it doesn't provide the guarantees that I need". You could have then explained why and we could have had this fixed a long time ago. The requirements that OHCI need weren't apparent to me when I originally designed it, but now we've found a small modification which does give you the guarantees that you need. I'm sorry this whole thread had to come down to this email, but this has to come to an end, especially when accusations like being uncooperative are being tossed around wantonly. JE _______________________________________________________________ Have big pipes? SourceForge.net is looking for download mirrors. We supply the hardware. You get the recognition. Email Us: [EMAIL PROTECTED] _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
