> You are arguing for an atomic kernel operation. Is that really fair?
No, actually, I'm simply pointing out that saying "there would be no possibilities for errors" was, as it stood, incorrect -- and your argument depended on it being correct. It's a minor point, but I thought I made that clear in my post. > You're not leaving my argument any flexibility to do anything at all. No, I'm simply not letting you get away with anything :-) You need to *admit* that it is an additional complexity, and that complexity has costs, and then show that the benefits outweight the costs. You can probably do that -- but saying "it can't fail" when it *can* is very weak ground. > Essentially, you are telling me to shut up. Horrors, no, not at all. If I didn't think you had *something*, I'd just ignore you, like I do some other people here. I'm not trying to push your buttons, I'm just pointing out the problems I see. Keep in mind that I do want better source tools -- but I've seen lots of "clever hacks" too, and that's not likely to be good enough. They have to *last*. I think you should consider this (and some other feedback that I've seen here) and perhaps go out of your way to read your respondents email a bit more closely, and try to assume less antagonism on their part. It's another easy trap to fall into (usually someone does, here or on debian-devel, probably once a month or so.) Sometimes it helps to imagine you're talking to people over a beer. > It's a superset - so it might be overkill - not optimized for Ah, see, I *don't* think dpkg is a superset of what you need. You've already shown a few examples of how dpkg *doesn't* work well for this case (the way of handling install-in-non-system-dir stuff, for example.) Not only is it "not optimized for performance" it's simply *missing* lots of things that would be important to a good source package design -- and since they're missing at the design level, the fact that details *appear* to be sufficient is illusory. > I already demonstrated that dpkg works just fine in user space. It Demonstrating that it works in a particular example with a bunch of advance setup only shows that it *takes* a bunch of hacking to make it work, not that it actually helps. > Anyways, I'm not interested in discussing this stuff anymore. It's > been interesting, but not very productive. Glad it's been interesting, sorry you haven't gotten more out of it...
