Hi Marcel, > > Also, I don't really buy the *need* for this since you're just removing > > a few kernel/user roundtrips here when new devices are discovered, a > > rare event. The parsing isn't really any more complicated for the > > userspace side. > > that is an argument that is coming to bite you. Forcing multiple > roundtrips or even collecting multiple split message for some ancient > legacy client behavior is just silly. If clients provide larger > buffers, we should start using them. I'm not arguing legacy/old client behaviour. > I have proven a long time ago that round-trips are causing delays and > creating visible user experience issues. Look up my DHCP presentation > from either LinuxCon or PlumbersConf. One round-trip leads to another > and at some point you end up with seconds wasted because you want to > sit here and ignore efforts in improving the situation. Comparing network roundtrips to local kernel access isn't exactly a very good comparison. > > And finally, I also see no reason to send out many KB of data for what > > might in the end (e.g. in iw) just be a debug message. > > Actually iw is just a dev tool. It should not be run in production and > so that is not an argument. Any proper client that cares about your > WiFi connections will want this information. Again, this isn't an argument. I said wpa_s is an example. Any other number of tools works, even wpa_s. Heck, probably even iwd, when configured to not care about some devices (unless you can't even make it ignore devices, which I'd consider a deficiency in its own right). > > But really I think the thing that kills this proposal is the fact that > > it reintroduces a message size limit (even if higher now) that we're > > somewhat likely to hit in the future. > > Maybe we need to accept that current nl80211 API is broken and start > over. Or we should at least start deprecating commands and replacing > them with new ones that are doing a better job for clients that > actually behave properly. I know you love throwing things away and rewriting them, but you're not going to solve the problem. I suggest you re-read my email and actually reply to it, rather than throwing out bullet points. Frankly, I'm tired of having a discussion where all you do is accuse me of not caring about the problem, but then you don't even respond to any arguments. johannes
