On Sun, May 08, 2022 at 10:21:28AM +0100, David CARLIER wrote: > On Sun, 8 May 2022 at 09:57, Willy Tarreau <w...@1wt.eu> wrote: > > > > On Sun, May 01, 2022 at 03:33:17PM +0100, David CARLIER wrote: > > > Hi here a little patch to set idle time for SO_KEEPALIVE socket option. > > > > Now merged, thanks. > > > > David, one comment though, your commit messages keep missing a lot of > > crucial information for reviewers and debuggers, and I had to spend time > > documenting myself on keep-alive on MacOS to figure the differences and/or > > what the impacts or limitations of the patch could be. I finally found and > > put that information into the commit message, but it would be much easier > > if you could put it yourself after your development since you actually > > have access to the information I had to seek, and know the reasoning > > behind your choice. > > > > Thanks, fair point :-) I ll take it in account
Thanks ;-) > even tough I was certain for this one it would not break anything. Quite frankly, experience tells us we really never know. If MacOS in the future version would define the TCP_KEEPIDLE macro, that would be sufficient to break the build for example. > > The new choice was explained again. Finally, two weeks later I got a > > report of breakage when using external code linked at run time, the > > keyword registration didn't work anymore due to a mistake in this last > > patch. Fortunately, the design decision was explained and I could > > restart from this, figure my mistake and make sure that the 3rd attempt > > at a fix this time addressed all 3 identified use cases: > > Do you have a reasonable numbers of macOs users or is it a rare occurrence ? I really don't know, here it was on the CI. Outside of the CI, breakage is usually reported in this order: - Linux - FreeBSD - MacOS - others, in random order Thus I guess it translates the number of occurrences in field. I suspect that a number of admins have macs and are used to build it locally to test their configs, and that it's probably a non-negligible part of the MacOS reports. But as usual, I could be totally wrong ;-) Willy