On Sat, May 30, 2020 at 09:59:41AM -0600, Theo de Raadt wrote: > Paul Irofti <[email protected]> wrote: > > > > A few more notes below. > > > > I addressed all the comments. Here is the updated diff. This includes > > the rename to usertc that I suggested. > > I want to see support for quite a few more architectures, especially > those which are very different, because changing format of the shared > page later will be very painful.
Sure. Your call. This last diff will help that as discussed with kettenis@. We now have usertc.c which should be the only place that needs to be touched by each arch. Let's see. I am currently looking at doing this on an octeon or a loongson. Kettenis said he will do arm64. > > The libc bump is there because it helps me switch more easily between > > versions. > > That is bogus. Minors are used for visible ABI additions, majors are > used for ABI deletions or API changes visible as ABI. Please don't > argue for a vague extension of the rules again. I do not know what you are taking about here. I am not looking at any extension of the rules, nor was I in the past. The whole issue of bumping I leave it up to you and whoever understands these rules. Some developers said this is not required, including kettenis@, and this is why I justified the bump in my diff. That and it might also help others quickly test the diff. > In essence this introduction requires no major or minor crank becuase > it just starts selecting a different backend which is newly supplied. > But as soon as the back-end is changed, the version number will barely > help, since code which can't match it has to revert to the non-optimized > path. > > I don't believe you can shortcut this by supporting 1 architecture and > casting a prayer it's going to be fine. I am not trying to shortcut anything. I am in no rush for anything.
