On Fri, Aug 12, 2016 at 1:53 AM, Roderick <hru...@gmail.com> wrote: ... > What did change here from OpenBSD 4.8 to the current versions? Is it an > esential change?
Yes, the previous situation with <machine/endian.h> and <sys/endian.h> was confusing (code was including the wrong header and not getting the optimized version that was expected) and there was finally a proposal to provide this functionality that was accepted for the next version of POSIX. When the existing situation is bad and there's a good direction finally settled on, moving that way is the Right Thing. Why did libressl break? Because the old <endian.h> setup assumed you pulled in <sys/types.h> first, which was a BSD-specific bug, and once we fixed that bug no one noticed when other changes meant LibreSSL stopped avoiding that bug. To keep that from happening, Brent would have to have an even longer list of setups to test on. Is that the best use of Brent's time, to make the code more complicated and do more testing for a release that is officially "out of support" and for which none of the users are interested in doing that testing? No! Wait, make that HELL NO! ... > But the question: is really till now, after so many years BSD, > no standard for getting this parameter? Or does things change from > time to time for the sake of change? There's been an official standard for years, but getting OpenBSD in line with it has required lots of work. You're running a version before many of those changes, a version that *doesn't* provide the de jure standard method! OpenBSD is going to continue moving forward and discarding the baggage of the past as fast as we can: walking consists of falling forward and catching yourself over and over again. If you need stability, to be able to stand still for years, then OpenBSD is not a good match for your needs. Philip Guenther
