On Fri, 6 Jun 2025, Fred Wright wrote:
On Fri, 6 Jun 2025, Riccardo Mottola wrote:
Fred Wright wrote:
The latest legacy-support-devel (v20250530) is a release candidate for the
new version, so it would be useful to get some broader testing. The most
significant changes are the complete rewrite of all the clock functions
for better speed, accuracy, and compatibility.
Since there are some cases where the architecture is relevant, and my G5
is inoperative, it would be useful if someone could test on a G5,
preferably running 10.5. That's just a matter of running "port test
legacy-support-devel" while targeting ppc64 as the architecture. Since
build_arch is usually set to ppc rather than ppc64 on a G5 (for good
reason), this would require either overriding build_arch on the command
line or testing +universal (provided that universal_archs includes ppc64).
Since the latter is also unlikely, the first approach is probably the most
convenient.
would it be of interest on 10.5 x86-64 too?
I could also on G4 and regular x86. No G5 though.
As far as arch/OS combinations go, I have those covered, though more testing
is always welcome.
The current status of this is:
1) In investigating the G5 problem, I found a somewhat minor (and easily
fixed) performance bug in the new mach_continuous_time() and friends.
2) I'm working with Sergey to figure out the G5 problem, which is fairly
bizarre. If there's someone else with a G5 that can be set up for
(nonprivileged) remote SSH access, that might save having to work around
Sergey's schedule.
So v20250530 is no longer a release candidate, but the only anticipated
changes are for the two issues above, so that's no reason not to continue
testing it.
To expedite the new release without being blocked on the G5 problem
(more precisely, the ppc64 problem), there's now a new -devel version
(v20250608) with only two changes since v20250530:
1) The aforementioned performance bug is fixed.
2) The sleep-time accounting is disabled in ppc64 builds. Since previous
versions of legacy-support didn't provide any sleep-time accounting at
all, this isn't really a regression.
In view of the limited scope of the changes, this won't need the usual
one-week soak time before the release, but some amount of field testing
wouldn't hurt.
Fred Wright
