Re: Still seeing Failed assertion: "p[i] == 0" on armv7 buildworld

On Nov 13, 2025, at 01:52, Ronald Klop <ronald@FreeBSD.org> wrote:

> Op 13-11-2025 om 10:07 schreef Carl Shapiro:
>> Ronald Klop <ronald@FreeBSD.org> writes:
>>> My thought was triggered by this as a build of opendjk11 failed with a
>>> jemalloc error.
>>> https://lists.freebsd.org/archives/freebsd-pkg-fallout/2025-September/804963.html
>> Is this build failure very reproducible?  Is there more of a stack trace
>> to go with it?
>> When the jemalloc witness code observes a locking error the process
>> should abort immediately with a SIGABRT.  However, there is SIGBUS
>> reported in the build output prior to the witness error which makes it
>> look like OpenJDK may have been handling a signal while the witness code
>> was running.  If malloc is somehow being called from a signal handler
>> that is asking for trouble.
>> Here's a closed issue from the old jemalloc repository about a witness
>> error when malloc was called from a signal handler
>> https://github.com/jemalloc/jemalloc/issues/1224
> 
> 
> Hi,
> 
> I only have this example. I don't run armv7 myself.
> Unfortunately the armv7 pkg builders don't run that often.
> 
> This is the only failure on main-armv7, but AFAIK no new pkg build run for main-armv7 has happened since.
> https://portsfallout.com/fallout?port=java%2Fopenjdk&maintainer=&env=armv7&category=&flavor=

armv7 is tier 2 and so does not have the guarantee: "Official binary
packages  or third party software will be provided by the ports team."

It turns out that various armv7 builder activities have been suspended
as of the tier 1's having to deal with 13.5, 14.3, 15.0, and main for
latest and quarterly on the same builder machines that otherwise would
also be building all that for armv7 as well. It is tied to how long it
takes for the 3 ampere*'s to do builds and how few builders there are
vs. the count of variations built: large latencies between rebuilds,
well beyond just how long the specific prior build took because of
other builds that also take a long time.

https://people.freebsd.org/~dbaio/pkg-master-report.html

shows the most recent builds for armv7 latest and quarterly ending
at about:

freebsd:13:armv7:32:el:eabi:hardfp latest    2025-10-05 04:47:33Z
freebsd:15:armv7:32:el:eabi:hardfp latest    2025-09-26 05:59:09Z (for prerelease)
freebsd:14:armv7:32:el:eabi:hardfp latest    2025-09-23 04:16:17Z
freebsd:13:armv7:32:el:eabi:hardfp quarterly 2025-09-16 08:07:05Z
freebsd:14:armv7:32:el:eabi:hardfp quarterly 2025-09-03 11:10:32Z

(Sending to the world's distribution servers was after those times.)

The kmods_* builds are separate and armv7 are still built with
the others. But kmods_* cover an incomplete subset of the full
set of kernel specific port-packages, if I understand right.

> I just noticed that the full build log is also already gone from the pkg build server.


Side note on use of:

https://people.freebsd.org/~dbaio/pkg-master-report.html

Clicking column titles in the order of "subsorts the next title to be
clicked" allows control the overall subsort order across the columns.


===
Mark Millard
marklmi at yahoo.com