On Mon, 18 Dec 2017 at 23:41:50 +0000, Simon McVittie wrote:
> In 2.54.2-3 I've cut that test down to 1% of the usual repetitions on
> arm64, which seems like enough to prove that this functionality does
> basically work. The fact that it's far slower than on other CPUs is
> still a bug, but hopefully one that can be downgraded to important.
Not doing enough repetitions made the test fail on the reproducible
builds infrastructure, where the main thread starved the other threads,
causing an assertion failure:
FAIL: closures
==============
START: /build/1st/glib2.0-2.54.2/debian/build/deb/tests/refcount/.libs/closures
init 0xaaaad66ca810
.
stopping
stopped
**
ERROR:../../../../../tests/refcount/closures.c:289:main: assertion failed:
(seen_thread1 != FALSE)
FAIL closures (exit status: 134)
I've changed this to 10% of the usual repetitions on arm64 for 2.54.2-4.
Hopefully that will be OK (it should take something like 20 minutes on
arm-arm-04, which is slow but not awful).
smcv
