On 3/22/26 14:55, Bruno Haible wrote:
Hi Dennis,
a 1997 Digital DEC AlphaStation 600
...
Alarm clock
FAIL test-regex (exit status: 142)
Anyways, I think the test should be safe to ignore. Here is the alarm
snippet from the test:
#if HAVE_DECL_ALARM
/* In case a bug causes glibc to go into an infinite loop.
The tests should take less than 10 s on a reasonably modern CPU. */
int alarm_value = 1000;
signal (SIGALRM, SIG_DFL);
alarm (alarm_value);
#endif
I don't think your system has a "reasonably modern CPU", no offense. But
maybe Paul, who wrote this comment, has a different definition than
mine. :)
I wouldn't ignore/discard this report immediately because
- A test that should take less than 10 seconds on a modern CPU
should take less than 100 seconds on an old alpha machine.
Not 1000 seconds.
- Even on my slowest QEMU-emulated VMs, I don't recall having seen this
test fail.
My gut feeling is that it could be some gcc compiler bug.
If you want to spend time narrowing it down, I would modify the 'alarm_value'
and the CFLAGS. Or run the test under gdb and see what it is doing after 900
seconds.
I hacked in 60000 as the number and all tests pass just fine. I time the
process of all the tests on this old machine and the diff is not much :
(1) with the single failure due to ALARM
real 3771.58
user 2600.41
sys 593.13
(2) with the hacked in 60000 value
real 3150.73
user 2270.45
sys 426.53
I *may* extract the tarball fresh and redo the whole game again
just to be more clear. However I get a good result and am happy.
--
--
Dennis Clarke
RISC-V/SPARC/PPC/ARM/CISC
UNIX and Linux spoken
