Boring notabug report, but..
(Dr. Dr., perf acts all dainbramaged when you shoot it in the head)
reproducer: perf record -a -g& sleep 1;whatever;killall perf
Discovered because AIM7 needs input, and does not run properly when
profiled via the more conventional perf record -a -g -- ./multitask -nl
-f. Ergo, start perf record, run AIM7, kill perf. Say SIGINT instead,
and it works as expected. SIGTERM leaves perf.data in a bad state that
report can't deal with, but it'll happily keep trying forever.
Only mentioned because some other user might try to workaround input
nogo thingy and meet the same.
-Mike
On 4/29/13 4:00 AM, Mike Galbraith wrote:
> Say SIGINT instead,
> and it works as expected. SIGTERM leaves perf.data in a bad state that
> report can't deal with, but it'll happily keep trying forever.
header is usually updated on_exit which requires a clean exit. perhaps
we should add SIGTERM to the signal list.
David
On Mon, 2013-04-29 at 07:27 -0400, David Ahern wrote:
> On 4/29/13 4:00 AM, Mike Galbraith wrote:
> > Say SIGINT instead,
> > and it works as expected. SIGTERM leaves perf.data in a bad state that
> > report can't deal with, but it'll happily keep trying forever.
>
> header is usually updated on_exit which requires a clean exit. perhaps
> we should add SIGTERM to the signal list.
Yeah. Graceful exit, and idiot proofing for free :)
-Mike