Use the POSIX timers API rather than itimer on FreeBSD

[ararslan]

POSIX.1-2008 marks getitimer/setitimer obsolete in favor of timer_gettime/timer_settime. Additionally, POSIX.1-2017 marks SIGPROF obsolete. Thus we can make simply have FreeBSD use the same signals code paths as Linux, which already uses the timer API and uses SIGUSR1 rather than SIGPROF. The code conditional on the HAVE_ITIMER flag, as well as the flag itself, have been removed since they're no longer used.

Inexplicably fixes #48714 for me.

[ararslan]

Weird, I did not get this locally:

cmdlineargs                                      (4) |         failed at 2023-03-21T02:00:25.487
Error During Test at /usr/home/julia/.buildkite-agent/builds/freebsd12-amdci6-1/julialang/julia-master/julia-044401cfcd/share/julia/test/cmdlineargs.jl:303
  Test threw exception
  Expression: string(cpu_threads) == read(`$exename --threads auto -e $code`, String) == read(`$exename --threads=auto -e $code`, String) == read(`$exename -tauto -e $code`, String) == read(`$exename -t auto -e $code`, String)
  failed process: Process(`/usr/home/julia/.buildkite-agent/builds/freebsd12-amdci6-1/julialang/julia-master/julia-044401cfcd/bin/julia -Cnative -J/usr/home/julia/.buildkite-agent/builds/freebsd12-amdci6-1/julialang/julia-master/julia-044401cfcd/lib/julia/sys.so --depwarn=error --check-bounds=yes -g1 --startup-file=no --startup-file=no --color=no -t auto -e 'print(Threads.threadpoolsize())'`, ProcessSignaled(6)) [0]
   
  Stacktrace:
   [1] pipeline_error
     @ ./process.jl:565 [inlined]
   [2] read(cmd::Cmd)
     @ Base ./process.jl:449
   [3] read(cmd::Cmd, #unused#::Type{String})
     @ Base ./process.jl:458
   [4] macro expansion
     @ /usr/home/julia/.buildkite-agent/builds/freebsd12-amdci6-1/julialang/julia-master/julia-044401cfcd/share/julia/stdlib/v1.10/Test/src/Test.jl:478 [inlined]
   [5] top-level scope
     @ /usr/home/julia/.buildkite-agent/builds/freebsd12-amdci6-1/julialang/julia-master/julia-044401cfcd/share/julia/test/cmdlineargs.jl:303

[giordano]

We've seen that error on other platforms as well. CC @gbaraldi

[gbaraldi]

Yeah, we usually get a segfault here when it errors, but it seems we got a sigabrt.

[vtjnash]

Do you want to use SIGPROF still, via setting:

    sigprof.sigev_signo = SIGPROF;

[ararslan]

Do you want to use SIGPROF still

Nah, might as well just follow suit with what Linux is doing and use SIGUSR1 since it seems to work just fine and SIGPROF is apparently obsolete now. Also, I'm not sure why this works now when it was not working for me earlier as in our discussion in Slack, but ¯\_(ツ)_/¯

We've seen that error on other platforms as well.

Which ones? Is it sporadic or consistent? (I assume sporadic given that CI passed on other platforms in this PR.) We could try rerunning the FreeBSD tests and see whether it happens again.

Yeah, we usually get a segfault here when it errors, but it seems we got a sigabrt.

Huh. Well, I guess this isn't worse at least...

[ararslan]

The cmdlineargs test failure didn't occur on the run triggered by the merge so I guess it's sporadic here as well

[giordano]

I opened an issue about the cmdlineargs test failure: #49154