Hi,

I tried to implement linux timer with timer_create and signals, but an exception always fired (External:SIGUSR1).
It was a lcl program, but I narrowed it to a console one. Here is the lines without extra codes of releasing (the error appears before),
What can be wrong with the code below?

Thanks in advance,

Max

==================================

program ConsTest;
uses ... libc, baseunix;
procedure PosixTimerHandler(signal: longint; info: psiginfo; context:
psigcontext);cdecl;
begin
  Now; // <- Never stops here.
end;
procedure TMyApplication.DoRun;
var
  ...
  fnewact: SigActionRec;
  fits: itimerspec;
  sev: sigevent;
  timerid: timer_t;
begin
...
  { add your program here }
  FillChar(fnewact, sizeof(SigActionRec),0);
  fnewact.sa_handler := @PosixTimerHandler;
  fnewact.sa_flags:=SA_SIGINFO;
  if fpsigaction(SIGUSR1,@fnewact, Nil)  0 then
    raise Exception.Create('sigaction failed');
  sev.sigev_notify := SIGEV_SIGNAL;
  sev.sigev_signo := SIGUSR1;
  timer_create(CLOCK_REALTIME, @sev, @timerid);
  fits.it_value.tv_sec := 1;
  fits.it_value.tv_nsec := 0;
  fits.it_interval.tv_sec := 1;
  fits.it_interval.tv_nsec := 0;
  timer_settime(timerid, 0, @fits, Nil);
  ReadLn(); // <- raise External:SIGUSR1. With debug fpc rtl, the line stops
at "call  psysinfo" inside FpSysCall

I assume threads are used for these timers.
If so, then this is normal, signal USR1 is used
by the linux threading mechanism, if I remember correctly.

Michael.

I also thought that the signal could be occupied by some api, but changing it to SIGRTMIN + 1, +2,+3 just changes the exception to "External:SIG35", "External:SIG36"...
Maybe I should apply the blocking mask for other signal for this api since it probably uses threads?

Max

Hmm, it seems that it's not a 'wrong' exception, I executed the same program from console and it worked. So there's a range of signals that is 'wrong' for the debugger/ide and it stops catching them. Can I somehow tell it to ignore some, for example 'SIG35'? I tried to add it to Language exceptions ignore list as 'External: SIG35', 'External:SIG35' (without space), SIG35. Nothing helped.

Max

It's not considered as "wrong" by gdb. By default it simply catches all signals. If you type "cont" at that point, the program will continue execution and the signal will be delivered to the program.

handle SIG35 nostop noprint pass

Jonas

Jonas, where am I supposed to add this options?

I tried adding
--eval-command="handle SIGUSR1 nostop noprint pass" to
Debugger_Startup_Options

but that didn't help. Actually nothing happened until I restarted the IDE. After the restart the first run showed
---------------
Initialization output:
&"Undefined command:\"\".Try \"help\".\n"
----------

To ensure the string is correct I tried
gdb --eval-command="handle SIGUSR1 nostop noprint pass" from terminal and it worked.
Is it possible that some procedure makes c-style string translation (" -> \") while actually it isn't necessary here?

Max

That's what you can enter on the gdb console after you start it from the command line. I've never debugged using Lazarus, so I can't help with that.

I tried adding --eval-command="handle SIGUSR1 nostop noprint pass"

While this won't solve your problem, as Michael said you should not use SIGUSR1 because it's probably already used by the system.

Jonas

SIG35 behaves the same. Strangely, I looked in system monitor for gdb command-line, it looked ok

/usr/bin/gdb -silent -i mi -nx --eval-command="handle SIG35 nostop noprint pass"

but after the same Initialization output dialog it still stops with "External:SIG35"
Actually looking at the -i mi options, I see it's machine interface with all these escaping sequences, so probably lazarus didn't undestand something, but it's strange it still stops

Thanks anyway

Max

Most commands are different in the machine interface. "handle SIG35 nostop noprint pass" is probably only valid when using the default interface. I don't know what the MI equivalent is.

Jonas

If I invoke the same full line in the terminal the output

$ /usr/bin/gdb -silent -i mi -nx --eval-command="handle SIG35 nostop noprint
pass"

=thread-group-added,id="i1"
~"Signal Stop\tPrint\tPass to program\tDescription\n" ~"SIG35 No\tNo\tYes\t\tReal-time event 35\n"

So looks like (Stop = No) gdb got the command.
Jonas, I see that you probably have your reasons not to use gdb inside the lazarus, but it was quite decent for me until this strange unavoidable moment :) It still is, hope I will find a work around

Max

Max,

This works (for me)

1) Add a file like :
/Users/helly/fos_buildsys/gdbinit

2) Content of file is:
handle SIGUSR1 pass noprint nostop

3) Add in Lazarus at

Debuger_Startup_Options : --command /Users/helly/fos_buildsys/gdbinit

4) Depending on lazarus version -> Reset Debugger via Menu / or restart

---

I use this to ignore signals, and it works here. (afair under linux/freebsd the same way)
I debug a multithreaded server hith kevent/kqueue waiting on the signals. (hup,kill,usr1,int) in a commandline daemon. I am on OSX Lion, fpc 2.5.1 trunk (2 weeks ago) lazarus trunk (2-3 weeks) ago.

helmut

Helmut, worked like magic! Thanks :)
In case Graeme see this post (he hates over-quoting desperately), I intentionally did not cut the quote to make this useful information exist twice everywhere ;)

Max