Re: How to interrupt a MT calculation?

[Bill Allombert <allomber@math.u-bordeaux.fr>]

On Sat, Jul 01, 2023 at 02:50:15PM +0200, Bill Allombert wrote:
> On Sat, Jul 01, 2023 at 01:30:35PM +0200, Loïc Grenié wrote:
> > On Sat 1 Jul 2023 aT 12:39, Bill Allombert wrote:
> > 
> > > On Sat, Jul 01, 2023 at 01:04:49AM +0200, Jean-Luc Arnaud wrote:
> > > > So, it seems like error does not work the same way on Linux (and probably
> > > > Windows) versus Mac OS, or is there a bug in the 2.16.0 Alpha version?
> > >
> > > No this is a limitation of pthread on macos:
> > > <
> > > https://stackoverflow.com/questions/43109438/macos-macoss-pthread-pthread-cancel-asynchronous-not-working
> > > >
> > >
> > 
> >      Definitely not the recommended way, but could a SEGV help?
> > 
> > install("getpid","l")
> > extern(Str("kill -SEGV ", getpid()))
> 
> That would not help: PARI will catch the SEGV and then call pthread_cancel to cancel the threads.
> No, your best bet is to make new_chunk or set_avma a cancellation point, but this slows down GP.

Jean-Luc,
you can try the following patch, it should fix this, but it will slowdown GP a bit.
(you will need to do 'make clean')

Cheers,
Bill

diff --git a/src/headers/parisys.h b/src/headers/parisys.h
index ee80d5b7c..0aa92a584 100644
--- a/src/headers/parisys.h
+++ b/src/headers/parisys.h
@@ -61,5 +61,6 @@ BEGINEXTERN
 ENDEXTERN
 #define CHECK_CTRLC if (win32ctrlc) dowin32ctrlc();
 #else
-#define CHECK_CTRLC
+void  pthread_testcancel(void);
+#define CHECK_CTRLC pthread_testcancel();
 #endif