longjmp - was: seemingly random "nesting of readline input" warnings

Barry Rowlingson B.Rowlingson@lancaster.ac.uk
Mon, 04 Nov 2002 13:03:32 +0000

Luke Tierney wrote:

> Looks like another case where our expedient but unfortunately too
> crude approach of handling Control-C on UNIX-likes with longjmp is
> biting us.  We really do need to bite the bullet and fix this properly.
> Maybe Duncan and I can coordinate getting a handle on this.

  please do! It still affects Mosix systems, since the signal handling 
in longjmp causes a system call on the home machine. Mosix tries to 
migrate processes to less-loaded machines, but R's constant setjmp'ing 
means the processes get stuck on the machines where they are started from.

  When I install a new version of R on our mosix cluster, I also compile 
a binary with setjmp/longjmp defined as the versions that do not 
preserve signal state. This produces a binary that migrates better under 
Mosix but has broken Ctrl-C handling. I then alter the R script so that 
my modified binary runs if and only if you do "R BATCH". Seems a fair 
solution at the moment.

  Barry Rowlingson
  Maths and Stats
  Lancaster University

r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch