On 29/01/13 10:29, Mark Kirkwood wrote:
On 25/01/13 13:56, Mark Kirkwood wrote:
On 25/01/13 13:49, Tom Lane wrote:
Mark Kirkwood <mark.kirkw...@catalyst.net.nz> writes:
On 25/01/13 13:06, Tom Lane wrote:
Unless libR can be coerced into not screwing up our signal handlers,
I'd say that PL/R is broken beyond repair. That would be
unfortunate.
It looks like Joe has run into something similar with libR stealing
SIGINT, he reinstalls it. A simple patch along the same lines for
SIGUSR1 (attached) seems to fix the issue.
This certainly is not good enough, for either signal. What happens if
the signal arrives while libR still has control? These things being
asynchronous with respect to the receiving backend, we certainly can't
assume that that won't happen.
Why does libR think it should be messing with these handlers in the
first place?
Agreed - I will see if I can work out why.
Looking at the R source (which reminds me of Postgres, nice, clear
code...), I saw quite a bit of code involving signal handlers - and
nothing along the lines of "don't set them up if you are in embedded
mode". So clearly more investigation is needed, and we really need to
take this up on the R mailing list I think. Joe - is this something
you would like to do? I am happy do it, but my knowledge of R is
measured in hours...
Like a dog with a bone - sometimes it's hard to let go... given the
quite readable nature of the R source I managed to find what I *think*
is the "don't do R signals switch" (which I'd obviously missed before).
Making use of this seems to fix the original bug - and possibly the
SIGINT stealing too.
Patch attached to set the variable (R_SignalHandlers = 0), and remove
the SIGINT workaround.
Cheers
Mark
*** plr.c.orig 2011-08-30 10:42:56.000000000 +1200
--- plr.c 2013-01-29 12:56:53.995936063 +1300
***************
*** 31,37 ****
* plr.c - Language handler and support functions
*/
#include "plr.h"
- #include "libpq/pqsignal.h"
PG_MODULE_MAGIC;
--- 31,36 ----
***************
*** 369,374 ****
--- 368,378 ----
atexit(plr_atexit);
/*
+ * Try to stop R using its own signal handlers
+ */
+ R_SignalHandlers = 0;
+
+ /*
* When initialization fails, R currently exits. Check the return
* value anyway in case this ever gets fixed
*/
***************
*** 392,401 ****
R_Interactive = false;
#endif
- /*
- * R seems to try to steal SIGINT in recent releases, so steal it back
- */
- pqsignal(SIGINT, StatementCancelHandler); /* cancel current query */
plr_pm_init_done = true;
}
--- 396,401 ----
--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs
