On Wed, May 16, 2012 at 02:46:17PM +0300, Heikki Linnakangas wrote:
> On 16.05.2012 14:30, Mark Cave-Ayland wrote:
> >On 16/05/12 11:39, Heikki Linnakangas wrote:
> >
> >>However, if you're absolutely positively sure that the library function
> >>can tolerate that, you can set "ImmediateInterruptOK = true" before
> >>calling it. See e.g PGSemaphoreLock() on how that's done before starting
> >>to sleep on a semapgore.
> >
> >Hi Heikki,
> >
> >Yes that appears to work fine on the surface - just testing now to
> >determine what state everything is left in when queries are aborted
> >prematurely.
>
> You're unlikely to catch all problems just by testing. I wouldn't
> trust that it's safe unless the library authors specifically
> mentions that it is safe to longjump out of the function at any
> time. Note for example that if the library function calls back to
> palloc/pfree, then it's not safe, because interrupting those
> functions is not safe.
I'm right now getting the external library into using memory allocator
wrappers so to provide an syntetized OOM condition in an aim to have a
more predictable answer to that.
But CHECK_FOR_INTERRUPTS doesn't return, right ?
Is there another macro for just checking w/out yet acting upon it ?
--strk;
,------o-. | __/ | Delivering high quality PostGIS 2.0 ! | / 2.0 | http://strk.keybit.net -
http://vizzuality.com`-o------'