[postgis-devel] Bogus PROJ4 memory context definition crashes Postgres

[Tom Lane]

Mark Cave-Ayland <mark.cave-ayland at ilande.co.uk> writes:
> On Wed, 2007-07-11 at 15:03 -0400, Tom Lane wrote:
>> If an "out of memory" condition occurs while using transform(),
>> Postgres will crash every time, because the memory manager code
>> will try to dump memory context stats to the postmaster log ...
>> and the Postgis code doesn't provide a stats method for its custom
>> memory context type.  Providing this method type is not optional.

> Oops. That code is my fault - I'll see if I can come up with a patch to
> fix this. Is this easy to simulate just with a large palloc()?

All you need is a stats-printing routine.  The minimum version would
look something like

static void
PROJ4SRSCacheStats(MemoryContext context)
{
    fprintf(stderr,
            "%s: PROJ4 context\n",
            context->name);
}

though perhaps there are useful bits of info you could add to the
printout --- maybe the srid would be worth including for instance.

>> I do not think the reset and is_empty methods are optional either,
>> although I don't have such a simple example of where they might
>> get called.

> Will look at these too. The only reason the MemoryContext exists in the
> first place is because we need to be able to cache library handles for
> PROJ.4, and so we need a hook when the backend exits in order to free
> all the handles we have used. Do you know of a better way of doing this,
> preferably with compatibility down to PostgreSQL 7.2?

Actually I thought it was a pretty cute solution.  You just need to
support all the methods that anyone might call.  I don't think they
need to do much --- it's probably all right if reset() does nothing,
and is_empty() can just always return false.

			regards, tom lane