Em ter., 29 de ago. de 2023 às 20:06, Michael Paquier
escreveu:
> On Thu, Aug 24, 2023 at 07:02:40PM -0700, Peter Geoghegan wrote:
> > FWIW I'm pretty sure that it's impossible to run into problems here in
> > practice -- the minheap is allocated by palloc(), and the high
> > watermark number of
On Thu, Aug 24, 2023 at 07:02:40PM -0700, Peter Geoghegan wrote:
> FWIW I'm pretty sure that it's impossible to run into problems here in
> practice -- the minheap is allocated by palloc(), and the high
> watermark number of free pages is pretty small. Even still, I agree
> with your conclusion. Th
On Thu, Aug 24, 2023 at 6:44 PM David Rowley wrote:
> I agree with this. I think Ranier's patch is good and we should apply
> it and backpatch it.
FWIW I'm pretty sure that it's impossible to run into problems here in
practice -- the minheap is allocated by palloc(), and the high
watermark number
On Thu, Aug 24, 2023 at 6:18 PM Michael Paquier wrote:
> Still that looks entirely different to me. Here we have a problem
> where the number of free blocks stored may cause an overflow in the
> internal routine retrieving a free block, but your other thread
> is about long being not enough on Wi
On Fri, 25 Aug 2023 at 13:19, Michael Paquier wrote:
> Still that looks entirely different to me. Here we have a problem
> where the number of free blocks stored may cause an overflow in the
> internal routine retrieving a free block, but your other thread
> is about long being not enough on Wind
On Thu, Aug 24, 2023 at 05:33:15PM -0700, Peter Geoghegan wrote:
> I am in favor of fixing the problem. I don't quite recall what it was
> that made the discussion stall last time around.
I think that you mean this one:
https://www.postgresql.org/message-id/CAH2-WznCscXnWmnj=STC0aSa7QG+BRedDnZsP=j
On Thu, Aug 24, 2023 at 4:47 PM Michael Paquier wrote:
> > patch attached.
>
> Yeah, it looks like you're right here. Will do something about that.
This is a known issue. It has been discussed before.
I am in favor of fixing the problem. I don't quite recall what it was
that made the discussion
On Thu, Aug 24, 2023 at 02:46:42PM -0300, Ranier Vilela wrote:
> With Windows both *long* and *int* are 4 bytes.
> But with Linux *long* is 8 bytes and *int* are 4 bytes.
And I recall that WIN32 is the only place where we treat long as 4
bytes.
> patch attached.
Yeah, it looks like you're right
Hi,
nFreeBlocks stores the number of free blocks and
your type is *long*.
At Function ltsGetFreeBlock is locally stored in
heapsize wich type is *int*
With Windows both *long* and *int* are 4 bytes.
But with Linux *long* is 8 bytes and *int* are 4 bytes.
patch attached.
best regards,
Ranier Vi
