On 12/5/25 15:50, Achilleas Mantzios wrote:
Dear All, Dear Tom
On 5/11/25 16:20, Tom Lane wrote:
Achilleas Mantzios <a.mantzios@cloud.gatewaynet.com> writes:
We use are own version of DBmirror, we run our replication in a highly
fine grained manner. So every upgrade I have to make the code compile
and test. Up to PostgreSQL 17, I only got minor compilation problems
that I managed to resolve fairly easily. However this didn't prove to be
the case with PostgreSQL 18beta1, it proved harder to compile and as my
fears were verified, it has serious problems.
My question : is 18's SPI stabilized ? Can I start work on our version
of DBmirror ? Or wait for 18beta2 or -RC ?
If you think there are changes we need to make, you'd better get
specific sooner not later. I'm not aware of any large fixes that
are pending, cf
https://wiki.postgresql.org/wiki/PostgreSQL_18_Open_Items
I attach
a) our old source (pending.c.orig), as of PostgreSQL 17 (tested for some 7 months, so pretty well tested),
b) the compilation errors when compiled against 18beta1, and
c) the patch that I came up with, which seems (in my minimal testing) to yield correct results on 18beta1.
The majority of serious warnings have to do with de-toasting arrays and the PK's int2vector , while the error has to do with getting column details such as attisdropped and attname.
Please have a look, and share your thoughts. I haven't touched serious C coding till I first wrote the above sometime in 2004 with a bunch of additions some years ago.
Hi again
just to close this, it seems that the main issue for the compilation fail was a change introduced in this commit : d28dff3f6cd6a7562fb2c211ac0fb74a33ffd032 and had to do with access to TupleDesc->attrs which does not exist anymore. Some browsing in the contrib dir provided the hint I needed (to use : TupleDescAttr ).
The funny thing here is that deepseek spotted the issue before I even pasted any piece of code, just mentioning. It would be nice if those SPI level changes were mentioned somewhere.
regards, tom lane
