Thread: implementing another hash join...

Hi,
Im new to postgres and I'm about to implement the algorithm "double 
pipelined hash join". I've started using SPI functions but soon I 
realised that  I 'm  not at the right way (am I?). Is there a way that 
I'm not obliged to change code of  the optimizer of postgres but to 
write a seperate code as a new function?I've been told by my supervisor 
that I should use the function that is doing the hashing of a relation.  
There is a chaos !!!!Please help!!I'm reading the documentation of 
postgres and some '.c .h'  files (mainly the createplan.c  at the 
directory "~/src/backend/optimizer/plan/") to figer out  what to do and 
from where to start.
Regards!!!!
Martha

Martha Chronopoulou <mchron@aueb.gr> writes:
> Im new to postgres and I'm about to implement the algorithm "double 
> pipelined hash join". I've started using SPI functions but soon I 
> realised that  I 'm  not at the right way (am I?).

Not even close.  You need to be implementing new executor node types
(see backend/executor) and teaching the planner about the properties
of this join method.  SPI is at a completely different level of
abstraction ...

> Is there a way that I'm not obliged to change code of the optimizer of
> postgres

Nope.  The set of join types it knows about is pretty much hard-wired.
        regards, tom lane