Thread: Possible bug in PostgreSQL 9.2 stable: TwoPhaseGetDummyBackendId()

I have looked at the Postgres 9.2 stable and Postgres 9.2 beta 3 git  
archives and this bug still appears to be present.

TwoPhaseGetDummyProc returns a PGPROC*. In 9.0, it was safe for  
TwoPhaseGetDummyBackendId() to cast this to a GlobalTransaction  
because the GlobalTransactionData structure's first element was always  
a PGPROC structure. However, in 9.2 this is no longer true. Despite  
the mismatch that now exists between the structures,  
TwoPhaseGetDummyBackendId() still attempts to cast a PGPROC* to a  
GlobalTransaction to extract the member 'dummyBackendId'. Rewriting  
this function to use the GlobalTransaction structure that is found in  
TwoPhaseState->prepXacts[] appears to fix the problem. In practice  
this appears to result in TwoPhaseGetDummyBackendId() returning  
invalid values (such as 0) instead of the expected range above  
MaxBackends. I'm not sure why this hasn't caused problems with  
multixact.c's use of this function.

Is this a known bug? Would it be helpful to submit a patch?

Regards,

Robert Ross

Robert Ross <rross@translattice.com> writes:
> I have looked at the Postgres 9.2 stable and Postgres 9.2 beta 3 git  
> archives and this bug still appears to be present.

> TwoPhaseGetDummyProc returns a PGPROC*. In 9.0, it was safe for  
> TwoPhaseGetDummyBackendId() to cast this to a GlobalTransaction  
> because the GlobalTransactionData structure's first element was always  
> a PGPROC structure. However, in 9.2 this is no longer true.

Clearly an oversight in commit ed0b409d22346b1b027a4c2099ca66984d94b6dd
:-(.  Will fix, thanks for the report!
        regards, tom lane