On Thu, Dec 17, 2009 at 7:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> If we're going to start moving these checks around we need a very
> well-defined notion of where permissions checks should be made, so that
> everyone knows what to expect. I have not seen any plan for that.
> Removing one check at a time because it appears to not be necessary
> in the code paths you've looked at is not a plan.
I'm not completely familiar with the existing code structure here, but
it sort of seems like in general you might want to divide up the
processing of a statement into a parse analysis phase, a permissions
checking phase, and an execution phase. The parse analysis seems to
be mostly separated out into transformXyz() functions, but the
permissions checking is mixed in with the execution. Disentangling
that seems like a job and a half.
...Robert