On 5/15/20 5:41 PM, Tom Lane wrote:
"David G. Johnston" <david.g.johnston@xxxxxxxxx> writes:
Yes, SQL and pl/pgsql have very different behaviors when it comes to
compilation and execution. In particular SQL performs parsing earlier
(during creation - just like it does for views) and links the textual query
to its parse result earlier. For pl/pgsql none of that happens until the
function is called. Because of this pl/pgsql allows for ambiguous sql text
to exist and be concretely resolved during execution while SQL does not.
I don't think that's accurate. SQL functions are stored as plain text,
just like any other non-C-coded function, and they are not parsed until
execution.
There are big differences from plpgsql of course. For one, it's
possible for a SQL function to be "inlined" into the calling query,
in which case parsing happens during planning of the calling query.
But other than that, I'd expect the execution-time search path
to determine how a SQL function behaves.
Since Rob didn't provide any details, it's far from clear what's
going wrong for him.
regards, tom lane
Did my message with a sql and plgpsql versions not come through?
I cannot create a plain sql function unless the search_path covers any
table mentioned. Not the case when using plpgsql - no path needed.
I'm ok(ish) with that, unless I've missed some detail.
rjs
