Re: Suggestion: Issue warning when calling SET TRANSACTION outside transaction block

On Thu, Oct 3, 2013 at 8:32 PM, Bruce Momjian <bruce(at)momjian(dot)us> wrote:
> On Thu, Oct 3, 2013 at 11:50:09AM +0530, Amit Kapila wrote:
>> > I looked at this but could not see how to easily pass the value of
>> > 'isTopLevel' down to the SELECT. All the other checks have isTopLevel
>> > passed down from the utility case statement.
>>
>> Yes, we cannot pass isTopLevel, but as isTopLevel is used to decide
>> whether we are in function (user defined) call, so if we can find
>> during statement execution (current case set_config execution) that
>> current statement is inside user function execution, then it can be
>> handled.
>> For example, one of the ways could be to use a mechanism similar to
>> setting of user id and sec context used by fmgr_security_definer() (by
>> calling function SetUserIdAndSecContext()), once userid and sec
>> context are set by fmgr_security_definer(), later we can use
>> InSecurityRestrictedOperation() anywhere to give error.
>>
>> For current case, what we can do is after analyze
>> (pg_analyze_and_rewrite), check if its not a builtin function (as we
>> can have functionid after analyze, so it can be checked
>> fmgr_isbuiltin(functionId)) and set variable to indicate that we are
>> in function call.
>>
>> Any better or simpler idea can also be used to identify isTopLevel
>> during function execution.
>>
>> Doing it only for detection of transaction chain in set_config path
>> might seem to be more work, but I think it can be used at other places
>> for detection of transaction chain as well.
>
> I am also worried about over-engineering this.

I had tried to think hard but could not come up with a simpler
change which could have handled all cases.
We can leave the handling for set_config() and proceed with patch
as Andres already given a reason where set_config() can be useful
within a
statement as well.

> I will wait to see if
> anyone else would find top-level detection useful, and if not, I will
> just apply my version of that patch that does not handle set_config.

I had verified the patch once again and ran regression, everything looks fine.

With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com