Re: [GENERAL] dblink: rollback transaction

Joe Conway wrote:
> Tom Lane wrote:
>> Joe Conway <mail(at)joeconway(dot)com> writes:
>>> I like the idea in general, but maybe instead there should be a new
>>> overloaded version of the existing function names that accepts an
>>> additional bool argument. Without the argument, behavior would be as
>>> it is now; with it, you could specify the old or new behavior.
>>
>> Um, maybe I'm confused about the context, but aren't we talking about C
>> function names here? No overloading is possible in C ...
>
> I was thinking in terms of overloaded SQL function names. For example,
> in addition to dblink_exec(text) and dblink_exec(text,text) we create
> dblink_exec(text,bool) and dblink_exec(text,text,bool).
>
> Currently both SQL versions of dblink_exec are implemented by a single C
> level function. But yes, we'd need another C level function to support
> the new SQL functions because there would be no way to distinguish the 2
> two-argument versions otherwise. (Actually, now I'm wondering if we
> could use a single C function for all four SQL versions -- between
> PG_NARGS() and get_fn_expr_argtype() we should be able to figure out how
> we were called, shouldn't we?)

The attached implements the new overloaded SQL functions as discussed
above (i.e. start with existing argument combinations, add a new bool
argument to each). I ended up with a single C function (by making use of
number and type of the arguments) for each overloaded SQL function name.

I'll commit in a day or two if there are no objections.

Thanks,

Joe