2009/5/25 Pavel Stehule <pavel.stehule@gmail.com>:
> 2009/5/24 Tom Lane <tgl@sss.pgh.pa.us>:
>> Pavel Stehule <pavel.stehule@gmail.com> writes:
>>> ==Steps==
>>> 1. add hook to analyser (transform stage) to substitute unknown
>>> columnref by param - when analyser detect unknown columnref, then call
>>> callback, that returns possible para node or NULL (when external
>>> environment doesn't have a variable). Returned param should be typed
>>> or unknown (for polymorphic params).
>>
>> IMHO the hook definition should support both the case of external
>> variables taking precedence over query variables and vice versa.
>> I don't think the core parser should be forcing that decision. In any
>> case we'd probably need both options for plpgsql, so as to be able to
>> support both traditional and Oracle-compatible behavior.
>
> good idea
>
the problem is place for hook variable - it should be in ParseState
because we need call hooked functions from environments that install
hook.
Pavel
>>
>> I'd be inclined to do that by letting the hook function interpose
>> itself between the parser and the regular transformColumnRef processing,
>> so that it can call the regular transformColumnRef processing either
>> before or after doing its external lookups. Giving it control only
>> after the regular processing fails would mean there's no way to let
>> external variables take precedence.
>>
>
> ok
>
>>> 2. add special modes to sql parser:
>>
>> None of those seem like a good idea to me. The only part that seems
>> useful is warning about conflicts between external variables and query
>> variables. That can be implemented by the hook function itself, if we
>> define the hook behavior as above.
>>
>
> there is minimal one necessary - for polymorphic variables, we know
> name, but we don't know type. And without types, we can't to transform
> correctly functions.
>
> regards
> Pavel Stehule
>
>
>> regards, tom lane
>>
>
