Tom Lane wrote:
>Andreas Pflug <pgadmin@pse-consulting.de> writes:
>
>
>>Maybe my proposal wasn't clear enough:
>>Just as an index references a pg_class entry by it's OID, not some value
>>identifying it's physical storage, all objects might continue
>>referencing columns by attnum.
>>
>>
>
>That's exactly the same thing I am saying. Your mistake is to assume
>that this function can be combined with identification of a (changeable)
>logical column position. It can't. Changeability and immutability are
>just not compatible requirements.
>
>
In the mind of a programmer, a ALTER COLUMN doesn't create a new column,
but merely changes some attributes of an existing column. In this sense,
changeability and immutability are not controversal.
Digging deeper:
TupDesc contains an array of physical attr descriptions, and to access a
column description attnum is taken as index into that array (taken from
fastgetattr).
return fetchatt(tupleDesc->attrs[attnum-1], ...)
The physical location can easily reordered if there's an additional
array, to translate attnum into the array index.
return fetchatt(tupleDesc->attrs[tupleDesc->attrpos[attnum-1]] ...
For sure, reordering (i.e. changing the attrpos array) may only be
performed as long as the column isn't referenced.
Regards,
Andreas