This patch provides a preliminary implementation of the logical column
ordering functionality recently proposed on -hackers. This patch has a
number of known issues, so please don't apply it.
Current functionality implemented:
- Added attpos to the system catalogs & bumped the catalog
version: initdb required
- COPY TO/FROM obey logical column ordering
- When expanding "SELECT * ...", we make sure to obey logical
column order (see below, however)
- If INSERT is given an empty column list, it will produce the
necessary implicit column list in sorted by attpos
- (unrelated) improve a bunch of comments in pg_class.h,
remove some dead code from pg_attribute.h, refactor some
code in access/common/printtup.c
Remaining work to do:
- Change the name of the pg_attribute attribute to something
other than 'attpos', per the discussion on -hackers. I don't
believe we've settled on the right name yet.
- The code for actually sorting the columns in attpos-order is
duplicated a few times -- this was just done for the sake of
convenience, I'm going to clean this up and stick it in a
single, shared location in the new patch.
- The INSERT change causes the alter_table regression tests to
fail (the regression.diff) is attached. This appears to be
the logical column ordering interacting with inheritance,
but I haven't yet taken an in-depth look at it.
- When processing a "SELECT *", for example, the actual data
columns are returned in the right order, but the
RowDescription messages sent by libpq are not (i.e. they are
sent in attnum-order, not attpos).
In SendRowDescriptionMessage() and related printtup code, I
took a look at trying to sort the attributes we're about to
send RowDescription messages for by their 'attpos', but the
TupleDesc that we have for the return type doesn't included
a filled-in attpos.
I spent a while trying to see how feasible it would be to
fill in the result-type-TupleDesc with an attpos where
available, but couldn't figure out an easy way to do
this. Are there any suggestions on how this should be best
done?
- I haven't yet implemented ordering-by-attpos for the output
columns of a join, or the alias->column matching in the FROM
alias list (per Tom's email to -hackers). I saw a couple
places where I might be able to do this, but I'm quite sure
what the correct spot is. Tom, do you have any suggestions?
Any comments would be gratefully appreciated.
-Neil