I wrote:
> I've attached version 9, whose biggest change is to address the above
> points of review. I pushed all of the catalog header build logic into
> catalog Makefile to avoid creating a separate symbol file. This
> involved putting the distprep logic there as well. Enough of the
> structure changed that one or two names didn't make sense anymore, so
> I changed them.
>
> As suggested, the conversion script is now part of the patchset and
> not committed to the repo. To run the conversion, save everything to a
> directory and update the dir vars at the top of
> apply-bootstrap-data-patches.sh accordingly.
>
> A couple things to note that I didn't do:
> -With all the new generated headers, the message "Writing ..." is now
> quite verbose. It might be worth changing that.
> -I'm not sure if I need to change anything involving "make install".
> -I haven't tested the MSVC changes.
> -I didn't change any clients to actually use the new headers directly.
> That might be too ambitious for this cycle anyway.
>
> While this goes through review, I'll get a head start rebasing the
> human readable OIDs and data compaction patches.
It didn't take that long to rebase the remaining parts of the
patchset, so despite what I said above I went ahead and put them in
version 10 (attached), this time via scripted bulk editing rather than
as large patches. Changes since the last patchset that contained these
parts:
-Split out the generation of pg_type OID symbols into its own patch.
-Remove single-purpose magic behaviors.
-Ditto for the ability to abbreviate attribute names. I decided the
added complexity and possible confusion wasn't worth the space
savings.
-Add some more OID macros for pg_aggregate and pg_range that I missed before.
Also, more generally, I cleaned up the apply-patches script and edited
its comments and commit messages.
Tom Lane wrote:
> In 0010, you relabel the types of some OID columns so that genbki.pl
> will know which lookup to apply to them. That's not such a problem for
> the relabelings that are just macros and genbki.pl converts back to
> type OID in the .bki file. But you also did things like s/Oid/regtype/,
> and that IS a problem because it will affect what client code sees in
> those catalog columns. We've discussed changing those columns to
> regfoo types in the past, and decided not to, because of the likelihood
> of breaking client queries. I do not think this patch gets to change
> that policy. So the way to identify the lookup rule needs to be
> independent of whether the column is declared as Oid or an Oid alias type.
> Perhaps an explicit marker telling what transformation to make, like
>
> Oid rngsubtype BKI_LOOKUP(pg_type);
>
> would work for that.
This is also done (now in 0007).
-John Naylor
