On 26 November 2017 at 22:56, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp> writes:
>> [ fix-rewrite-tlist-v4.patch ]
>
> I started reviewing this patch. I did not much like the fact that it
> effectively moved rewriteTargetListUD to a different file and renamed it.
> That seems like unnecessary code churn, plus it breaks the analogy with
> rewriteTargetListIU, plus it will make back-patching harder (since that
> code isn't exactly the same in back branches). I see little reason why
> we can't leave it where it is and just make it non-static. It's not like
> there's no other parts of the rewriter that the planner calls.
>
I wonder if, years from now, it might look a bit odd that
rewriteTargetListUD() is doing part of work of preptlist.c, is only
called from there, and yet is located in the rewriter.
Aside from having a similar name to rewriteTargetListIU(), what
rewriteTargetListUD() does seems more like what
preprocess_targetlist() does for rowmarks. The fact that
rewriteTargetListIU() intentionally only applies to the parent,
whereas preprocess_targetlist() and now rewriteTargetListUD() apply to
each child, further destroys any similarity between
rewriteTargetListUD() and rewriteTargetListIU().
The point about back-patching is a reasonable one though, so I won't
mind either way.
A separate point -- it might be marginally more efficient to have the
work of rewriteTargetListUD() done after expand_targetlist() to avoid
the possible renumbering of the resjunk entries.
Regards,
Dean