Re: [PATCHES] WITH RECUSIVE patches 0723

Tatsuo Ishii <ishii(at)postgresql(dot)org> writes:
> Reviewers, please let me know if you find problems with the
> patches. If none, I would like to commit this weekend.

Given that everyone who has tested this has found a different way to
crash it, and that the frequency of crash reports shows no signs of
slowing down, I have to think that committing it is premature.

I tried to look through the patch just now and failed to make any
sense of it, because of the complete absence of documentation.
Two unexplained examples added to the SELECT reference page don't
do it for me. I want to see an explanation of exactly what behaviors
are intended to be provided (and, in view of the long TODO list that
was posted awhile back, what isn't provided). And there needs to be
more than zero internal documentation. A README file, or perhaps
a very long file header comment, needs to be provided to explain
what's supposed to happen, when, and where when processing a recursive
query. (For comparison look at the README.HOT file that was created
to explain the HOT patch --- something at about that level of detail
would help this patch a lot. Or consider adding a section to
chapter 43 in the SGML docs.)

We really can't accept a patch that is so poorly documented as to
be unreviewable. Unreviewable also means it'll be unmaintainable
going forward.

regards, tom lane