Re: indxpath.c refactoring

This simple refactoring patch moves all of the code associated with
Partial Index planning into a single file. This isolates it from other
code changes that may take place over the next few months, as well as
setting the scene for a number of changes that will hopefully take place
with that code.

This should allow work to continue on this undisturbed; more work will
happen within the 8.1 window, so please approve this patch to cvstip
now.

Changes likely on this code are
- redesigning the low level routines to allow them to perform their
stuff for both Partial Index and Constraint-based elimination (aka
Partitioning...)
- allowing equivalence class re-write
- allow handling for r-tree operators
- maybe GIST also??

Better to do this change now, than try to do it when code gets more
complex.

- Passes make check on current cvstip.

- No functional code changes of any kind, just moving between modules.

Changes:
- code removal from indxpath.c
- code addition to new module predtest.c
- no changes required to header files
- update Makefile

Best Regards, Simon Riggs

Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> This simple refactoring patch moves all of the code associated with
> Partial Index planning into a single file.

Applied with some revisions -

* I put the new predtest.c file into optimizer/util, which seemed a more
plausible location than /path, and gave it its own header file.

* I thought check_partial_indexes() should stay in indxpath.c; it is an
application of the theorem-proving module to indexes, not part of the
prover proper.

* I took the opportunity to rename pred_test() to predicate_implied_by(),
which seems more clear about what it's doing and in which direction.

As for the future...

> - redesigning the low level routines to allow them to perform their
> stuff for both Partial Index and Constraint-based elimination (aka
> Partitioning...)

The other main entry point should probably be called something like
predicate_refuted_by?

> - allow handling for r-tree operators

I don't see anyone very excited about r-tree these days; have you
noticed that no one has stepped up to repair the known semantic errors?
I wouldn't spend any time on it in the prover.

> - maybe GIST also??

The trouble with GiST is that the logical structure of each opclass
could be completely different from the next --- it's certainly not
possible to have a simple fixed table of implications. Again, we have
much more important fish to fry, anyway ...

regards, tom lane

> I don't see anyone very excited about r-tree these days; have you
> noticed that no one has stepped up to repair the known semantic errors?
> I wouldn't spend any time on it in the prover.

That sort of thing is always self-fulfilling. If rtrees were fast,
logged and rad, then more people would use them :)

Chris

Christopher Kings-Lynne <chriskl(at)familyhealth(dot)com(dot)au> writes:
>> I don't see anyone very excited about r-tree these days; have you
>> noticed that no one has stepped up to repair the known semantic errors?
>> I wouldn't spend any time on it in the prover.

> That sort of thing is always self-fulfilling. If rtrees were fast,
> logged and rad, then more people would use them :)

The prophecy I'd like to see become self-fulfilling is that we get GIST
to production quality and then r-trees wither on the vine because
there's no reason to use them. I'm not aware of any reason to prefer
r-tree to an equivalent GIST opclass...

regards, tom lane

On Fri, 2005-06-10 at 18:33 -0400, Tom Lane wrote:
> Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> > This simple refactoring patch moves all of the code associated with
> > Partial Index planning into a single file.
>
> Applied with some revisions -
>
> * I put the new predtest.c file into optimizer/util, which seemed a more
> plausible location than /path, and gave it its own header file.
>
> * I thought check_partial_indexes() should stay in indxpath.c; it is an
> application of the theorem-proving module to indexes, not part of the
> prover proper.
>
> * I took the opportunity to rename pred_test() to predicate_implied_by(),
> which seems more clear about what it's doing and in which direction.
>

Many thanks.

> As for the future...
>
> > - redesigning the low level routines to allow them to perform their
> > stuff for both Partial Index and Constraint-based elimination (aka
> > Partitioning...)
>
> The other main entry point should probably be called something like
> predicate_refuted_by?

Will do.

Best Regards, Simon Riggs