Re: WIP: Faster Expression Processing v4

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Andres Freund <andres(at)anarazel(dot)de>
Cc: Heikki Linnakangas <hlinnaka(at)iki(dot)fi>, Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: WIP: Faster Expression Processing v4
Date: 2017-03-23 21:40:55
Message-ID: 8011.1490305255@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Stylistic thought ... I am wondering if it wouldn't be a good idea
to replace EEOP_CASE_WHEN_STEP, EEOP_CASE_THEN_STEP, EEOP_COALESCE,
and EEOP_ARRAYREF_CHECKINPUT with instructions defined in a less
usage-dependent way as

EEOP_JUMP unconditional jump
EEOP_JUMP_IF_NULL jump if step result is null
EEOP_JUMP_IF_NOT_NULL jump if step result isn't null
EEOP_JUMP_IF_NOT_TRUE jump if step result isn't TRUE

One could imagine later filling out this set with the other BoolTest
condition types, but that seems to be all we need right now.

These are basically just renamings of the step types that exist now,
although EEOP_ARRAYREF_CHECKINPUT would have to drop its not-very-
necessary Assert(!op->d.arrayref.state->isassignment). Well, I guess
I should say that they're renamings of the semantics that I have
for these steps in my working copy; for instance, I got rid of
casewhen.value/casewhen.isnull in favor of letting CASE WHEN expressions
evaluate into the CASE's final output variable.

At least to me, I think the compiling code would be more readable
this way. I find WHEN_STEP and THEN_STEP a bit odd because they are
emitted after, not before, the expressions you'd think they control.
ARRAYREF_CHECKINPUT is pretty vaguely named, too.

regards, tom lane

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Michael Paquier 2017-03-23 21:54:59 Re: Potential data loss of 2PC files
Previous Message Thomas Munro 2017-03-23 21:39:24 Re: Measuring replay lag