Re: Test code is worth the space

On 2015-08-15 03:35, Jim Nasby wrote:
>
> I setup a simple example of this with 64 variations of TAP tests, BLKSZ
> and WAL blocksize. Unfortunately to make this work you have to commit a
> .travis.yml file to your fork.
>
> build: https://travis-ci.org/decibel/postgres/builds/75692344
> .travis.yml: https://github.com/decibel/postgres/blob/master/.travis.yml
>
> Looks like we might have some problems with BLKSZ != 8...

I went through several of those and ISTM that most test failures are to
be expected:
a) order of resulting rows is different for some of the joins
b) planner behaves differently because of different number of pages
c) tablesample returns different results because it uses ctids as random
source
d) toaster behaves differently because of different threshold which is
calculated from BLKSZ
e) max size of btree value is exceeded with BLKSZ = 4 in the test that
tries to create deep enough btree

We could fix a) by adding ORDER BY to those queries but I don't see how
to fix the rest easily or at all without sacrificing some test coverage.

However I see one real error in the BLKSZ = 4 tests - two of the merge
join queries in equivclass die with "ERROR: could not find commutator
for operator" ( example build output
https://travis-ci.org/decibel/postgres/jobs/75692377 ). I didn't yet
investigate what's causing this.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services