Horton

Thread

Commits

  1. Add a regression test script dedicated to exercising system views.

  1. Create a separate test file for exercising system views

    Tom Lane <tgl@sss.pgh.pa.us> — 2017-01-29T21:02:21Z

    In connection with the "pg_hba_file_settings view patch" thread, I was
wondering where we could logically insert a regression test case for that
view.  I realized that there is no natural home for it among the existing
regression tests, because it's not really connected to any SQL language
feature.  The same is true for a number of other built-in views, and
unsurprisingly, most of them are not exercised anywhere :-(.

Accordingly, I propose creating a new regression test file whose charter
is to exercise the SRFs underlying system views, as per attached.
I don't think we desperately need new tests for views that expand to
simple SQL, but these test cases correspond directly to code coverage
for C functions, so they seem worthwhile.

I did not do anything about testing the various pg_stat_xxx views.
Those could be added later, or maybe they deserve their own home.
(In many cases, those would need something smarter than the basic
count(*) technique used here, because the C functions are invoked
in the view's SELECT list not in FROM, so the planner would throw
away those calls.)

Comments/objections?

			regards, tom lane

  2. Re: Create a separate test file for exercising system views

    Andres Freund <andres@anarazel.de> — 2017-01-29T21:12:19Z

    Hi,

On 2017-01-29 16:02:21 -0500, Tom Lane wrote:
> I did not do anything about testing the various pg_stat_xxx views.
> Those could be added later, or maybe they deserve their own home.
> (In many cases, those would need something smarter than the basic
> count(*) technique used here, because the C functions are invoked
> in the view's SELECT list not in FROM, so the planner would throw
> away those calls.)

I've previously wished there were a portable equivalent of \o /dev/null
that'd not be perfect, but it'd still exercise more than what we
currently have.  Alternatively casting the entire row to text should
allow to use count(*) trickery in some of the cases at least.

> Comments/objections?

Sounds like a good idea here.

Greetings,

Andres Freund

  3. Re: Create a separate test file for exercising system views

    Michael Paquier <michael.paquier@gmail.com> — 2017-01-30T05:02:12Z

    On Mon, Jan 30, 2017 at 6:02 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> In connection with the "pg_hba_file_settings view patch" thread, I was
> wondering where we could logically insert a regression test case for that
> view.  I realized that there is no natural home for it among the existing
> regression tests, because it's not really connected to any SQL language
> feature.  The same is true for a number of other built-in views, and
> unsurprisingly, most of them are not exercised anywhere :-(.
>
> Accordingly, I propose creating a new regression test file whose charter
> is to exercise the SRFs underlying system views, as per attached.
> I don't think we desperately need new tests for views that expand to
> simple SQL, but these test cases correspond directly to code coverage
> for C functions, so they seem worthwhile.
>
> I did not do anything about testing the various pg_stat_xxx views.
> Those could be added later, or maybe they deserve their own home.
> (In many cases, those would need something smarter than the basic
> count(*) technique used here, because the C functions are invoked
> in the view's SELECT list not in FROM, so the planner would throw
> away those calls.)
>
> Comments/objections?

Nice idea to group those tests in the same file. I am not noticing any
issues with the patch proposed.
-- 
Michael

  4. Re: Create a separate test file for exercising system views

    Ashutosh Bapat <ashutosh.bapat@enterprisedb.com> — 2017-01-30T06:42:27Z

    On Mon, Jan 30, 2017 at 2:32 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> In connection with the "pg_hba_file_settings view patch" thread, I was
> wondering where we could logically insert a regression test case for that
> view.  I realized that there is no natural home for it among the existing
> regression tests, because it's not really connected to any SQL language
> feature.  The same is true for a number of other built-in views, and
> unsurprisingly, most of them are not exercised anywhere :-(.
>
> Accordingly, I propose creating a new regression test file whose charter
> is to exercise the SRFs underlying system views, as per attached.
> I don't think we desperately need new tests for views that expand to
> simple SQL, but these test cases correspond directly to code coverage
> for C functions, so they seem worthwhile.
>
> I did not do anything about testing the various pg_stat_xxx views.
> Those could be added later, or maybe they deserve their own home.
> (In many cases, those would need something smarter than the basic
> count(*) technique used here, because the C functions are invoked
> in the view's SELECT list not in FROM, so the planner would throw
> away those calls.)
>
> Comments/objections?
>

I think this is good in the given infrastructure. Thanks for working on it.
-- 
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company