Обсуждение: [PATCH] We install pg_regress and isolationtester but not pg_isolation_regress

On 2018-Jun-01, Craig Ringer wrote:

> On 28 May 2018 at 15:06, Craig Ringer <craig@2ndquadrant.com> wrote:
> 
> > Per topic, the Pg makefiles install pg_regress (for use by extensions) and
> > htey install the isolationtester, but they don't install
> > pg_isolation_regress.
> >
> > We should install it too.

I happened to come across this thread by accident, and I tend to agree
that we need to install both isolationtester and pg_isolation_regress to
the pgxs dirs, just like we do pg_regress.  I can't find that
isolationtester is installed anywhere though; maybe that changed after
you posted this.  Anyway, this version of the patch installs both.

I did search for evidence in the Makefile's git log that would remove a
recipe for installing isolationtester; couldn't find anything.

> I'm wondering if I should add ISOLATION support to PGXS too, like we have
> REGRESS .

This was covered by Michael's d3c09b9b1307 a few months after you
posted.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

On 2020-Sep-30, Craig Ringer wrote:

> On Tue, 29 Sep 2020 at 22:09, Alvaro Herrera <alvherre@2ndquadrant.com>
> wrote:

> > I happened to come across this thread by accident, and I tend to agree
> > that we need to install both isolationtester and pg_isolation_regress to
> > the pgxs dirs, just like we do pg_regress.  I can't find that
> > isolationtester is installed anywhere though; maybe that changed after
> > you posted this.  Anyway, this version of the patch installs both.

> I think rules were added to allow the isolation tester to be installed with
> an explicit make -C src/test/isolation install, but not added to the
> default "install" target.
> 
> I agree it should just be installed.

Pushed, thanks.

On 2020-Oct-15, Alvaro Herrera wrote:

> Pushed, thanks.

I forgot to mention that I considered backpatching this and decided not
to, but only because it might confuse packagers if they see unrecognized
files in the installation.  I realize now that c3a0818460a8 was
back-patched.  Any opinions on backpatching?  

Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> I forgot to mention that I considered backpatching this and decided not
> to, but only because it might confuse packagers if they see unrecognized
> files in the installation.  I realize now that c3a0818460a8 was
> back-patched.  Any opinions on backpatching?  

We've added new installed files in minor releases before, true.
But I agree it's something to do only when pretty important, and I'm
not sure this clears the bar.  TAP tests (the facility added by that
other patch) seem way more commonly useful than isolation tests.

Quickly counting the uses in our existing in-core extensions, I see

TAP_TESTS: 3 contrib, 5 src/test/modules
ISOLATION: 1 contrib, 3 src/test/modules

Other than src/test/modules/brin, the ISOLATION users don't look
much like real extensions (rather than test scaffolding), either.
If you discount test scaffolding modules then the use-counts are
more like 4 to 1.

So I'm -0.1 or so on backpatching.

            regards, tom lane

On Thu, Oct 15, 2020 at 01:06:54PM -0400, Tom Lane wrote:
> Other than src/test/modules/brin, the ISOLATION users don't look
> much like real extensions (rather than test scaffolding), either.
> If you discount test scaffolding modules then the use-counts are
> more like 4 to 1.

Out of core, the only thing I can see with isolation tests is rum, but
it uses a workaround to have an access to the necessary binaries.
--
Michael

Hi hackers,

> Any opinions on backpatching?

> Other than src/test/modules/brin, the ISOLATION users don't look
> much like real extensions (rather than test scaffolding), either.

> Out of core, the only thing I can see with isolation tests is rum, but
> it uses a workaround to have an access to the necessary binaries.

I just wanted to let you know that TimescaleDB uses
pg_isolation_regress and occasionally there are reports from some
suffering/puzzled users/developers, e.g. [1]. Not 100% sure if it
makes investing the time into backpatching worth it. However if
someone could do it, it would be nice.

[1]: https://github.com/timescale/timescaledb/issues/1655

-- 
Best regards,
Aleksander Alekseev

On Wed, Apr 28, 2021 at 12:44:45PM +0300, Aleksander Alekseev wrote:
> I just wanted to let you know that TimescaleDB uses
> pg_isolation_regress and occasionally there are reports from some
> suffering/puzzled users/developers, e.g. [1]. Not 100% sure if it
> makes investing the time into backpatching worth it. However if
> someone could do it, it would be nice.

FWIW, I am not really sure that this is important enough to justify a
back-patch, even it is true that there have been cases in the past
where extra binaries got added in minor releases.
--
Michael

Michael Paquier <michael@paquier.xyz> writes:
> On Wed, Apr 28, 2021 at 12:44:45PM +0300, Aleksander Alekseev wrote:
>> I just wanted to let you know that TimescaleDB uses
>> pg_isolation_regress and occasionally there are reports from some
>> suffering/puzzled users/developers, e.g. [1]. Not 100% sure if it
>> makes investing the time into backpatching worth it. However if
>> someone could do it, it would be nice.

> FWIW, I am not really sure that this is important enough to justify a
> back-patch, even it is true that there have been cases in the past
> where extra binaries got added in minor releases.

Yeah, I think adding a binary in a minor release is a Big Deal to
packagers.  I doubt that the case here justifies that.

            regards, tom lane

On Thu, Apr 29, 2021 at 4:41 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Michael Paquier <michael@paquier.xyz> writes:
> > On Wed, Apr 28, 2021 at 12:44:45PM +0300, Aleksander Alekseev wrote:
> >> I just wanted to let you know that TimescaleDB uses
> >> pg_isolation_regress and occasionally there are reports from some
> >> suffering/puzzled users/developers, e.g. [1]. Not 100% sure if it
> >> makes investing the time into backpatching worth it. However if
> >> someone could do it, it would be nice.
>
> > FWIW, I am not really sure that this is important enough to justify a
> > back-patch, even it is true that there have been cases in the past
> > where extra binaries got added in minor releases.
>
> Yeah, I think adding a binary in a minor release is a Big Deal to
> packagers.  I doubt that the case here justifies that.

+1.

Given the number of complaints from people lacking it since the binary
was first created, I can't see how that's a priority that justifies
that.

-- 
 Magnus Hagander
 Me: https://www.hagander.net/
 Work: https://www.redpill-linpro.com/