pgsql: Make PostgreSQL::Test::Cluster compatible with all live branches

Andrew Dunstan <andrew@dunslane.net> writes:

Make PostgreSQL::Test::Cluster compatible with all live branches

prairiedog is unhappy [1]https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=prairiedog&amp;dt=2022-03-30%2021%3A23%3A58, seemingly as a result of this patch:

t/001_repl_stats....Can't locate parent.pm in @INC (@INC contains: ../../src/test/perl/ . /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl . /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl .) at ../../src/test/perl//PostgreSQL/Test/Cluster.pm line 2921.
BEGIN failed--compilation aborted at ../../src/test/perl//PostgreSQL/Test/Cluster.pm line 2921.

I think this means that its old Perl version misinterprets

use parent -norequire, qw(PostgreSQL::Test::Cluster);

as a request to include "parent.pm". Is there a more
backwards-compatible way to spell that?

regards, tom lane

[1]: https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=prairiedog&amp;dt=2022-03-30%2021%3A23%3A58

On Mar 30, 2022, at 8:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

Make PostgreSQL::Test::Cluster compatible with all live branches

prairiedog is unhappy [1], seemingly as a result of this patch:

t/001_repl_stats....Can't locate parent.pm in @INC (@INC contains: ../../src/test/perl/ . /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl . /usr/local/perl5.8.3/lib/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/5.8.3 /usr/local/perl5.8.3/lib/site_perl/5.8.3/darwin-2level /usr/local/perl5.8.3/lib/site_perl/5.8.3 /usr/local/perl5.8.3/lib/site_perl .) at ../../src/test/perl//PostgreSQL/Test/Cluster.pm line 2921.
BEGIN failed--compilation aborted at ../../src/test/perl//PostgreSQL/Test/Cluster.pm line 2921.

I think this means that its old Perl version misinterprets

use parent -norequire, qw(PostgreSQL::Test::Cluster);

as a request to include "parent.pm". Is there a more
backwards-compatible way to spell that?

Not that I know of. Googling tells me it was a core module from 5.10.1. I can revert tomorrow if necessary :-(

Cheers

Andrew

Andrew Dunstan <andrew@dunslane.net> writes:

On Mar 30, 2022, at 8:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
I think this means that its old Perl version misinterprets
use parent -norequire, qw(PostgreSQL::Test::Cluster);
as a request to include "parent.pm". Is there a more
backwards-compatible way to spell that?

Not that I know of. Googling tells me it was a core module from 5.10.1. I can revert tomorrow if necessary :-(

Hmm. I checked not too long ago, and these are the buildfarm
animals that would be broken:

prairiedog | 2022-03-23 05:04:02 | configure: using perl 5.8.3
anole | 2022-03-24 10:40:08 | configure: using perl 5.8.8
gharial | 2022-03-24 18:32:22 | configure: using perl 5.8.8
locust | 2022-03-19 22:17:53 | configure: using perl 5.8.8
gaur | 2022-03-19 15:00:54 | configure: using perl 5.8.9

I could update prairiedog and gaur (... probably), but hard to say
about the others.

It seems kind of sad for a backwards-compatibility patch to force a
major move of our minimum Perl version. OTOH, if there's no other
good way to do it, maybe we gotta.

Another possibility is that maybe parent.pm could be successfully
installed into a 5.8.x installation. I can't speak for those
other animals, but mine at least are already frankencritters.

regards, tom lane

On Mar 30, 2022, at 9:12 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

On Mar 30, 2022, at 8:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

I think this means that its old Perl version misinterprets
use parent -norequire, qw(PostgreSQL::Test::Cluster);
as a request to include "parent.pm". Is there a more
backwards-compatible way to spell that?

Not that I know of. Googling tells me it was a core module from 5.10.1. I can revert tomorrow if necessary :-(

Hmm. I checked not too long ago, and these are the buildfarm
animals that would be broken:

prairiedog | 2022-03-23 05:04:02 | configure: using perl 5.8.3
anole | 2022-03-24 10:40:08 | configure: using perl 5.8.8
gharial | 2022-03-24 18:32:22 | configure: using perl 5.8.8
locust | 2022-03-19 22:17:53 | configure: using perl 5.8.8
gaur | 2022-03-19 15:00:54 | configure: using perl 5.8.9

I could update prairiedog and gaur (... probably), but hard to say
about the others.

It seems kind of sad for a backwards-compatibility patch to force a
major move of our minimum Perl version. OTOH, if there's no other
good way to do it, maybe we gotta.

Another possibility is that maybe parent.pm could be successfully
installed into a 5.8.x installation. I can't speak for those
other animals, but mine at least are already frankencritters.

I think it can be salvaged. See https://fastapi.metacpan.org/source/CORION/parent-0.238/lib/parent.pm

Cheers

Andrew

On 3/30/22 21:12, Tom Lane wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

On Mar 30, 2022, at 8:19 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
I think this means that its old Perl version misinterprets
use parent -norequire, qw(PostgreSQL::Test::Cluster);
as a request to include "parent.pm". Is there a more
backwards-compatible way to spell that?

Not that I know of. Googling tells me it was a core module from 5.10.1. I can revert tomorrow if necessary :-(

I must have been half asleep when I posted this.

There are three ways out of this that I can see:

. carry a copy of parent.pm in src/test/perl (It's very small)

. use the older and heavier 'base' module which goes back to 5.004, and
does much the same thing (and a lot more)

. just do directly what parent.pm's import() does, as in the attached,
which I have tested down to version 10.

On the whole I think I prefer the last.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

Andrew Dunstan <andrew@dunslane.net> writes:

There are three ways out of this that I can see:
. carry a copy of parent.pm in src/test/perl (It's very small)
. use the older and heavier 'base' module which goes back to 5.004, and
does much the same thing (and a lot more)
. just do directly what parent.pm's import() does, as in the attached,
which I have tested down to version 10.

On the whole I think I prefer the last.

I like #3 too, given that it's such a small patch. I've verified
that it allows regression tests to run on perl 5.8.8, which is the
oldest I have handy other than prairiedog's 5.8.3 (but that machine
is tied up and I don't feel like trying to test this concurrently
with a buildfarm run).

regards, tom lane

On 3/31/22 12:50, Tom Lane wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

There are three ways out of this that I can see:
. carry a copy of parent.pm in src/test/perl (It's very small)
. use the older and heavier 'base' module which goes back to 5.004, and
does much the same thing (and a lot more)
. just do directly what parent.pm's import() does, as in the attached,
which I have tested down to version 10.
On the whole I think I prefer the last.

I like #3 too, given that it's such a small patch. I've verified
that it allows regression tests to run on perl 5.8.8, which is the
oldest I have handy other than prairiedog's 5.8.3 (but that machine
is tied up and I don't feel like trying to test this concurrently
with a buildfarm run).

I think it's safe enough. Pushed.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com