Thread: prevent 006_transfer_modes.pl from leaving files behind
The other pg_upgrade tests chdir to tmp_check prior to running pg_upgrade
to avoid leaving behind delete_old_cluster.{sh,bat}. 006_transfer_modes.pl
should, too. However, this test is a little different because it loops
over all of the available transfer modes and runs pg_upgrade for each one
supported by the platform. From my testing and code analysis, it seems
sufficient to change the directory once at the beginning of the test, but
we could alternatively save the current directory and change back to it in
each iteration to be safe.
Thoughts?
--
nathan
Attachment
On Mon, Apr 07, 2025 at 04:45:52PM -0500, Nathan Bossart wrote:
> The other pg_upgrade tests chdir to tmp_check prior to running pg_upgrade
> to avoid leaving behind delete_old_cluster.{sh,bat}. 006_transfer_modes.pl
> should, too. However, this test is a little different because it loops
> over all of the available transfer modes and runs pg_upgrade for each one
> supported by the platform. From my testing and code analysis, it seems
> sufficient to change the directory once at the beginning of the test, but
> we could alternatively save the current directory and change back to it in
> each iteration to be safe.
>
> Thoughts?
Hmm. Doing one chdir at the beginning of the test should be OK,
because we don't do a move to a different directory within the test
for each transfer mode tested. So your patch looks like the simplest
thing to do to avoid the generation of these files. Note that
delete_old_cluster.sh would be left around even if not using a VPATH
build with ./configure (your commit message does not mention that).
Even if .gitignore discards it, the file is here.
--
Michael
Attachment
On 2025-04-07 Mo 7:41 PM, Michael Paquier wrote:
> On Mon, Apr 07, 2025 at 04:45:52PM -0500, Nathan Bossart wrote:
>> The other pg_upgrade tests chdir to tmp_check prior to running pg_upgrade
>> to avoid leaving behind delete_old_cluster.{sh,bat}. 006_transfer_modes.pl
>> should, too. However, this test is a little different because it loops
>> over all of the available transfer modes and runs pg_upgrade for each one
>> supported by the platform. From my testing and code analysis, it seems
>> sufficient to change the directory once at the beginning of the test, but
>> we could alternatively save the current directory and change back to it in
>> each iteration to be safe.
>>
>> Thoughts?
> Hmm. Doing one chdir at the beginning of the test should be OK,
> because we don't do a move to a different directory within the test
> for each transfer mode tested. So your patch looks like the simplest
> thing to do to avoid the generation of these files. Note that
> delete_old_cluster.sh would be left around even if not using a VPATH
> build with ./configure (your commit message does not mention that).
> Even if .gitignore discards it, the file is here.
I don't think that matters. In non-vpath builds we expect the source
directory to be scribbled on. All sorts of things might be left around.
cheers
andrew
--
Andrew Dunstan
EDB: https://www.enterprisedb.com
Andrew Dunstan <andrew@dunslane.net> writes:
> On 2025-04-07 Mo 7:41 PM, Michael Paquier wrote:
>> delete_old_cluster.sh would be left around even if not using a VPATH
>> build with ./configure (your commit message does not mention that).
>> Even if .gitignore discards it, the file is here.
> I don't think that matters. In non-vpath builds we expect the source
> directory to be scribbled on. All sorts of things might be left around.
It's okay as long as .gitignore ignores it and "make clean" removes it.
When/if we give up makefile builds, a lot of these maintenance
issues will go away...
regards, tom lane
On 2025-04-08 Tu 10:17 AM, Tom Lane wrote: > Andrew Dunstan <andrew@dunslane.net> writes: >> On 2025-04-07 Mo 7:41 PM, Michael Paquier wrote: >>> delete_old_cluster.sh would be left around even if not using a VPATH >>> build with ./configure (your commit message does not mention that). >>> Even if .gitignore discards it, the file is here. >> I don't think that matters. In non-vpath builds we expect the source >> directory to be scribbled on. All sorts of things might be left around. > It's okay as long as .gitignore ignores it and "make clean" removes it. > > When/if we give up makefile builds, a lot of these maintenance > issues will go away... Maybe some will, but I observed the present issue on fairywren which is using meson. cheers andrew -- Andrew Dunstan EDB: https://www.enterprisedb.com
Committed. -- nathan