Re: TAP tests aren't using the magic words for Windows file access

Hello hackers,

Are there any plans to backport the patch to earlier versions
of the Postgres?
https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=114541d58e5970e51b78b77b65de16210beaab43

We rarely see the issue with the pg_ctl/004_logrotate test on
the REL_12_STABLE branch. On my notebook I can easily reproduce
the "Permission denied at src/test/perl/TestLib.pm line 259"
error with the small change below. But the same test on the
13th version and the 12th version with the TestLib patch does
not fail.

diff --git a/src/bin/pg_ctl/t/004_logrotate.pl 
b/src/bin/pg_ctl/t/004_logrotate.pl                                      
                                                                         
index bc39abd23e4..e49e159bc84 100644                                    
                                                                          
                                               --- 
a/src/bin/pg_ctl/t/004_logrotate.pl                                      
                                                                          
                                           +++ 
b/src/bin/pg_ctl/t/004_logrotate.pl                                      
                                                                          
                                           @@ -72,7 +72,7 @@ for (my 
$attempts = 0; $attempts < $max_attempts; $attempts++)                   
                                                                          
                      {                                                   
                                                                          
                                                                          
  $new_current_logfiles = slurp_file($node->data_dir . 
'/current_logfiles');                                                    
                                                                   last 
if $new_current_logfiles ne $current_logfiles;                           
                                                                          
                                  -       usleep(100_000);                
                                                                          
                                                                          
       +       usleep(1);                                                 
                                                                          
                                                      }                   
                                                                          
                                                                          
                                                                          
                                                                          
                                                                         
note "now current_logfiles = $new_current_logfiles";


On 2019-11-22 20:22, Andrew Dunstan wrote:
> On 11/22/19 3:55 AM, Juan José Santamaría Flecha wrote:
>> 
>> On Fri, Nov 22, 2019 at 9:00 AM Michael Paquier <michael@paquier.xyz
>> <mailto:michael@paquier.xyz>> wrote:
>> 
>>     On Thu, Nov 21, 2019 at 08:09:38PM +0100, Juan José Santamaría
>>     Flecha wrote:
>>     > I think Perl's open() is a bad candidate for an overload, so I
>>     will update
>>     > the previous patch that only touches slurp_file().
>> 
>>     FWIW, I don't like much the approach of patching only 
>> slurp_file().
>>     What gives us the guarantee that we won't have this discussion 
>> again
>>     in a couple of months or years once a new caller of open() is 
>> added
>>     for some new TAP tests, and that it has the same problems with
>>     multi-process concurrency?
>> 
>> 
>> I agree on that, from a technical stand point, overloading open() is
>> probably the best solution for the reasons above mentioned. My doubts
>> come from the effort such a solution will take and its
>> maintainability, also taking into account that there are not that many
>> calls to open() in "src/test/perl".
>> 
>> 
> 
> 
> I think the best course is for us to give your latest patch an outing 
> on
> the buildfarm and verify that the issues seen with slurp_file 
> disappear.
> That shouldn't take us more than a week or two to see - drongo has had 
> 6
> such failures in the last 11 days on master. After that we can discuss
> how much further we might want to take it.
> 
> 
> cheers
> 
> 
> andrew

--
regards,

Roman Zharkov