BUG #15065: ActivePerl 5.24.3 breaks PG compilation on Windows - Mailing list pgsql-bugs

From PG Bug reporting form
Subject BUG #15065: ActivePerl 5.24.3 breaks PG compilation on Windows
Date
Msg-id 151861437507.1449.1024662815700270598@wrigleys.postgresql.org
Whole thread Raw
Responses Re: BUG #15065: ActivePerl 5.24.3 breaks PG compilation on Windows
List pgsql-bugs
The following bug has been logged on the website:

Bug reference:      15065
Logged by:          Heath Lord
Email address:      heath.lord@crunchydata.com
PostgreSQL version: 9.6.7
Operating system:   Windows
Description:

   With the latest release of ActivePerl 5.24.3 they are installing both a
.a and a .lib with their installation.  While this normally wouldn't be an
issue, the logic that is being used in the PG source checks for one or the
other and if it finds both it will fail with an error stating that it could
not find them.  
   In the postgres/src/tools/msvc/Mkvcbuild.pm file the following logic
exists:

# ActivePerl 5.16 provided perl516.lib; 5.18 provided libperl518.a
                my @perl_libs =
                  grep { /perl\d+\.lib$|libperl\d+\.a$/ }
glob($perl_path);
                if (@perl_libs == 1)
                {
                        $plperl->AddLibrary($perl_libs[0]);
                }

  The issue is that with the latest ActivePerl release the if statement
results in a value of 2 which causes this to fail and it falls through to
the else which is the failure case.  This causes all of the Windows builds
to fail if using the latest version of ActivePerl.
    There are two possible solutions that I am proposing:

1) [msvc_perl_lib_check_for_two_use_second.patch] 
             Add an elsif clause and check if @perl_libs is 2 and then use
the libperl###.a which I have tested and everything compiles fine and passes
all tests.  

2) [msvc_perl_lib_check_inclusive_use_first_found.patch] 
       Change the if statement to check for 1 or more perl libs found and
then always use the first one that was found.  I have also built and tested
that this passes all regression.

   Both of these options work and have been built and tested.  The resulting
generated binaries are of identical size as well.  

    I am not sure given the fact that this breaks all windows builds using
this version of ActivePerl if this change needs to be backpatched or not?
Thank you in advance for any help or assistance given to get one of these
fixes into the PG code base.

   I will be attaching the patches referenced above in a follow up reply
after the bug has been submitted.

-Heath


pgsql-bugs by date:

Previous
From: PG Bug reporting form
Date:
Subject: BUG #15064: Deadlock not detected on standby.
Next
From: Heath Lord
Date:
Subject: Re: BUG #15065: ActivePerl 5.24.3 breaks PG compilation on Windows