Re: [HACKERS] pl/perl extension fails on Windows

On Fri, Jul 28, 2017 at 7:22 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Ashutosh Sharma <ashu.coek88@gmail.com> writes:
>> Thanks for the patch. I am seeing some compilation errors on Windows
>> with the patch. Below are the errors observed,
>> ...
>> I did spent some time to find reason for these compilation errors and
>> could eventually find that some of the Windows specific functions
>> inside plperl module are calling Perl APIs without fetching the perl
>> interpreter's context using dTHX.
>
> Ah, thanks.  I just stuck that in where compiler errors were telling
> me to, so I didn't realize there were Windows-specific functions to
> worry about.
>
>> Moreover, I have also tested this patch along with the patch to import
>> switches used by perl into plperl and together it fixes the server
>> crash issue. Also, now, the interpreter size in both perl and plperl
>> module are the same thereby generating the same key on both plperl and
>> perl module. Thanks.
>
> Excellent.  So this looks like the avenue to pursue.
>
> As far as the question of which symbols to import goes: on my own
> machines I'm seeing a lot of things like
>
> $ perl -MConfig -e 'print $Config{ccflags}'
> -D_REENTRANT -D_GNU_SOURCE -fno-strict-aliasing -pipe -fstack-protector -I/usr/local/include -D_LARGEFILE_SOURCE
-D_FILE_OFFSET_BITS=64
>
> $ perl -MConfig -e 'print $Config{ccflags}'
>  -Ae -D_HPUX_SOURCE -Wl,+vnocompatwarnings -DDEBUGGING -I/usr/local/include -D_LARGEFILE_SOURCE
-D_FILE_OFFSET_BITS=64
>
> I'm really quite afraid to import symbols like _LARGEFILE_SOURCE and
> _FILE_OFFSET_BITS into plperl; those *will* break things if they
> are different from what core Postgres is built with.  Moreover,
> I came across a relevant data structure in perl.h:
>
> /* These are all the compile time options that affect binary compatibility.
>    Other compile time options that are binary compatible are in perl.c
>    Both are combined for the output of perl -V
>    However, this string will be embedded in any shared perl library, which will
>    allow us add a comparison check in perlmain.c in the near future.  */
> #ifdef DOINIT
> EXTCONST char PL_bincompat_options[] =
> #  ifdef DEBUG_LEAKING_SCALARS
>                              " DEBUG_LEAKING_SCALARS"
> #  endif
> #  ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
>                              " DEBUG_LEAKING_SCALARS_FORK_DUMP"
> #  endif
> #  ifdef FAKE_THREADS
>                              " FAKE_THREADS"
> #  endif
> #  ifdef MULTIPLICITY
>                              " MULTIPLICITY"
> #  endif
> ... lots more ...

Thanks for sharing this information. I too had a look into
'PL_bincompat_options' data structure in perl.h and i didn't see any
macro name starting with underscore. Based on this information, we can
now confidently say that we do not need any -D switches starting with
underscore for binary compatibility purpose.

>
> Assuming that the Perl crew know what they're doing and this list is
> complete, I notice that not one of the symbols they show as relevant
> starts with an underscore.  So I'm thinking that my previous semi-
> joking idea of absorbing only -D switches for names that *don't*
> start with an underscore was actually a good solution.

Yes, it was a good solution indeed.

If that
> turns out to not be enough of a filter, we could consider looking
> into perl.h to extract the set of symbols tested in building
> PL_bincompat_options and then intersecting that with what we get
> from Perl's ccflags.  But that would be a lot more complex, so
> I'd rather go with the simpler filter rule for now.

Okay, as per your suggestion, I have modified my earlier patches that
imports the -D switches used by perl into plperl accordingly i.e. it
now ignores the switches whose name starts with underscore. Please
find plperl_win_v3 and plperl_linux_v2 patches attached with this
email.

>
> (BTW, you never did tell us exactly what defines you're getting
> out of Perl's flags on the problem installation.)

I am really sorry about that. I just missed that in my earlier email.
Here are the defines used in the perl where i could reproduce the
issue,

C:\Users\ashu>perl -MConfig -e "print $Config{ccflags}"
-nologo -GF -W3 -O1 -MD -Zi -DNDEBUG -GL -fp:precise -DWIN32
-D_CONSOLE -DNO_STRICT -DWIN64 -DCONSERVATIVE  -DUSE_SITECUSTOMIZE
-DPERL_TEXTMODE_SCRIPTS -DPERL_IMPLICIT_CONTEXT -DPERL_IMPLICIT_SYS

--
With Regards,
Ashutosh Sharma
EnterpriseDB:http://www.enterprisedb.com

-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers