Horton

Re: pl/perl extension fails on Windows

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
Cc: Robert Haas <robertmhaas@gmail.com>, Ashutosh Sharma <ashu.coek88@gmail.com>, Dave Page <dpage@pgadmin.org>, Sandeep Thakkar <sandeep.thakkar@enterprisedb.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2017-07-27T22:50:53Z
Lists: pgsql-hackers

Attachments

Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> On 07/27/2017 04:33 PM, Tom Lane wrote:
>> So I was trying to figure a way to not include XSUB.h except in a very
>> limited part of plperl, like ideally just the .xs files.  It's looking
>> like that would take nontrivial refactoring though :-(.  Another problem
>> is that this critical bit of the library API is in XSUB.h:

> That's the sort of thing that prompted me to ask what was the minimal
> set of defines required to fix the original problem (assuming such a
> thing exists)
> We haven't used PERL_IMPLICIT_CONTEXT to date, and without ill effect.
> For example. it's in the ExtUtils::Embed::ccopts for the perl that
> jacana and bowerbird happily build and test against.

Well, actually, PERL_IMPLICIT_CONTEXT is turned on automatically in any
MULTIPLICITY build, and since it changes all the Perl ABIs, we *are*
relying on it.  However, after further study of the Perl docs I noticed
that we could dispense with XSUB.h if we defined PERL_NO_GET_CONTEXT
(which turns the quoted stanza into a no-op).  That results in needing to
sprinkle plperl.c with "dTHX" declarations, as explained in perlguts.pod.
They're slightly tricky to place correctly, because they load up a pointer
to the current Perl interpreter, so you have to be wary of where to put
them in functions that change interpreters.  But I seem to have it working
in the attached patch.  (One benefit of doing this extra work is that it
should be a bit more efficient, in that we load up a Perl interpreter
pointer only once per function called, not once per usage therein.  We
could remove many of those fetches too, if we wanted to sprinkle the
plperl code with yet more THX droppings; but I left that for another day.)

Armed with that, I got rid of XSUB.h in plperl.c and moved the
PG_TRY-using functions in the .xs files to plperl.c.  I think this would
fix Ashutosh's problem, though I am not in a position to try it with a
PERL_IMPLICIT_SYS build here.

			regards, tom lane

Commits

  1. MSVC: Test whether 32-bit Perl needs -D_USE_32BIT_TIME_T.

  2. Further tweaks to compiler flags for PL/Perl on Windows.

  3. Absorb -D_USE_32BIT_TIME_T switch from Perl, if relevant.

  4. PL/Perl portability fix: absorb relevant -D switches from Perl.

  5. PL/Perl portability fix: avoid including XSUB.h in plperl.c.