From: Tom Lane <tgl@sss.pgh.pa.us>
To: "=?utf-8?B?aWxtYXJp?=" <ilmari@ilmari.org>
cc: "=?utf-8?B?V2VpIFN1bg==?=" <936739278@qq.com>,
        "=?utf-8?B?cGdzcWwtYnVncw==?=" <pgsql-bugs@lists.postgresql.org>
Subject: Re: Build error in Solaris
In-reply-to: <1176583.1644323289@sss.pgh.pa.us>
References: <tencent_2EFCC8BA0107B6EC0F97179E019A8A43C806@qq.com>
 <1176583.1644323289@sss.pgh.pa.us>
Comments: In-reply-to Tom Lane <tgl@sss.pgh.pa.us>
	message dated "Tue, 08 Feb 2022 07:28:09 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1290985.1644357181.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 08 Feb 2022 16:53:01 -0500
Message-ID: <1290986.1644357181@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1290986.1644357181%40sss.pgh.pa.us>
Precedence: bulk

I wrote:
> Hm.  I saw something similar on my old dinosaur pademelon, but wrote it
> off as "they don't care about pre-C99 compilers anymore".  But now I
> dug into ppport.h a bit more, and I think it's actually busted for
> non-gcc compilers: eval_pv expands into something that tries to use
> STMT_START/STMT_END inside a plain parenthesized expression (via
> croak_sv).

> Maybe the bleeding edge Devel::PPPort would do better?  I see

Nope, 3.64 is also broken.

I traced this to this stanza, originally added in Devel-PPPort 3.54:

/* Older Perl versions have broken croak_on_error=3D1 */
#if (PERL_BCDVERSION < 0x5031002)
# ifdef eval_pv
#  undef eval_pv
#  if defined(PERL_USE_GCC_BRACE_GROUPS)
#   define eval_pv(p, croak_on_error) ({ SV *_sv =3D Perl_eval_pv(aTHX_ p,=
 0); D_PPP_CROAK_IF_ERROR(croak_on_error); _sv; })
#  else
#   define eval_pv(p, croak_on_error) ((PL_Sv =3D Perl_eval_pv(aTHX_ p, 0)=
), D_PPP_CROAK_IF_ERROR(croak_on_error), PL_Sv)
#  endif
# endif
#endif

D_PPP_CROAK_IF_ERROR goes to croak_sv which, in older Perl
versions, produces a statement block.  So anything that won't
accept gcc-like statements within expressions blows up.

Now, we don't actually *care* about the case this is trying
to fix, because all our eval_pv usages pass constant-FALSE
for croak_on_error.  (Possibly not unrelated to the bug
implied by the above comment ... but I digress.)  Therefore,
pending a resolution to the issue I just filed [1], I suggest
that we just dike this stanza out of our copy of ppport.h.

			regards, tom lane

[1] https://github.com/Dual-Life/Devel-PPPort/issues/216