MIME-Version: 1.0
References: 
 <CA+hUKG+ixUUYOGRcuZpkk5pmJZaUQv6VCPAjdTZXFP5vA8jxcA@mail.gmail.com>
 <854785.1772206539@sss.pgh.pa.us>
 <rc5bkn7ikwo6cir5w4jilxtqeiczdup7ytooav7lufa77brz5p@54gexczvq25p>
In-Reply-To: <rc5bkn7ikwo6cir5w4jilxtqeiczdup7ytooav7lufa77brz5p@54gexczvq25p>
From: Thomas Munro <thomas.munro@gmail.com>
Date: Wed, 11 Mar 2026 13:05:22 +1300
Message-ID: 
 <CA+hUKG+-=A-xGCKecQF8kG=agMJLoK31gutaS+nVLK=iL-oWMQ@mail.gmail.com>
Subject: Re: A stack allocation API
To: Andres Freund <andres@anarazel.de>
Cc: Tom Lane <tgl@sss.pgh.pa.us>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CA%2BhUKG%2B-%3DA-xGCKecQF8kG%3DagMJLoK31gutaS%2BnVLK%3DiL-oWMQ%40mail.gmail.com>
Precedence: bulk

On Sat, Feb 28, 2026 at 7:02=E2=80=AFAM Andres Freund <andres@anarazel.de> =
wrote:
> I wish msvc implemented something akin to gcc/clang's
> attribute(cleanup(cleanup_function)), but it doesn't look like it
> does. Obviously it would if we were to compile with C++, but I don't thin=
k
> anybody has appetite for the work it'd need to get there.

Well, it does have __try/__finally, but the future of the idea for C2y
looks like Go or Zig:

    #include <stddefer.h>

    p =3D malloc();
    defer free(p);

It's already available in bleeding edge GCC and Clang 22, based on the
TS 25755 draft[1].  Unfortunately for the real world, it's annoyingly
difficult to come up with *nice* looking macros that can expand to the
dialects we'd need to cover even if we were OK with abandoning the
notion of supporting C < 2y compilers other than the 3 we talk about,
due to syntactic structure and semantic differences, presumably:

1. C2y defer
2. C++ RAII
3. GCC/Clang cleanup attributes
4. MSVC __try/__finally

But with big honking framing macros BEGIN_MAGIC_PIXIE_DUST(), ...
pg_defer_pfree(p) (assuming we'd want a style with a pathway to the
future standard?); ... END_MAGIC_PIXIE_DUST(), I think it's probably
doable...  There are lots of people trying to suffer through
portable-enough-for-me and just-enough-functionality-for-my-project,
maybe some inspiration[2]...

[1] https://www.open-std.org/JTC1/SC22/WG14/www/docs/n3734.pdf
[2] https://antonz.org/defer-in-c/