From: Tom Lane <tgl@sss.pgh.pa.us>
To: Chao Li <li.evan.chao@gmail.com>
cc: pgsql-hackers <pgsql-hackers@lists.postgresql.org>,
        Robert Haas <robertmhaas@gmail.com>
Subject: Re: astreamer_lz4: fix bug of output pointer advancement in
 decompressor
In-reply-to: <0594CC79-1544-45DD-8AA4-26270DE777A7@gmail.com>
References: <0594CC79-1544-45DD-8AA4-26270DE777A7@gmail.com>
Comments: In-reply-to Chao Li <li.evan.chao@gmail.com>
	message dated "Mon, 02 Mar 2026 17:17:40 +0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1531717.1772644615.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Wed, 04 Mar 2026 12:16:55 -0500
Message-ID: <1531718.1772644615@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1531718.1772644615%40sss.pgh.pa.us>
Precedence: bulk

Chao Li <li.evan.chao@gmail.com> writes:
> There have been a couple of LZ4-related patches recently, so I spent som=
e time playing with the LZ4 path and found a bug in astreamer_lz4_decompre=
ssor_content().

Yup, that's clearly wrong.  I failed to reproduce a crash with the
test hack you suggested, but no matter.  Pushed with some cosmetic
editorialization.

The track record of all this client-side-compression logic is
really quite awful :-(.  Another thing that I'm looking askance
at is the error handling, or rather lack of it:

        ret =3D LZ4F_decompress(mystreamer->dctx,
                              next_out, &out_size,
                              next_in, &read_size, NULL);

        if (LZ4F_isError(ret))
            pg_log_error("could not decompress data: %s",
                         LZ4F_getErrorName(ret));

        ... continue on our merry way ...

I suspect whoever wrote this thought pg_log_error is equivalent
to elog(ERROR), but it's not; it just prints a message.  It seems
highly unlikely to me that continuing onwards will result in a
good outcome.  I'm a bit inclined to s/pg_log_error/pg_fatal/
throughout these files, at least in places where there's no
visible effort to handle the error.

			regards, tom lane