From: Tom Lane <tgl@sss.pgh.pa.us>
To: Michael Paquier <michael@paquier.xyz>
cc: Thomas Munro <thomas.munro@enterprisedb.com>,
        Alvaro Herrera <alvherre@2ndquadrant.com>,
        PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: Allowing printf("%m") only where it actually works
In-reply-to: <20180925071636.GK1354@paquier.xyz>
References: <8770.1533931511@sss.pgh.pa.us> <16896.1533935042@sss.pgh.pa.us>
 <14925.1534040609@sss.pgh.pa.us> <1437.1534100930@sss.pgh.pa.us>
 <13183.1534624490@sss.pgh.pa.us> <30022.1534705920@sss.pgh.pa.us>
 <20180912070739.GK25160@paquier.xyz> <1304.1536774001@sss.pgh.pa.us>
 <20180914020256.GD29332@paquier.xyz> <13585.1537809527@sss.pgh.pa.us>
 <20180925071636.GK1354@paquier.xyz>
Comments: In-reply-to Michael Paquier <michael@paquier.xyz>
	message dated "Tue, 25 Sep 2018 16:16:36 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <4417.1537882226.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 25 Sep 2018 09:30:26 -0400
Message-ID: <4418.1537882226@sss.pgh.pa.us>
Precedence: bulk

Michael Paquier <michael@paquier.xyz> writes:
> On Mon, Sep 24, 2018 at 01:18:47PM -0400, Tom Lane wrote:
>> Well, we have to change the code somehow to make it usable in frontend
>> as well as backend.  And we can *not* have it do exit(1) in libpq.
>> So the solution I chose was to make it act the same as if FormatMessage
>> were to fail.  I don't find this behavior unreasonable: what is really
>> important is the original error code, not whether we were able to
>> pretty-print it.  I think the ereport(FATAL) coding is a pretty darn
>> bad idea even in the backend.

> Ok.  I won't fight hard on that.  Why changing the error message from
> "could not load netmsg.dll" to "unrecognized winsock error" then?  The
> original error string is much more verbose to grab the context.

As the code stands, what you'll get told about is the error code
returned by the failed LoadLibrary call; the original winsock error
code is reported nowhere.  I think that's backwards.

We could possibly write something like

sprintf(wserrbuf, "winsock error %d (could not load netmsg.dll to translat=
e: error code %lu)", err, GetLastError())));

but I'm unconvinced that that's useful.

			regards, tom lane