MIME-Version: 1.0
In-Reply-To: 
 <CAM2FmMo5YbxtAa_ZQhP99NqHod=UUyfXoCLrio2DJ344Wb81kQ@mail.gmail.com>
References: 
 <CAM2FmMo5YbxtAa_ZQhP99NqHod=UUyfXoCLrio2DJ344Wb81kQ@mail.gmail.com>
From: Josh Kupershmidt <schmiddy@gmail.com>
Date: Tue, 13 Nov 2012 22:49:12 -0500
Message-ID: 
 <CAK3UJRHi90DyuS0dUGMRf68fov82QB5Mb650kkms0eGzDd3e5g@mail.gmail.com>
Subject: Re: pg_dump -t '"Table"' for cmd.exe
To: Mike Toews <mwtoews@gmail.com>
Cc: pgsql-docs@postgresql.org
Content-Type: text/plain; charset=ISO-8859-1

On Mon, Nov 12, 2012 at 8:45 PM, Mike Toews <mwtoews@gmail.com> wrote:
> I'm not sure if this is worth documenting, but pg_dumping mixed-case tables
> with the '-t table' option appears to not be accurately documented for
> cmd.exe. Here are my four attempts, with only the last as success:

I believe this complaint is basically the same as this recent one:
  http://archives.postgresql.org/pgsql-bugs/2012-10/msg00044.php

and this behavior confused me recently as well. I agree the behavior
should be documented, though it should be documented for all our
utilities which accept a --table argument and pass it on as a
supposedly already-escaped identifier.

Incidentally, I'm really not fond of the existing behavior because it:
 a.) is counterintuitive
 b.) requires the user to escape table names themselves, and how were
they even supposed to have a chance of programmatically doing this
properly before we exposed PQescapeIdentifier()
 c.) opens possible SQL injection holes, e.g.
  reindexdb --table='foo; ALTER ROLE limited WITH superuser'

Josh