From: Tom Lane <tgl@sss.pgh.pa.us>
To: Pavel Raiskup <praiskup@redhat.com>
cc: Robert Haas <robertmhaas@gmail.com>,
        "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>,
        Devrim =?ISO-8859-1?Q?G=DCND=DCZ?= <devrim@gunduz.org>
Subject: Re: plruby: rb_iterate symbol clash with libruby.so
In-reply-to: <1772202.Bn4MKCBuZf@nb.usersys.redhat.com>
References: <4738198.8KVIIDhgEB@nb.usersys.redhat.com>
 <20461.1541269186@sss.pgh.pa.us>
 <CA+TgmoYbguBrde0XrXpdHkFZcB3Bjc+oeR2T8Xjb1k+rzJCrPQ@mail.gmail.com>
 <1772202.Bn4MKCBuZf@nb.usersys.redhat.com>
Comments: In-reply-to Pavel Raiskup <praiskup@redhat.com>
	message dated "Tue, 06 Nov 2018 13:43:03 +0100"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <22880.1541515786.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 06 Nov 2018 09:49:46 -0500
Message-ID: <22881.1541515786@sss.pgh.pa.us>
Precedence: bulk

Pavel Raiskup <praiskup@redhat.com> writes:
> On Monday, November 5, 2018 9:06:41 PM CET Robert Haas wrote:
>> On Sat, Nov 3, 2018 at 2:20 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>> Is it realistic we could rename red-black tree methods from 'rb_*' to=
 e.g.
>>>> 'rbt_*' to avoid this clash?

>>> I don't have a huge objection to renaming the rbtree functions, other
>>> than the precedent it sets ...

>> Maybe prefixing with pg_ would better than rb_ to rbt_. ...
>> It's certainly not very surprising that Ruby has symbols starting with
>> rb_...

> I now realized that there's rb_block_call() alternative for rb_iterate()
> Ruby call -- which fortunately doesn't collide with PostgreSQL internals=
.
> It means that for sufficiently new Ruby there exists some solution (not
> that something similar can not re-appear elsewhere).

Yeah.  The long and short of this is that we're trampling on namespace
that reasonably belongs to Ruby --- if they had some functions named
"pg_something" and complained about a collision with libpq, would we
change?  Nope.  So really we should rename these.

After looking at the code a bit I like the idea of s/rb/rbt/g
better than s/rb/pg_rb/g.  The latter seems verbose, and it would
also open the question of whether we need to rename rbtree.h/.c,
which would be an additional level of complication I don't want.
The rbt approach will allow skipping some other renamings that
would be needed for consistency if we use pg_rb.

Barring objections I'll go make this happen shortly.

It's too late for this week's releases, but Pavel could pick up the commit
once it happens and carry it as a Fedora patch until the next releases.

			regards, tom lane