From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
cc: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>,
        Ian Barwick <ian.barwick@2ndquadrant.com>,
        "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Subject: Re: PG12, PGXS and linking pgfeutils
In-reply-to: <20190509173911.GA16616@alvherre.pgsql>
References: <20190509173911.GA16616@alvherre.pgsql>
Comments: In-reply-to Alvaro Herrera <alvherre@2ndquadrant.com>
	message dated "Thu, 09 May 2019 13:39:11 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <27314.1557424032.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Thu, 09 May 2019 13:47:12 -0400
Message-ID: <27315.1557424032@sss.pgh.pa.us>
Precedence: bulk

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-May-09, Tom Lane wrote:
>> I'm not especially on board with the idea of moving FE-specific error
>> handling code into libpgcommon, as that breaks the concept that
>> src/common/ is broadly for code that can work in either frontend or
>> backend contexts.  However, we already have a few violations of that
>> rule: common/Makefile already has
>> =

>> # A few files are currently only built for frontend, not server
>> OBJS_FRONTEND =3D $(OBJS_COMMON) fe_memutils.o file_utils.o restricted_=
token.o
>> =

>> So maybe the answer is to move these logging support functions into
>> src/common, in a file that's only built for frontend.

> I wonder if a better solution isn't to move the file_utils stuff to
> fe_utils.  Half of it is frontend-specific.  The only one that should be
> shared to backend seems to be fsync_fname ... but instead of sharing it,
> we have a second copy in fd.c.

Hm, if file_utils is the only thing in common/ that uses this, and we
expect that to remain true, that would fix the issue.  But ...

The thing I was looking at was mainly fe_memutils, which is justifiably
here on the grounds that it provides backend-like palloc support and
thereby eases the task of making other common/ modules work in both
contexts.  If we built elog/ereport emulations on top of Peter's logging
functions, there'd be a very clear case for having that in common/.
Peter didn't do that for v12, but I hope we get there at some point.

			regards, tom lane