Date: Sun, 16 Jun 2019 19:45:45 +0200 (CEST)
From: Fabien COELHO <coelho@cri.ensmp.fr>
To: Tom Lane <tgl@sss.pgh.pa.us>
cc: Tomas Vondra <tomas.vondra@2ndquadrant.com>,
    Dean Rasheed <dean.a.rasheed@gmail.com>,
    Tomas Vondra <tomas.vondra@postgresql.org>,
    pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Rework the pg_statistic_ext catalog
In-Reply-To: <4352.1560698164@sss.pgh.pa.us>
Message-ID: <alpine.DEB.2.21.1906161933240.21115@lancre>
References: <E1hcHz2-0005mY-5A@gemulon.postgresql.org>
 <8833.1560647898@sss.pgh.pa.us> <20156.1560650744@sss.pgh.pa.us>
 <CAEZATCWWX5O57d=MHY=MqyJe-xHG2pbi75LKLjo6aZLP3zsaow@mail.gmail.com>
 <20190616085651.d2hxuiqlojewed5j@development>
 <20190616100328.shl2hhuy2pup6ges@development>
 <20190616112021.2mncnuj26sbsbnzx@development>
 <14081.1560693790@sss.pgh.pa.us> <4352.1560698164@sss.pgh.pa.us>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII; format=flowed
Precedence: bulk


Hello Tom,

>> I see that Coelho's two bleeding-edge-clang animals are reporting that
>> failure too.  Normally I'd just ignore those two;

And you'd be right.

>> they break pretty regularly.  Maybe you're using an 
>> almost-bleeding-edge clang?
>
> Oh --- I managed to reproduce that failure locally on RHEL6 (nothing
> bleeding-edge anywhere), but it went away after make-clean-and-rebuild.

> I'm suspicious that the underlying cause was failure to recompile
> something that knows the value of CATALOG_VERSION_NO.

Hmmm. My animals compile out of the source tree, not sure how this could 
happen.

> This theory is insufficient to explain why Coelho's animals failed,
> though.  Maybe it would, if you also posit a ccache screwup?  But it's
> not obvious from their configurations that they're using ccache at all.

Indeed, no ccache. The compilers change often (well, once a week), I do 
not want anything kept across compiler versions. The animals are really 
testing the compilers as much as they are testing postgres.

However, there is an autoconf cache, but I do not see why it would have 
such an effect.

So no clue.

-- 
Fabien.