From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
cc: Dean Rasheed <dean.a.rasheed@gmail.com>,
        Michael Paquier <michael@paquier.xyz>,
        PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
        laetitia.avrot@gmail.com
Subject: Re: pgsql: Add support for hyperbolic functions, as well as log10().
In-reply-to: <9aae40da-9572-a82b-a283-d2158ed6ab0c@2ndQuadrant.com>
References: <E1h3nUY-0000sM-Vf@gemulon.postgresql.org>
 <20190313022159.GR13812@paquier.xyz> <24813.1552447002@sss.pgh.pa.us>
 <20190313050331.GT13812@paquier.xyz> <17820.1552514192@sss.pgh.pa.us>
 <CAEZATCX-mFwmx0NKbGsYJ6RAo8Hb=c+_CeNNXWQ7-BmK2_XVWw@mail.gmail.com>
 <6989.1552538487@sss.pgh.pa.us>
 <9aae40da-9572-a82b-a283-d2158ed6ab0c@2ndQuadrant.com>
Comments: In-reply-to Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
	message dated "Thu, 14 Mar 2019 08:54:49 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <8794.1552590496.1@sss.pgh.pa.us>
Date: Thu, 14 Mar 2019 15:08:16 -0400
Message-ID: <8795.1552590496@sss.pgh.pa.us>
Precedence: bulk

Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> On 3/14/19 12:41 AM, Tom Lane wrote:
>> So far, no other failures ...

> I have replicated this on my Msys2 test system.
> I assume it's a bug in the mingw math library. I think jacana is the
> only currently reporting mingw member :-( The MSVC members appear to be
> happy.
> I have several releases of the mingw64 toolsets installed on jacana -
> I'll try an earlier version to see if it makes a difference.

Yeah, it would be interesting to know whether it's consistent across
different mingw versions.

So far, though, jacana is still the only buildfarm animal that's having
trouble with those tests as of c015f853b.  I want to wait another day or
so in hopes of getting more reports from stragglers.  But assuming that
that stays true, I do not feel any need to try to work around jacana's
issues.  We already have proof of two deficiencies in their
hyoerbolic-function code, and considering the tiny number of test cases
we've tried, it'd be folly to think there are only two.  I don't want
to embark on a project to clean that up for the sake of one substandard
implementation.

I feel therefore that what we should do (barring new evidence) is either

1. Remove all the inf/nan test cases for the hyoerbolic functions, on
the grounds that they're not really worth expending buildfarm cycles on
in the long run; or

2. Just comment out the one failing test, with a note about why.

I haven't got a strong preference as to which.  Thoughts?

			regards, tom lane