[postgis-devel] [postgis-tickets] r14592 - Patch from Sebastiaan Couwenberg to fix test_wkb_out_point failure on hppa & mips.

[Regina Obe]

Greg,

I committed to both trunk and branch 2.2.  I blindly trusted Sebastian's
judgement on things since it was just a regression test (not code change),
he had tested on platforms I have no access to,
and the tests still passed under my mingw64 install.  I'm not knowledgeable
enough in C to judge code.

Perhaps Paul , Sandro, or others can chime in on this.

Thanks,
Regina

-----Original Message-----
From: Greg Troxel [mailto:gdt at ir.bbn.com] 
Sent: Tuesday, January 12, 2016 7:44 PM
To: Regina Obe <lr at pcorp.us>
Cc: postgis-devel at lists.osgeo.org
Subject: Re: [postgis-tickets] r14592 - Patch from Sebastiaan Couwenberg to
fix test_wkb_out_point failure on hppa & mips.


Regina Obe <lr at pcorp.us> writes:

> Author: robe
> Date: 2016-01-12 16:36:35 -0800 (Tue, 12 Jan 2016) New Revision: 14592
>
> Modified:
>    branches/2.2/liblwgeom/cunit/cu_out_wkb.c
> Log:
> Patch from Sebastiaan Couwenberg to fix test_wkb_out_point failure on hppa
& mips.
> references #3426

I'm a little surprised at changes going onto the release branch without
first being on trunk and having review and then getting cherrypicked onto
the release branch.

> +/* parisc and mips (at least some processors) have a different nan 
> +representation from other arches. */ #if !defined(__hppa__) && 
> +!defined(__mips__) # define nan_val( v1, v2)  v1 #else # define 
> +nan_val( v1, v2)  v2 #endif

This seems quite awkward.  It would seem the tests should be grounded in the
standards (presumably postgis requires IEEE754 to start and doesn't work on
the vax :-).  Isn't there some way to start with a nan expressed as a
double, and then to convert that to a sql representation?  Or to write an
"isnan_wkb" function that takes the string/hex representation, extracts a
double, and then uses the C99-specified isnan(3) procedures.
Encoding of IEEE754 values into hex and having them in the tests seems
nonrobust, and it seems extra fragile to have switches by arch.