standards/51292: [PATCH] add ecvt()/fcvt()/gcvt() functions (SUSv3)

[Alexey Zelkin]

The following reply was made to PR standards/51292; it has been noted by GNATS.

From: Alexey Zelkin <phantom at FreeBSD.org.ua>
To: "Sergey A. Osokin" <osa at freebsd.org.ru>
Cc: FreeBSD-gnats-submit at freebsd.org
Subject: Re: standards/51292: [PATCH] add ecvt()/fcvt()/gcvt() functions (SUSv3)
Date: Fri, 25 Apr 2003 15:19:37 +0300

 hi,
 
 On Fri, Apr 25, 2003 at 12:38:51PM +0400, Sergey A. Osokin wrote:
 
 > > > >Number:         51292
 > > > >Category:       standards
 > > > >Synopsis:       [PATCH] add ecvt()/fcvt()/gcvt() functions (SUSv3)
 > > 
 > > Few questions related to code:
 > > 
 > > 1. What's a reason to have some LANG handling logic here ?
 > 
 > Because standart says:
 > The radix character is determined by the current locale.
 > Follow this way: check LANG, then set locale.
 
 First sentence is correct, but second one is wrong.  radix character
 is a 'property' of LC_NUMERIC locale category.  So, having LANG=C
 and LC_NUMERIC=ru_RU.KOI8-R should use russian locale radix character.
 
 Actually, all these cases should be handled by setlocale() itself.  And
 after call to 'setlocale()' you should use lconv() to receive current
 radix character.
 
 I also would object to using of setlocale() from libc's function internally.
 It's a application's problem to decide which locale to use and [fge]cvt()
 should use locale previously set by application.
 
 > > 2. Did you check correctness of dtoa()'s usage here ?  It's not easy
 > > and after last netlib's import it become more uneasy.
 
 From reading of SUSv3's section related to these functions I'd suggest
 following way to check a function (and maybe write a regression-test
 for these funcs).
 
 1. declare some set of floating values (more than one)
 2. and try comapring result of gcvt() against of result of printf("%g").
 same for others.
 3. check these results against predefined values (if you'll write a
 regression test program)