From: Paul Brannan <pbrannan@...>
Date: 2004-07-03T02:45:41+09:00
Subject: bug in String#to_f in 1.9 (was Re: ruby-dev summary 23690-23762)

On Sat, Jun 26, 2004 at 08:38:51AM +0900, Masayoshi Takahashi wrote:
> Nakamura pointed out that the string "nan" and "inf" are
> used as NaN and Inf in some environments. His patch
> forced to use "NaN" and "Inf". He needs comments
> about this change.

I think it's a good idea to print NaN and Inf uniformly across all
platforms.

However, a thought occurred to me when I was considering this, that it
would be a shame if we print NaN to a string and then were unable to
read it back in, because to_f didn't handle the case change correctly.
It should work though, since #to_f is implemented in terms of strtod,
and strtod can handle "NAN" in any case.

However, it seems this doesn't work in the first place in 1.8 nor 1.9:

irb(main):001:0> (0.0/0).to_s.to_f
=> 0.0
irb(main):002:0> RUBY_VERSION
=> "1.9.0"
irb(main):004:0> RUBY_RELEASE_DATE
=> "2004-06-22"

But it does work on 1.6.8:

[pbrannan@zaphod images]$ irb
irb(main):001:0> (0.0/0).to_s.to_f
=> NaN
irb(main):002:0> "NAN".to_f
=> NaN
irb(main):003:0> "nan".to_f
=> NaN
irb(main):004:0> RUBY_VERSION
=> "1.6.8"

(Also, it looks to me like in rb_cstr_to_dbl, errno is being checked
incorrectly; errno is only set if +/- HUGE_VAL is returned, otherwise it
is left to the value it was set before strtod was called.  The fix
should be to zero out errno before calling strtod).

Paul