Re: Bizarre Range behavior

2009/8/5 Rick DeNatale <[e-mail]>

> On Wed, Aug 5, 2009 at 12:21 PM, Yukihiro Matsumoto<[e-mail]>
> wrote:
> > Hi,
> >
> > In message "Re: Bizarre Range behavior"
> >    on Wed, 5 Aug 2009 23:29:08 +0900, "David A. Black" <
> [e-mail]> writes:
> >
> > |If you make this change, how would you then accomplish the old
> > |version? In other words, if you wanted:
> > |
> > |   "2".."19"
> > |
> > |to obey ASCII/character code logic, would there still be a way?
> >
> > The point is the current behavior is not really obeying
> > ASCII/character code logic.  It's a half-cooked magic (comparison done
> > by dictionary order, but increment done by numerical order), so if we
> > can come up with the better logic, we can override, I think.  The
> > current logic is too weird so that I believe none uses it in the real
> > code.
> >
> > Since no one is using this logic, it doesn't matter if we change it,
> > or keep it if compatibility matters most.  I prefer moving forward.
>
> Well my vote is to cook it to use the methods of the endpoints rather
> than doing something special IF the endpoints are strings and both
> happen to be numerical.  That seems as half cooked to me as the
> current situation.
>
> And if you follow the "if they're numerical" path what do you do about
> things like
>
> ("0x32".."0xFE").to_a
> ("032".."0100").to_a
> ("032".."0x32").to_a
>
> I say let strings be strings and numbers be numbers!


Quite. I think consistency is important, and this is currently broken in
1.9:

$ irb1.8
>> ('!'..']').to_a
=> ["!", "\"", "#", "$", "%", "&", "'", "(", ")", "*", "+", ",", "-", ".",
"/", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]
>> '9'.succ
=> "10"

$ irb1.9
>> ('!'..']').to_a
=> ["!", "\"", "#", "$", "%", "&", "'", "(", ")", "*", "+", ",", "-", ".",
"/", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", ":", ";", "<", "=",
">", "?", "@", "A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L",
"M", "N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z", "[",
"\\", "]"]
>> '9'.succ
=> "10"

In 1.8, String#succ is consistently used for iteration; when '9'.succ yields
'10', it spots that the endpoint is no longer reachable and ends the
iteration. 1.9 seems to embody more special cases, as '9' is followed by ':'
in the above range even though '9'.succ is still '10'.

I think treating the alphanumerics as special is useful, e.g. I like that
'9'.succ is '10' and 'z'.succ is 'aa' rather than using dumb charcode-based
sequencing, but it is only useful if it is applied consistently and
predictably.

There is obviously room for debate over the above examples: the 1.9 one
could be viewed as more complete even though it doesn't apply #succ
consistently. A lot of these odd examples are edge cases that are unlikely
to come up in production code, but personally I find the rules used by 1.8
easier to understand. If I define my own class, Range will use its #succ and
#<=> methods to carry out its logic and I expect the same for built-in
classes.

--
James Coglan
http://jcoglan.com