make (a < b < c) illegal?

[Derek Parnell]

On Wed, 07 Feb 2007 19:02:06 -0800, Andrei Alexandrescu (See Website For
Email) wrote:

> Derek Parnell wrote:
>> On Wed, 07 Feb 2007 17:09:35 -0800, Andrei Alexandrescu (See Website For
>> Email) wrote:
>> 
>>> Derek Parnell wrote:
>>>> On Wed, 07 Feb 2007 16:55:15 -0800, Walter Bright wrote:
>>>>
>>>>> Right now, in D (as well as C and C++), when you see the expression:
>>>>>
>>>>> 	if (a < b < c)
>>>>>
>>>>> what is your first thought? Mine is that it was written by a newbie who 
>>>>> didn't realize that (a < b) returns true or false, and that it does NOT 
>>>>> mean ((a < b) && (b < c)). The odds approach certainty that this is a 
>>>>> logic error in the code, and even if it was intentional, it raises such 
>>>>> a red flag that it shouldn't be used anyway.
>>>>>
>>>>> Andrei has proposed (and I agreed) that this should be done away with in 
>>>>> the language, i.e. comparison operators should no longer be associative. 
>>>>>   It's a simple change to the grammar. If one really did want to write 
>>>>> such code, it could be done with parentheses:
>>>>>
>>>>> 	if ((a < b) < c)
>>>>>
>>>>> to get the original behavior. At least, that looks intentional.
>>>>>
>>>>> I don't think this will break existing code that isn't already broken.
>>>> First thought: Yes, your proposed change makes sense.
>>>>
>>>> Second thought: Why not make it do what the coder is wanting it to do?
>>>>    Namely, make the idiom:
>>>>
>>>>      expression1 relopA expression2 relopB expression3
>>>>
>>>>    translate as
>>>>
>>>>      ( auto temp = expression2,
>>>>       (expression1 relopA temp) && (temp relopB expression3) )
>>> What's the intended meaning of:
>>>
>>> a < b == c < d
>> 
>> Well that *obvious*!
>> 
>>  (((a < b) && (b == c)) < d)  <G>
>> 
>> Okay, okay, I see your point. But it would be useful (one day) to easily
>> code the idiom  (a op b) && (b op c),  no? 
>> 
>> How about someone knocking up a mixin template for expressions of this
>> format? I haven't got a clue how it could be done as the
>> mixin/template/meta-programming syntax and semantics of D is still so
>> obtuse and confusing to me that I can only do the very simplest things and
>> then only after many false starts.
> 
> It's very easy to implement as a straight template function:
> 
> ordered(a, b, c)
> 
> returns true iff a <= b <= c, and
> 
> strictly_ordered(a, b, c)
> 
> returns true iff a < b < c. The functions can be nicely accommodate 
> multiple arguments. Other functions can define STL-style in-range.

Right ... 

class Foo { . . . }
class Bar { . . . }

Foo f = new Foo;
Foo g = new Foo;
Bar b = new Bar;

if (ordered( f,b,g ) ) ...

BANG! The template doesn't know about Foo and Bar, so there is no match. In
fact, just getting a template for built-in types requires a few thousand
combinations of a,b,c signatures and aliases to make it palatable.

That is why I thought mixin-expressions might be more useful in this case -
working at the textual level rather than the template-matching level. But I
do know how to do it.

-- 
Derek
(skype: derek.j.parnell)
Melbourne, Australia
"Justice for David Hicks!"
8/02/2007 2:39:20 PM