[Haskell-cafe] lengthOP rewrite rules

[Ryan Ingram]

2008/12/18 Luke Palmer <lrpalmer at gmail.com>:
> On Thu, Dec 18, 2008 at 1:53 AM, Cetin Sert <cetin.sert at gmail.com> wrote:
>>
>> Hi,
>>
>> I tested the following, why does the rewrite rules not fire when using
>> tuples also in testRewrite2, testRewriteReverse2?
>
>> testRewrite2 :: a → (Bool,Bool)
>> testRewrite2 x = (pf,pi)
>>   where
>>     f  = replicate 2000 x
>>     i  = repeat         x
>>     lf = length f
>>     li = length i
>>     pf = lf > 300
>>     pi = li > 300
>>
>
> Why would you expect it to?  The compiler is free to inline lf and li to
> discover that the rule applies, but it is also free not to.  Applying all
> applicable rules while maintaining the ability to abstract is undecidable
> (big surprise). Thus the dependency on compiler cleverness I mentioned...

I'm agreeing with Luke here.  It's possible that the compiler decided
to inline f and i, and length, and determined that lf == 2000 and li
== _|_
Or it could have decided not to inline at all.  Or some other possibility.

If you specify {-# INLINE lf #-}, do the results change?  I suspect they might.

  -- ryan