Re: Pushing the `gnus-range-*' functions down into the C layer

[Stefan Monnier]

>> #define EQ(x, y) (XHASH (x) == XHASH (y))

Not only that, but also

   #define NILP(x)  EQ (x, Qnil)

So EQ is really used extensively in C and slowing it down is likely to
make a perceptible difference.   

>> /* Return a perfect hash of the Lisp_Object representation.  */
>> #define XHASH(a) (a)

> this case I agree, it should not be called XHASH because that implies it
> is used for structural equivalence, which is misleading. Also, hashing
> object should behave like deep copying - follow pointers.

Nothing says that hashing an object should behave like deep copying.
That's why make-hash-table has a :test parameter which can be `eq' or
`equal'.
As for why we have XHASH: because there's another definition of it for
when we use a union rather than an int for Lisp_Object:

    #define XHASH(a) ((a).i)

    
-- Stefan