The primitive and the primitive failure code should behave the same?
On 12/18/16 2:13 , stepharong wrote:
Sorry but how can we improve the situation concretely
Stef
On Wed, 14 Dec 2016 00:22:47 +0100, Andres Valloud
<avall...@smalltalk.comcastbiz.net> wrote:
Oh, ok... then the failure code isn't doing what was intended :).
On 12/13/16 15:16 , Martin McClure wrote:
On 12/13/2016 08:39 AM, Andres Valloud wrote:
I would have expected the floats to be a byte object of size 8. Why is
this conversion needed? Is somehow the primitive thinking the float
has
size 2? Or is the primitive hashing 32 bits at a time? The prim used
to be a C version of the byte oriented 1644525 multiplicative hash, did
this change?
The float is not a byte object of size 8, it's a word object of size 2.
When you feed it to #hashBytes:startingWith: the primitive fails. (I
didn't look at why -- maybe because it isn't a byte object?)
The alternative Smalltalk code treats it like it was a byte object, even
though it's not. So it iterates through the two words, multiplying each
by 1664525, then keeping only the low-order 28 bits of the result. But
since each word is a full 32 bits, the high-order four bits of each
word, which include the sign bit, can never affect the hash.
Regards,
-Martin
