Clojure

clj-1036-hasheq-for-biginteger-patch-v1.txt dated Sep 26 2012 makes BigInteger's return equal hash values for values considered equal by =.

It does the same for Float and Double types, which before returned different hash values for values considered equal by =

I went ahead and changed the :added metadata on bigint and biginteger, although I can see that without my change, the person who did that may have meant for the :added to go with the behavior of the function, not with the name. Paul's suggested change that I have in the patch is for the :added metadata to go with the name, not the function behavior. It is easy to remove that part of the patch if that change is not desired.

You can't just consider only the lower long of bigints. Also, what's the rationale for the float stuff?

clj-1036-hasheq-for-biginteger-patch-v2.txt dated Nov 13 2012 is identical to clj-1036-hasheq-for-biginteger-patch-v1.txt except that it addresses Rich's comment that for BigInt's and BigInteger values that don't fit in a long, their entire value must be hashed.

The rationale for the changes to hasheq for Float and Double types is the same as the rationale for the change for BigInteger: without that change, Float and Double types that are = can have different hasheq values.

Although you are correct that Double and Float are =, but have different hashes:

user=> (apply = [(Double. -1.0) (Float. -1.0)])
true
user=> (map hash [(Double. -1.0) (Float. -1.0)])
(-1074790400 -1082130432)

I could not get the same errant behavior out of PHM:

user=> (assoc clojure.lang.PersistentHashMap/EMPTY (Float. -1.0) :oops! (Double. -1.0) :one)
{-1.0 :one}

I haven't taken the time to investigate exactly what is happening here, but either way I think this ticket is very specifically about BigInteger and the Float/Double issue could be explored in another ticket.

I can open another ticket for the Float/Double issue if that is what people would prefer.

I think what is happening in the test case you give, Paul, is that the hash values for (Float. -1.0) and (Double. -1.0) happen to be the same in their least significant 20 bits, and PHM isn't using the upper bits where the hash values differ.

Clojure 1.5.0-beta without patch:
user=> (map #(format "%x" %) (map hash [(Float. -1.0) (Double. -1.0)]))
("bf800000" "bff00000")

There are other Float/Double values where this lucky accident doesn't happen, e.g.

Clojure 1.5.0-beta1 without patch:

user=> (= (Float. 1e9) (Double. 1e9))
true
user=> (map hash [(Float. 1e9) (Double. 1e9)])
(1315859240 1104006501)
user=> (assoc clojure.lang.PersistentHashMap/EMPTY (Float. 1e9) :oops! (Double. 1e9) :one)

With 1.5.0-beta1 plus patch clj-1036-hasheq-for-biginteger-patch-v2.txt:

user=> (= (Float. 1e9) (Double. 1e9))
true
user=> (map hash [(Float. 1e9) (Double. 1e9)])
(1315859240 1315859240)
user=> (assoc clojure.lang.PersistentHashMap/EMPTY (Float. 1e9) :oops! (Double. 1e9) :one)

Presumptuously changing status from Not Approved to Vetted, since patch clj-1036-hasheq-for-biginteger-patch-v2.txt should address the reasons that Rich marked the previous patch as Not Approved. Changing it to Vetted on the assumption that if Stuart Halloway marked the previous patch as Screened, the ticket itself is good enough to be Vetted.

Patches and tickets need to be better than this. Talks about BigInteger, changes hash for doubles. Lists problem but not approach, need to trawl through comments and code to see what's going on, etc.