-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 20.04.2015 20:02, Francis Avila wrote: > There's no contract, but strings, keywords, and symbols should hash > the same, and collections of these (vectors, lists, maps, sets) > should hash the same. > > It's difficult to hash numbers the same between Clojure and > Clojurescript. > > Clojurescript numbers are all doubles (because JS), so they should > in theory hash the same as Clojure doubles. Clojure hashes doubles > using Java's Double.hashCode(), which relies on knowing the exact > bits of the double. These bits are not available in Javascript (at > least not easily or without using typedarrays). I'm also not sure > if this particular implementation of hashCode is part of the Java > spec (i.e. implemented the same by all JDKs and JVMs.) > > Of course in practice the same clojure form (as read) will not hash > the same in clj and cljs because clj uses longs most of the time. > > You could take a hybrid approach where integers in cljs are hashed > like longs in Clojure. This works up to 52 bits, but longs with > more bits than that are not representable in Clojurescript. This is > an approach I was pursuing in my murmur3 hashing implementation for > cljs: > > http://dev.clojure.org/jira/browse/CLJS-754 > https://github.com/favila/clojurescript/blob/murmur3/src/cljs/cljs/core.cljs#L1111 > > https://github.com/favila/clojurescript/blob/murmur3/src/cljs/cljs/murmur3.cljs#L61 > > > In practice you will hash the same most of the time if you most > deal with integer numbers, but any doubles, bigdecimals, or large > integers will still hash differently. > > This is what happens in clojurescript now: > > (js-mod (Math/floor o) 2147483647)
Hi, I am the author of hasch (1) for cryptographic cross-platform hashing, which Herwig mentioned. The numeric types were a problem for me, too, as Francis describes. I guess that all the corner cases are difficult to catch, so I decided to treat all numbers like doubles in edn data. Still this is fragile due to floating point arithmetic differences to JVM integer arithmetic... JavaScript is really bad for numerical tasks sadly and I don't see a lightweight work-around. So once you numerically calculate the same thing on both runtimes and expect the hashed results to be the same, there could be trouble. Additionally there is no Character type in JavaScript, so you have to treat them like Strings as well. Normally different types with the same content should hash differently, except for seqs and vectors, to my current understanding. My implementation is just a recursive hashing scheme protocol, which allows to swap the hash function, so if you want something more lightweight than sha512, you could also use a cross-platform murmur implementation with hasch maybe. Maps and Sets are hashed elementwise and XORed afterwards, which might cause performance problems in your case. I haven't found a cheaper way to ensure against malicious collisions. Cheers, Christian (1) https://github.com/ghubber/hasch -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iQEcBAEBAgAGBQJVNh5tAAoJEKel+aujRZMk7mcH/05CV5ACKIZDJTEE7X38xG98 Sgxqc81JNKH5NFgcrkTfjqV2SlGROt7mxt3pVUEuLoiyE5U3CzeHBge0wpiBVmsB eougHX43a5EL2qJWYhJTlORBXidxiT5uWdonyKH7AVNcURcM0I6P0BH92NPgq4hE q8vQLdEJeqWoeHllLHb+te2sfmSArEESgNSMStgDZQF+J7ODJCKzako49UCtJxg1 FVjsHMK2+d50X9yTw4NP644PKssH3GdPBiqmyr0cx20jtpvn5x+q+xU5XaYbA0L4 Uk1FT38L4FoSRVzBMZfb0FsmF7WxdQOgKuQc3tVeLgJhcWSOjJ2js5Sd8gwMQgM= =lywe -----END PGP SIGNATURE----- -- Note that posts from new members are moderated - please be patient with your first post. --- You received this message because you are subscribed to the Google Groups "ClojureScript" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/clojurescript.
