remove residual hokey hash functions

[nikomatsakis]

Our current hashtables just take the hash modulo number of buckets to find the bucket. There is a good case to be made that multiplying by a suitable magic number, such as the golden ratio, will help spread the entropy around, particularly for poor hash functions. We should try this out and measure the results.

[dgryski]

However, if we're using (a keyed!) siphash as our base, we probably have about as good entropy as we're going to get. The magic multiplication seems to have more of an effect with poorly chosen hash functions -- the standard h = h*37 + c kind.

[nikomatsakis]

@dgryski yes. right now, of course, rustc at least uses some hokey hash fns everywhere.

[dgryski]

Were those not fixed as part of @graydon's changes to vec.rs and str.rs in fada46c ?

[mignon-p]

No, there are still lots of hokey hash functions in rustc. (Mostly along the theme of combining two or more integers via addition and shifting.) I made a list of them:

https://gist.github.com/3330996

(Since github's bug tracker doesn't seem to have a way to do attachments like I would in Bugzilla, I'm assuming gists are the answer for attachments.)

But I agree that the right approach would be to fix the hash functions, rather than try to compensate by doing mixing in the hashtable code.

[graydon]

Yikes. I'll remove these.