Replace murmur hash with the fastest hash

[blackswanny]

Here is another option. It is way faster than murmur and even murmur ver 2, which is used in another PR. The source code is at https://github.com/darkskyapp/string-hash
The original work is at http://www.cse.yorku.ca/%7Eoz/hash.html

[blackswanny]

I have yarn tests failing. Do you use screen snapshots to compare?
Should I run yarn run coverage -- -u? In this case all tests are passed.
I don't know how to debug this kind of tests.

Snapshot Summary
 › 206 snapshot tests failed in 26 test suites. Inspect your code changes or run with `yarn run coverage -- -u` to update them.

Test Suites: 26 failed, 16 passed, 42 total
Tests:       184 failed, 10 skipped, 487 passed, 681 total
Snapshots:   206 failed, 491 passed, 697 total
Time:        77.255s

[tkh44]

You are going to have massive snapshot failures because of the change in hash structure and algorithm. All the stylesheet snapshots will need to be updated.

try this command

jest --watch --no-cache

[tkh44]

Have you tested these 2 hashes side by side with large strings before doing all of this though? The performance difference might not warrant that much work.

[blackswanny]

I made a test. it's two strings with 10000 chars
According to this one, there is no difference, and they are equal. So it's either i did't test well or results we see for our project do not directly relate to hash itself, but to the way how hash is used and cached , for instance.
https://jsperf.com/csshash/1

[emmatown]

The benchmarks I showed in #544 had this hash function as string-hash, it was slower than the hashmumur2 function in #544 but faster than the current hashmumur2 function in Chrome. It was slower than both the hashmumur2 functions in Firefox and it was faster than both the hashmumur2 functions in Safari. I think that the hashmumur2 function in #544 is the best choice since it's faster in general and it's the same algorithm so if people have snapshots with hashes, they won't change.

[blackswanny]

@mitchellhamilton @tkh44 tests seems controversial. We have no exact winner. Once again my results on large strings.
Firefox - looser (2 times)
Chrome - even
Safari - winner (5 times)
We need to investigate deeper issue with hashes. Maybe it's not related with hash itself. However, on glamor in Chrome (as pretty common dev browser) old hash is 4 times slower, whereas according to these tests it should be even

https://jsperf.com/murmur-vs-djb2