fixed long cachekeys reported on issue #380

[trein]

No description provided.

[trein]

Ok. Locally all tests are passing, but they fail on Travis. Do you have any ideas?

[hanshasselberg]

Ruby 1.8.7 doesn't have the method base64digest: http://www.ruby-doc.org/stdlib-1.8.7/libdoc/digest/rdoc/Digest/Class.html.

[trein]

Good catch! I haven't noticed you guys are using Ruby 1.8.7. I will propose a new way to hash the cache_key then.

[Zapotek]

Zlib.crc32 worked for me, fast and small, and zlib is already required by that file.

[Zapotek]

@i0rek Just saw your remark on CRC32 as a hashing function on #380, is it about the collisions?
Also, unless you're worried about persistency, you can just use #hash and force it #to_s.

If none of these are good enough, you can use MurmurHash (which is what Ruby uses for strings) which is again fast and small.

[hanshasselberg]

@Zapotek I recently stumbled over this: http://stackoverflow.com/questions/996843/when-is-crc-more-appropriate-to-use-than-md5-sha1.

[Zapotek]

Hm, fair enough. Better safe than sorry then, there's also a MurmurHash3 implementation available as a gem which is faster than MurmurHash 1 or 2: https://github.com/funny-falcon/murmurhash3-ruby

I think I'll switch my project to MurmurHash3 too, heh. Thanks for the info man.

[hanshasselberg]

at least for Typhoeus I would prefer a stdlib solution...

[trein]

Hi guys.

Sorry for the delay. I've switched the hashing method from SHA1.base64digest to SHA1.hexdigest, which is available on ruby 1.8.7 stdlib. Let me know your thoughts.

[hanshasselberg]

Great! Since cache_key and hash are very similar now, could you alias one to another?
Btw I looked up the issue where the long cache keys were discussed: #345. The reason it was done that way is that Dalli can deal with long cache keys. Since other libs don't do that I think it is appropriate to switch to SHA1.

[trein]

Hi @i0rek,

Probably I haven't understood properly, but I'm currently calling cache_key from hash. Is that what you meant by alias?

def hash
    Zlib.crc32 cache_key
end

The only difference is that I'm encoding it using CRC as the return shouldn't be a string.

[hanshasselberg]

@trein I am sorry. That was confusing. Thank you for your work. I will merge and make the tiny change myself. Should have done that in the first place instead of bothering you.

[hanshasselberg]

I was talking without thinking. hash returns an int...
Thanks again for your work.