Performance improvements

[aectann]

Re-use hmac object instead of creating a new one from scratch on every generate() call.

40% shorter running time for generate() function with thread_safe set to False, or 30% running time improvement when thread_safe parameter is true.

Code used to measure running time:

from libthumbor import CryptoURL
import cProfile


def generate():
    base_url = 'http://example.com/images{}.jpeg'
    crypto = CryptoURL(key='example_key_here')
    for i in xrange(200001):
        crypto.generate(image_url=base_url.format(i))

if __name__ == '__main__':
    cProfile.run('generate()', sort='cumulative')

Before the change:

   10400079 function calls in 4.910 seconds

   Ordered by: cumulative time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    4.910    4.910 <string>:1(<module>)
        1    0.136    0.136    4.910    4.910 load_test.py:5(generate)
   200001    0.179    0.000    4.701    0.000 crypto.py:62(generate)
   200001    0.610    0.000    4.480    0.000 crypto.py:54(generate_new)

After, thread_safe=True:

   8400061 function calls in 3.483 seconds

   Ordered by: cumulative time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    3.483    3.483 <string>:1(<module>)
        1    0.153    0.153    3.483    3.483 load_test.py:11(generate)
   200001    0.220    0.000    3.225    0.000 crypto.py:72(generate)
   200001    0.430    0.000    2.966    0.000 crypto.py:62(generate_new)

After, thread_safe=False:

   7600057 function calls in 2.985 seconds

   Ordered by: cumulative time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    2.985    2.985 <string>:1(<module>)
        1    0.151    0.151    2.985    2.985 load_test.py:11(generate)
   200001    0.167    0.000    2.732    0.000 crypto.py:72(generate)
   200001    0.355    0.000    2.525    0.000 crypto.py:62(generate_new)

[guilhermef]

You still need to convert the key to bytes.

[aectann]

Good catch, addressed that in ea4843a.

[guilhermef]

Thanks @aectann, that's an awesome change.
I'll release a new version.