Hash#symbolize_keys! performance

Updated (http://gist.github.com/297772) to handle failures in the following test cases :

• test_symbolize_keys_preserves_keys_that_cant_be_symbolized
• test_symbolize_keys_preserves_fixnum_keys

methodmissing:~ lourens$ ruby hash_bench.rb
Rehearsal -----------------------------------------------------------------------
Hash#orig_symbolize_keys! , nil key 3.500000 0.160000 3.660000 ( 3.686203)
Hash#symbolize_keys! , nil key 0.510000 0.000000 0.510000 ( 0.506390)
Hash#orig_symbolize_keys! 2.000000 0.090000 2.090000 ( 2.096217)
Hash#symbolize_keys! 0.470000 0.000000 0.470000 ( 0.468481)
-------------------------------------------------------------- total: 6.730000sec

                                      user     system      total        real

Hash#orig_symbolize_keys! , nil key 3.520000 0.160000 3.680000 ( 3.753897)
Hash#symbolize_keys! , nil key 0.510000 0.000000 0.510000 ( 0.513103)
Hash#orig_symbolize_keys! 2.020000 0.090000 2.110000 ( 2.115938)
Hash#symbolize_keys! 0.470000 0.000000 0.470000 ( 0.481901)

Could you please show the data that generates those dubious benchmarks ?
For the (typical?) case where the keys are all strings to be symbolized, and not Fixnums, the original is about 25% faster for my tests (which do not factor out the overhead of duplicating the hash for each iteration of the destructive test.)

Hey,

Thanks for spotting this and apologies for the oversight.

I updated the gist in the original post ( http://gist.github.com/297755 ) with string hash cases and your spot on.

I'll ping Jeremy to revert in favor of the common case.

• Lourens