Hacker News Reader

Comment by bodhi | Hacker News Reader

In case you were wondering why, it is most likely* because `#join` is appending to a single string, whereas the `#reduce` call is creating intermediate strings for each step of the fold.

* I'd have to check the implementation of `#join` to be sure.

View on HN

I thought so too, but appending to a single string does even worse than reduce(:+):

  add 2  0.390000   0.000000   0.390000 (  0.390529)
  +=  2  0.540000   0.000000   0.540000 (  0.537450)

  add 3  0.530000   0.000000   0.530000 (  0.534131)
  +=  3  0.760000   0.000000   0.760000 (  0.752280)

  add 4  0.660000   0.000000   0.660000 (  0.668154)
  +=  4  0.960000   0.000000   0.960000 (  0.954727)

Code: https://gist.github.com/1562994

+= is not append

  str += append_str

is the equivalent to:

  str.dup << append_str

That is, it copies the string and then appends to the copy. Benchmarking the speed of a true append is difficult because in order to preserve the original string in the benchmark you must dup it anyhow (done here outside of the bench.report block). However, the bigger the strings get, the more pronounced the advantage of appending is.

                user     system      total        real
  add  2      0.260000   0.000000   0.260000 (  0.263414)
  join 2      0.320000   0.000000   0.320000 (  0.325341)
  append 2    0.230000   0.010000   0.240000 (  0.235669)

  add  3      0.500000   0.000000   0.500000 (  0.497219)
  join 3      0.840000   0.020000   0.860000 (  0.866401)
  append 3    0.260000   0.010000   0.270000 (  0.268676)

  add  4      0.360000   0.040000   0.400000 (  0.397778)
  join 4      0.970000   0.030000   1.000000 (  0.997321)
  append 4    0.280000   0.000000   0.280000 (  0.281020)

  add  5      0.460000   0.000000   0.460000 (  0.454780)
  join 5      0.910000   0.030000   0.940000 (  0.946600)
  append 5    0.330000   0.000000   0.330000 (  0.336214)

Code: https://gist.github.com/1563290

Reply on news.ycombinator.com