Method: Enumerable#sort_by

Defined in:

enum.c

#sort_by {|obj| ... } ⇒ Array #sort_by ⇒ Object

Sorts enum using a set of keys generated by mapping the values in enum through the given block.

If no block is given, an enumerator is returned instead.

%w{apple pear fig}.sort_by { |word| word.length}
              #=> ["fig", "pear", "apple"]

The current implementation of sort_by generates an array of tuples containing the original collection element and the mapped value. This makes sort_by fairly expensive when the keysets are simple.

require 'benchmark'

a = (1..100000).map { rand(100000) }

Benchmark.bm(10) do |b|
  b.report("Sort")    { a.sort }
  b.report("Sort by") { a.sort_by { |a| a } }
end

produces:

user     system      total        real
Sort        0.180000   0.000000   0.180000 (  0.175469)
Sort by     1.980000   0.040000   2.020000 (  2.013586)

However, consider the case where comparing the keys is a non-trivial operation. The following code sorts some files on modification time using the basic sort method.

files = Dir["*"]
sorted = files.sort { |a, b| File.new(a).mtime <=> File.new(b).mtime }
sorted   #=> ["mon", "tues", "wed", "thurs"]

This sort is inefficient: it generates two new File objects during every comparison. A slightly better technique is to use the Kernel#test method to generate the modification times directly.

files = Dir["*"]
sorted = files.sort { |a, b|
  test(?M, a) <=> test(?M, b)
}
sorted   #=> ["mon", "tues", "wed", "thurs"]

This still generates many unnecessary Time objects. A more efficient technique is to cache the sort keys (modification times in this case) before the sort. Perl users often call this approach a Schwartzian Transform, after Randal Schwartz. We construct a temporary array, where each element is an array containing our sort key along with the filename. We sort this array, and then extract the filename from the result.

sorted = Dir["*"].collect { |f|
   [test(?M, f), f]
}.sort.collect { |f| f[1] }
sorted   #=> ["mon", "tues", "wed", "thurs"]

This is exactly what sort_by does internally.

sorted = Dir["*"].sort_by { |f| test(?M, f) }
sorted   #=> ["mon", "tues", "wed", "thurs"]

Overloads:

• #sort_by {|obj| ... } ⇒ Array

  Yields:

  • (obj)

  Returns:

  • (Array)

# File 'enum.c', line 956

static VALUE
enum_sort_by(VALUE obj)
{
    VALUE ary, buf;
    NODE *memo;
    long i;
    struct sort_by_data *data;

    RETURN_SIZED_ENUMERATOR(obj, 0, 0, enum_size);

    if (RB_TYPE_P(obj, T_ARRAY) && RARRAY_LEN(obj) <= LONG_MAX/2) {
  ary = rb_ary_new2(RARRAY_LEN(obj)*2);
    }
    else {
  ary = rb_ary_new();
    }
    RBASIC_CLEAR_CLASS(ary);
    buf = rb_ary_tmp_new(SORT_BY_BUFSIZE*2);
    rb_ary_store(buf, SORT_BY_BUFSIZE*2-1, Qnil);
    memo = NEW_MEMO(0, 0, 0);
    OBJ_INFECT(memo, obj);
    data = (struct sort_by_data *)&memo->u1;
    data->ary = ary;
    data->buf = buf;
    data->n = 0;
    rb_block_call(obj, id_each, 0, 0, sort_by_i, (VALUE)memo);
    ary = data->ary;
    buf = data->buf;
    if (data->n) {
  rb_ary_resize(buf, data->n*2);
  rb_ary_concat(ary, buf);
    }
    if (RARRAY_LEN(ary) > 2) {
  RARRAY_PTR_USE(ary, ptr,
          ruby_qsort(ptr, RARRAY_LEN(ary)/2, 2*sizeof(VALUE),
         sort_by_cmp, (void *)ary));
    }
    if (RBASIC(ary)->klass) {
  rb_raise(rb_eRuntimeError, "sort_by reentered");
    }
    for (i=1; i<RARRAY_LEN(ary); i+=2) {
  RARRAY_ASET(ary, i/2, RARRAY_AREF(ary, i));
    }
    rb_ary_resize(ary, RARRAY_LEN(ary)/2);
    RBASIC_SET_CLASS_RAW(ary, rb_cArray);
    OBJ_INFECT(ary, memo);

    return ary;
}