I'd expect the following to produce output like this: >>> a [{:type=>"a"}, {:type=>"a"}] >>> b [{:type=>"b"}] Instead, I get: >>> a [{:type=>"b"}, {:type=>"a"}, {:type=>"a"}] >>> b [{:type=>"b"}, {:type=>"a"}, {:type=>"a"}] W

unrolling-wtf.rb

before = {
  "key1" => {:type => "a"},
  "key2" => {:type => "a"},
  "key3" => {:type => "b"},
}

after = Hash.new([])
before.each{|k,v| after[v[:type]] <<= v}

after.each{|k,v| puts ">>> #{k} #{v.inspect}"}

# I expect output like the following:
# >>> a [{:type=>"a"}, {:type=>"a"}]
# >>> b [{:type=>"b"}]

# That is, all values in each key of 'after' should have :type matching the key.

# Instead, I get:
# >>> a [{:type=>"b"}, {:type=>"a"}, {:type=>"a"}]
# >>> b [{:type=>"b"}, {:type=>"a"}, {:type=>"a"}]

The following Python does what I want; I still don't see why the Ruby doesn't...

before = {
    "key1": {"type": "a"},
    "key2": {"type": "a"},
    "key3": {"type": "b"},
}

after = {}
for k, v in before.items():
    after.setdefault(v["type"], []).append(v)

for k, v in after.items():
    print ">>> %s %s" % (k, v)

Sigh. This version of the Ruby works. It's more explicit (and less sugary), but I still don't see why my first crack at it fails.

before = {
  "key1" => {:type => "a"},
  "key2" => {:type => "a"},
  "key3" => {:type => "b"},
}

after = {}
for k, v in before
  type = v[:type]
  after[type] || after[type] = []
  after[type] << v
end