如何从提供的参数递归定义Ruby中的Hash？

Question

This snippet of code populates an @options hash. values is an Array which contains zero or more heterogeneous items. If you invoke populate with arguments that are Hash entries, it uses the value you specify for each entry to assume a default value.

def populate(*args)
  args.each do |a|
    values = nil
    if (a.kind_of? Hash)
      # Converts {:k => "v"} to `a = :k, values = "v"`
      a, values = a.to_a.first
    end

    @options[:"#{a}"] ||= values ||= {}
  end
end

What I'd like to do is change populate such that it recursively populates @options . There is a special case: if the values it's about to populate a key with are an Array consisting entirely of (1) Symbols or (2) Hashes whose keys are Symbols (or some combination of the two), then they should be treated as subkeys rather than the values associated with that key, and the same logic used to evaluate the original populate arguments should be recursively re-applied.

That was a little hard to put into words, so I've written some test cases. Here are some test cases and the expected value of @options afterwards:

populate :a
=> @options is {:a => {}}

populate :a => 42
=> @options is {:a => 42}

populate :a, :b, :c
=> @options is {:a => {}, :b => {}, :c => {}}

populate :a, :b => "apples", :c
=> @options is {:a => {}, :b => "apples", :c => {}}

populate :a => :b
=> @options is {:a => :b}

# Because [:b] is an Array consisting entirely of Symbols or
# Hashes whose keys are Symbols, we assume that :b is a subkey
# of @options[:a], rather than the value for @options[:a].
populate :a => [:b]
=> @options is {:a => {:b => {}}}

populate :a => [:b, :c => :d]
=> @options is {:a => {:b => {}, :c => :d}}

populate :a => [:a, :b, :c]
=> @options is {:a => {:a => {}, :b => {}, :c => {}}}

populate :a => [:a, :b, "c"]
=> @options is {:a => [:a, :b, "c"]}

populate :a => [:one], :b => [:two, :three => "four"]
=> @options is {:a => :one, :b => {:two => {}, :three => "four"}}

populate :a => [:one], :b => [:two => {:four => :five}, :three => "four"]
=> @options is {:a => :one,
                :b => {
                   :two => {
                      :four => :five
                      }
                   },
                   :three => "four"
                }
               }

It is acceptable if the signature of populate needs to change to accommodate some kind of recursive version. There is no limit to the amount of nesting that could theoretically happen.

Any thoughts on how I might pull this off?

Answer 1

So here's some simple code that works.

def to_value args
  ret = {}
  # make sure we were given an array
  raise unless args.class == Array
  args.each do |arg|
    case arg
    when Symbol
      ret[arg] = {} 
    when Hash
      arg.each do |k,v|
        # make sure that all the hash keys are symbols
        raise unless k.class == Symbol
        ret[k] = to_value v 
      end           
    else    
      # make sure all the array elements are symbols or symbol-keyed hashes
      raise         
    end     
  end
  ret
rescue
  args
end
def populate *args
  @options ||= {}
  value = to_value(args)
  if value.class == Hash
    @options.merge! value
  end
end

It does deviate from your test cases:

• test case populate :a, :b => "apples", :c is a ruby syntax error. Ruby will assume the final argument to a method is a hash (when not given braces), but not a non-final one, as you assume here. The given code is a syntax error (no matter the definition of populate ) since it assumes :c is a hash key, and finds an end of line when it's looking for :c 's value. populate :a, {:b => "apples"}, :c works as expected
• test case populate :a => [:one], :b => [:two, :three => "four"] returns {:a=>{:one=>{}}, :b=>{:two=>{}, :three=>"four"}} . This is consistent with the test case populate :a => [:b] .

Answer 2

Ruby isn't Perl, => works only inside real Hash definition or as final argument in method call. Most things you want will result in a syntax error.

Are you sure that populate limited to cases supported by Ruby syntax is worth it?