I’m going to cross-post this from the Rails group, because some of the
people here are Ruby ninjas and don’t read that forum, and I’d like help
getting this function optimized… its results are useful in a Rails
perspective, but it’s functionality has nothing to do with Rails at all.
class Hash
def to_params(parent = ‘’)
ret = ‘’
self.keys.each do |key|
if self[key].is_a? Hash
if parent == ‘’
ret += self[key].to_uri(key.to_s)
else
ret += self[key].to_uri(parent + “[#{key.to_s}]”)
end
else
if parent == ‘’
ret += “#{key}=#{self[key]}&”
else
ret += “#{parent}[#{key}]=#{self[key]}&”
end
end
end
return ret.chomp(’&’)
end
end
Anybody got any optimizations(either quicker speed, or less text and
comparable speed) for that one?
Thanks.
ret += self[key].to_uri(key.to_s)
else
ret += self[key].to_uri(parent + "[#{key.to_s}]")
s/to_uri/to_params
Originally wrote it with a different function name.
On Feb 8, 12:42 pm, Luke I. [email protected] wrote:
if parent == ''
end
end
return ret.chomp('&')
end
end
I realize the code is there, but I’m having trouble figuring out what
it’s used for. Could you describe what ‘parent’ is, and what it means
to have nested hashes?
One guess I have at improving the speed it to change:
self.keys.each{ |key|
# repeated function calls needed for self[key]
}
to
self.each_pair{ |key,value|
# direct references to value
}
On Feb 8, 1:03 pm, Luke I. [email protected] wrote:
For {:test => ‘testing’} the return should be test=testing
For {:test => {:test => ‘testing’}} it should be test[test]=testing
Any nested hashes past that point should just continue to be added on as
an array reference:
{:test => {:test => {:test => ‘testing’}}}
=> test[test][test] = testing
So, the parent attribute is really just for the recursive calls?
Is the output of the code correct for these cases:
puts( { :foo=> ‘bar’, :jim=>‘jam time’ }.to_params )
#=> jim=jam time&foo=bar
puts( { :foo => { :bar=>1, :jim=>‘jam’ } }.to_params )
#=> foo[jim]=jam&foo[bar]=1
puts( { :foo =>
{ :bar=>{ :jim=>‘jam’, :jar=>‘jib’ }, :jim=>‘jam’ } }.to_params )
#=> foo[jim]=jam&foo[bar][jim]=jam&foo[bar][jar]=jib
(Note the unencoded space in the first example.)
I wrote another function that basically does the exact same thing, only
tries to do it in as short a space as possible: here’s what I got (and
also benchmark times for the two of them)
I also tried splitting the original function into two seperate functions
so it didn’t have to do as many comparisons.
require ‘benchmark’
class Hash
#to avoid spamming more than necessary, see original email in thread
for definition of to_params
def to_params2(parent = ‘’)
self.keys.inject(’’) do |k, v|
(self[v].is_a? Hash) ?
(parent == ‘’ ? k += self[v].to_params2(v.to_s) : k +=
self[v].to_params2(parent + “[#{v.to_s}]”)) :
(parent == ‘’ ? k += “#{v}=#{self[v]}&” : k +=
“#{parent}[#{v}]=#{self[v]}&”)
end
end
def to_params3()
ret = ‘’
self.each_pair do |key, value|
if value.is_a? Hash
ret += value.to_params3_with_parent(key.to_s)
else
ret += “#{key}=#{value}&”
end
end
return ret.chomp(’&’)
end
def to_params3_with_parent(parent)
ret = ‘’
self.each_pair do |key, value|
if value.is_a? Hash
ret += value.to_params3_with_parent(parent + “[#{key.to_s}]”)
else
ret += “#{parent}[#{key}]=#{value}&”
end
end
return ret.chomp(’&’)
end
end
n = 100000
h = {:user => {:subuser => {:name => ‘test’}, :name => ‘test2’}, :name
=> ‘test3’}
Benchmark.bm do |x|
x.report { n.times do; h.to_params; end }
x.report { n.times do; h.to_params2; end }
x.report { n.times do; h.to_params3; end }
end
Here’s the results:
user system total real
5.132000 0.000000 5.132000 ( 5.174000) #original
5.803000 0.000000 5.803000 ( 5.844000) #with inject
4.868000 0.000000 4.868000 ( 4.880000) #split into two functions
Can anyone do better than the last one?
Is the output of the code correct for these cases:
Yes.
Noting the unencoded space: not a big deal. I can encode the whole
string later, or throw in an encode in the parameterization process, but
that’s not relevant to just optimizing the base functionality.
On 08.02.2007 20:42, Luke I. wrote:
if parent == ''
end
end
return ret.chomp('&')
end
end
Anybody got any optimizations(either quicker speed, or less text and
comparable speed) for that one?
Use << instead of += in all places.
replace the line
self.keys.each do |key|
with
each do |key, value|
Replace “self[key]” with “value” then.
Maybe change the major if then else with case when end in order to more
easily adjust to special treatment of other types than Hashes.
If you need more efficiency improvements, extract the “if parent=’’”
from the loop, make it a top level decision and have two iterations (if
and else branch).
And, make the string / stream to append to a parameter. That way you
don’t need to create potentially large strings during recursion before
you append them but you can directly append - you basically just have
one.
Typing left as an exercise for the reader.
Kind regards
robert
On 08.02.2007 21:43, Robert K. wrote:
And, make the string / stream to append to a parameter. That way you
don’t need to create potentially large strings during recursion before
you append them but you can directly append - you basically just have one.
PS: Forgot to mention that the last one might be one of the improvements
that bring most benefits together with using <<. The pattern is
def meth(out = ‘’)
…
…
# recursion
another.meth(out)
…
out
end
Have fun!
robert
And, make the string / stream to append to a parameter. That way you
don’t need to create potentially large strings during recursion before
you append them but you can directly append - you basically just have
one.
This one was great: using two functions as I did in one of the earlier
emails, however, still provides a significant speed bump over a
top-level if-else decision on parent==’’.
I changed the two-function thing to pass the returned string as a param,
and re-wrote the original function using exactly your suggestions,
giving these benchmarks:
robert 4.414000 0.000000 4.414000 ( 4.461000)
two-fun 4.088000 0.000000 4.088000 ( 4.097000)
Luke I. wrote:
And, make the string / stream to append to a parameter. That way you
don’t need to create potentially large strings during recursion before
you append them but you can directly append - you basically just have
one.
This one was great: using two functions as I did in one of the earlier
emails, however, still provides a significant speed bump over a
top-level if-else decision on parent==’’.
I changed the two-function thing to pass the returned string as a param,
and re-wrote the original function using exactly your suggestions,
giving these benchmarks:
robert 4.414000 0.000000 4.414000 ( 4.461000)
two-fun 4.088000 0.000000 4.088000 ( 4.097000)
Oh, and I used str.concat: should I have used something else?
Luke I. wrote:
Here’s the results:
user system total real
5.132000 0.000000 5.132000 ( 5.174000) #original
5.803000 0.000000 5.803000 ( 5.844000) #with inject
4.868000 0.000000 4.868000 ( 4.880000) #split into two functions
Can anyone do better than the last one?
That wasn’t fair at all (you know i couldn’t resist, do you?)
def to_params4()
result = ‘’
stack = []
each do |key, value|
Hash === value ? stack << [key, value] : result <<
“#{key}=#{value}&”
end
stack.each do |parent, hash|
hash.each do |key, value|
if Hash === value
stack << ["#{parent}[#{key}]", value]
else
result << “#{parent}[#{key}]=#{value}&”
end
end
end
result.chop
end
as you can see i unrolled the recursion, the benefit isn’t that
high but it was fun to code.
(and yes, i know i shouldn’t modify an array i’m iterating over,
but hey it seems to work (appending seems to be fine))
cheers
Simon
On 08.02.2007 22:11, Luke I. wrote:
Oh, and I used str.concat: should I have used something else?
I’d use << - in that case you can also use StringIO and IO objects (i.e.
files). I don’t think it makes a performance difference.
Kind regards
robert