Ordered query_values

[sporkmonger]

Ordered query_values

[sporkmonger]

Yeah, I strongly agree. A fix will probably come in the next major release though, because my solution probably will be to change the return value to:

[['a', '1'], ['a', '2']]

This format both preserves order as well as allows for handling query strings with multiple occurrences of a single key as above. But because it's a breaking change, it probably won't land until a major version release. I might make it available under a branch earlier though.

[sporkmonger]

Actually, bvandenbos just sent me a pull request that resolves this.

[sporkmonger]

Note that his patch requires you to explicitly indicate you want a :flat_array. In the future this will probably become the only format the method returns, but for now you can indicate you want it as an alternative notation.

[igrigorik]

Awesome - thanks!

[igrigorik]

Hmm, almost.. There are still two problems:

addressable/lib/addressable/uri.rb

Line 1422 in 7b13500

hash.sort.inject([]) do |accu, (key, value)|

addressable/lib/addressable/uri.rb

Line 1503 in 7b13500

new_query_values.sort! # Useful default for OAuth and caching

Need to drop the .sort call on 1422, and remove 1503 (or make optional / conditional)

        url = 'http://a.com/?'+('a'..'z').to_a.shuffle.map {|e| "#{e}=#{e}"}.join("&")

Above case should result in same url post query_values parse + query_parse=

[sporkmonger]

I was actually going to ask bvandenbos for some test cases to prove he'd gotten the sort order correct. But then I was lazy and didn't. Oh, well, as least my test-completeness instincts are working. I'll fix it right now.

[sporkmonger]

Bother. I just pushed a pair of test cases that cover some of the weird edge cases in some of the notation forms that addressable currently parses. Dropping the .sort call makes these tests fail in ruby 1.9.x. The .sort call on line 1422 is necessary. However I moved the sorting code on line 1503 up into the non-Array input block, so it will only sort if you give it a Hash input. Admittedly, this might be a bit aggressive in 1.9.x where order is preserved in Hash values, but you'll always have the Array option available if it matters. I'm not concerned about the occasional instance where someone has to do an extra .to_a call on their 1.9.x Hash to ensure order is preserved.

[igrigorik]

Why is 1422 necessary? Without removing that sort, can't make this pass:

https://gist.github.com/787989

[sporkmonger]

Hmmm. Looks like the .sort call is having an unintentional side-effect. The intent was to sort Hash values, not Array values. For instance: ?a[1]=one&a[0]=zero should parse to {'a' => ['zero', 'one']}.

[sporkmonger]

Actually, I added your test and the unmodified code passed just fine. Please take another look.

[igrigorik]

Interesting! You're right, works just fine. Oi! :-)

In which case, what are the odds of a point release to get this out in the wild?

[sporkmonger]

Done.

[igrigorik]

Sweet, thanks Bob!