Add session#fetch

[dmathieu]

This adds a #fetch method to the session object.

session.fetch(:value, 'default')

I've made this fetch behave like Hash#fetch, which means it takes either a default value, a block or no value.

[senny]

Let's also add some rDoc to explain the usage and maybe refer to Hash#fetch

[senny]

just saw that saw that Session is :nodoc:... wondering why that is...

[dmathieu]

I would definitely do that.
But since no other method in this class has documentation, I would think it might not be the best place.

[senny]

It looks like all the rDoc related to sessions is currently stored in base.rb. @fxn where should such methods be documented? As Session is :nodoc it looks like it's an implementation detail and not part of the public API. It's a bit sad that this way you won't be able to find the docs using the "Search" feature of api.rubyonrails.org:

[guilleiguaran]

Looks fine for me 👍

iirc, some core members are a bit skeptic about add new methods to params/session hashes but this shouldn't be te case since we have fetch already in params 😄

[splattael]

AFAIK, Hash#fetch does not modify its state when invoked with a default value or a block.

>> h = {}
=> {}
>> h.fetch(23) { :yo }
=> :yo
>> h.fetch(23)
KeyError: key not found: 23
    from (irb):25:in `fetch'
    from (irb):25
    from /home/ps/.rvm/rubies/ruby-2.0.0-p195/bin/irb:16:in `<main>'

Fetching a missing key in ActionDispatch::Request::Session modifies its state:

>> s = ActionDispatch::Request::Session.create(store, {}, {})
=> #<ActionDispatch::Request::Session:0x1976b88 not yet loaded>
>> s.fetch(23) { :yo }
=> :yo
>> s.fetch(23)
=> :yo

The second call should raise a KeyError.

The implementation of Session#fetch must not store the default value or the result of block.

[dmathieu]

Actually, I did this on purpose, as I would think this could lead to unexpected behaviors.

session.fetch(:read_articles, {})[21] = true

In this case, the saved value would be expected to be saved in the session.
There's no point in just having an anonymous hash object.

I agree I missed a test for this though. I can happily add it, or change the current behavior.

[rrrene]

I'm with @splattael on this one. The behaviour you implement might have very valid use cases (but the example you give here isn't one of them and seems artificial). But it is not okay if the feature is advertised as "I've made this fetch behave like Hash#fetch" and then it doesn't live up to this promise (and moreover has serious side-effects).

On a side note, I strongly disagree with this opinion:

There's no point in just having an anonymous hash object.

This is false on so many levels 😔

[rafaelfranca]

@dmathieu I agree with both about the different behavior between Hash#fetch and Session#fetch, but I also agree this is a nice thing. So I suggest to remove the sentence "fetch behaves like Hash#fetch." from changelog since it doesn't behave.

[dmathieu]

@splattael, @rrrene: I just pushed a fix to the concern you raised, which has just been merged. Thanks @rafaelfranca 💛