svenfuchs / routing-filter

routing-filter wraps around the complex beast that the Rails routing system is, allowing for unseen flexibility and power in Rails URL recognition and generation.
http://www.artweb-design.de
MIT License
464 stars 84 forks source link

Rails 3 rack-mount optimization doesn't work for constraints #15

Open clemens opened 14 years ago

clemens commented 14 years ago

It seems that the monkeypatches for the rack-mount optimizations don't work properly when using constraints in routes.

See this gist: https://gist.github.com/705804

URLs with subdomains are recognized correctly but not URLs without subdomains. As soon as I comment out the code generation stuff in routing filter, recognition works (but obviously routing filter doesn't work anymore).

I'm not 100% this is a problem specific to the constraints method – it could affect other aspects of routing, too.

svenfuchs commented 13 years ago

I see.

Can someone provide a failing test case, please?

clemens commented 13 years ago

Here it is: https://github.com/clemens/routing-filter/commit/01a9206ecf681f35d41982f6ee1cad7fb1c78f66.

clemens commented 13 years ago

Actually the constraints line should go before the other one because it's more specific (that's the rule in Rails 3, right?). What stays the same is that stuff is failing.

I still don't have a clue about the problem but one hint might be that constraint handling is a Rack middleware itself that relies on the X-Cascade header. So maybe routing filter needs to somehow take that into account? Just musing ...

svenfuchs commented 13 years ago

Ah, thanks for the test case, Clemens.

Will have a look asap, although ... that might be next year ;)

svenfuchs commented 13 years ago

Ok, I've had a look already.

The problem is related to the fact that Rack::Mount::RouteSet#call yields a block to recognize and this block calls return in order to break out of the optimized_each loop. Therefor it is not possible to actually wrap around that block. Whenever that return is called it will immediately return to the original method no matter how many filtering blocks are wrapped around that inner block. Yeah, that's just like using GOTO behaved in BASIC in 1982.

I'm not sure how to get around this. I can't believe this stuff is so fricking hard, but it seems it actually is.

dvandersluis commented 12 years ago

Has any progress on this issue ever been made? I'm running into the same thing myself and in trying to figure out a solution came to the same conclusions as you did a year ago, @svenfuchs.

ygor commented 12 years ago

I just ran into this problem too! Any ideas on how to fix it?

ygor commented 12 years ago

Ok, this fixes the problem I think, but of course is not a nice fix. I think an easy fix is not possible without refactoring parts of the core of this gem. Sven what do you think?

def recognize_with_filtering(request, &block)
  path, route, matches, params = request.env['PATH_INFO'], nil, nil, nil
  original_path = path.dup

  # first run of filters to possibly change the request path
  filters.run(:around_recognize, path, request.env) {{}}

  recognize_without_filtering(request) do |route, matches, params|
    request.env['PATH_INFO'] = original_path
    filters.run(:around_recognize, original_path, request.env) do
      params || {}
    end
    request.env['PATH_INFO'] = original_path
    if block_given?
      block.call(route, matches, params)
    else
      return route, matches, params
    end
  end
  nil
end