If you change “ActionController::AbstractRequest” to “ActionController::Request” in the fix you will be good to go.

Rails 2.3 swapped out CGI requests for Rack requests and removed the “Abstract” part of the class name.

The HTTP Accept header is disabled by default now. You should prefer the use of formatted URLs (such as /customers/1.xml) to indicate the format that you want. If you need the Accept headers, you can turn them back on with config.action_controller.user_accept_header = true.

http://guides.rubyonrails.org/2_2_release_notes.html section 6.3

def access_denied
+ request.format = :html if request.env['HTTP_USER_AGENT'] =~ /msie/i && (request.format.to_s =~ /(text|html|xml|js|\*)/).nil?
respond_to do |format|
format.html do
store_location
redirect_to new_sessions_path
end
# format.any doesn’t work in rails version < http://dev.rubyonrails.org/changeset/8987
# you may want to change format.any to e.g. format.any(:js, ml)
format.any do
request_http_basic_authentication ‘Web Password’
end
end
end

My copy of IE6 was sending a type through. It was “image/gif” and for some reason the above code didn’t fire and the http_auth dialog was displayed.

With my code, it works properly. What do you think?

Best of luck!

def action
  respond_to do |format|
    format.html { # render html }
    format.xml  { # render xml }
    format.any  { # render json }
  end
end

If you have an Accept header that has text/html, application/xml, application/json, */*, the action will deliver html.

If you have an Accept header that has application/xml, text/html, application/json, */*, the action will deliver xml (first priority in the Accept list that matches).

If you have an Accept header that has text/plain, */* then the format.any will be selected and deliver json (impractical but good for my example).

And lastly if you just have */* as the Accept header, the action will deliver the first match in the action’s respond_to, in this case, html.

Because, IE is not specifying html, xhtml in it’s list but is specifying a bunch of other formats or just */*, then in the first case it will deliver text (format.any) and in the latter case it will deliver html (correctly).

If I cannot rely on the Accept header and the header does not list html or xhtml, then I would say if I know the user agent is MSIE I should normalize it’s Accept header to be the catch-all */* if it is not already.

The Accept header in IE is one of two values: */*, or a string made up from the values of the “Accepted Documents” key in the system registry. That key is accessible to anything installed on the system, so some applications (e.g. Office) and add-ons (Flash/Silverlight, etc) shove their own values in there.

So would you suggest I normalize all IE6 HTTP_ACCEPT headers to be the catch-all?

I’m still feeling unsure about normalizing the header to the catch-all rather than knowing what IE6’s true capabilities are via the Accept header.

You can not rely on the Accept header (in any version of IE) to determine whether or not the request represents a top-level navigation.