In a normal Rails application with Devise, I can sign up or sign in as a user and then be redirected to the root_path of my application. In the action at this point, I am able to make reference to request.referer to determine where the request came from. This is incredibly helpful if you want to show people who have recently signed up a slightly different page because they have to confirm their accounts before they can see this action. All is sunshine and rainbows in the wonderful world of rails server.

However, when I run my Cucumber tests to test this particular feature, the Referer header is NOT set by rack-test and therefore my features fail. I believe this is a problem to do with rack-test, as I can see no code which would set the Referer header upon a redirect.

I have attempted to patch this myself and haven't been able to find a clean way to implement it yet.

in rack/test.rb:171, URI.parse is used to create a URI instance. Unfortunately, the uri library cannot handle URIs with unicode characters in them: http://bugs.ruby-lang.org/issues/1787

It'd be nice if rack-test avoided URI or was able to take advantage of something like https://github.com/sporkmonger/addressable which handles unicode URIs correctly. Even allowing URI-like objects to be passed into Rack::Test methods without being parsed (so we could use Addressable ourselves) would be great.

Querystring params with the special [] syntax seem to lose their values when passed through rack/test.

Browsing to http://localhost:3333/direct?box[]=1&box[]=2 gets params {"box"=>["1", "2"]}.
Hitting the same url in test gets params {"box"=>[nil, nil]}.

Example code is up here:
http://gist.github.com/399738

Is REQUEST_URI not getting set in the environment by design, or is it just not implemented?

When posting hash params to a server using rack-test, the ordering of elements can cause problems. This isn't a problem in a non-test scenario because the ordering in html determines the ordering of the param string, which gets parsed correctly by rack.
For example, in the following request, a change in the iterated order of the last hash param causes the "fairly" key to get associated with the previous hash when rack parses the generated query string. No idea how one would fix this short of using an OrderedHash in client's test code, but thought I'd mention it in case someone has an idea.

params = {'priorities' => [
    {"pattern" => "foo"},
    {"pattern" => "default"},
    {"pattern" => "bar", "fairly" => "true"}
]}
post "/queuepriority", params 

1.9
rack test generates: "priorities[][pattern]=foo&priorities[][pattern]=default&priorities[][pattern]=bar&priorities[][fairly]=true"
rack parses to: {"priorities"=>[{"pattern"=>"foo"}, {"pattern"=>"default"}, {"pattern"=>"bar", "fairly"=>"true"}]}

1.8
rack-test generates: "priorities[][pattern]=foo&priorities[][pattern]=default&priorities[][fairly]=true&priorities[][pattern]=bar"
rack parses to: {"priorities"=>[{"pattern"=>"foo"}, {"fairly"=>"true", "pattern"=>"default"}, {"pattern"=>"bar"}]}

I tried setting the Content-Type header like so:

header 'Content-Type', 'application/json'
get 'some/path'

However, rack-test stomps the content type header. First, #headers_for_environment converts it to HTTP_CONTENT_TYPE. Then #env_for sets the content type to application/x-www-form-urlencoded here. Since my content type header was converted to HTTP_CONTENT_TYPE, CONTENT_TYPE has no value and the ||= stomps it.

I'm willing to take a crack at fixing it and submitting a patch, but I'm not sure of the reason for the conversion to HTTP_CONTENT_TYPE. What's the best way to fix this?

Rack uses its parse_nested_query method to parse query strings, not parse_query: http://github.com/chneukirchen/rack/blob/master/lib/rack/request.rb#L265

Rack::Test uses the later: http://github.com/brynary/rack-test/blob/master/lib/rack/test.rb#L178

As the result of this inconsistency, query strings like 'a=1&a=2' are parsed to {:a => ['1', '2']} when testing with Rack::Test, but to {:a => '2'} when running the application with Rack (An explanation of why this approach has been chosen for Rack can be found here: http://groups.google.com/group/rack-devel/browse_thread/thread/61e0cfbfa0dd8dbd/c42e55d2cb6866dd?lnk=gst)

This patch fixes the issue: http://gist.github.com/587525

Thanks

Hi,

Thanks for this awesome library!

It seems that the conversion of arrays of hashes does not match with Rack's actual behaviour.

here's a spec that shows the issue. I basicallu reversed the direction of one of your existing specs for build_nested_query.

(to be added here: http://github.com/brynary/rack-test/blob/master/spec/rack/test/utils_spec.rb#L51)

    it "arrays of hashes conversion be correlate of Rack's parse_nested_query" do
      Rack::Utils.parse_nested_query("a[][b]=2&a[][c]=3").should == {"a" => [{"b" => "2"}, {"c" => "3"}]}
    end

The result of the spec is this:

expected: {"a"=>[{"b"=>"2"}, {"c"=>"3"}]},
     got: {"a"=>[{"b"=>"2", "c"=>"3"}]} (using ==)

Before I proceed with filing a bug against Rack, I'd like to know what you think the expected behaviour is.

My situation is that a cuke doesn't work with Rack::Test (the failure being to do with param parsing) but does work using the selenium driver, and in the actual app. This leads me to believe something is going screwy with Rack::Test...

Cheers,
Ian

In which ruby versions does String#encoding_aware? work? It was introduced in 0.5.5 and seems to break under 1.9.1-p378 and 1.8.7-p302. I'm running via cucumber and I suddenly had these "undefined method `encoding_aware?' for #String:0x5957c84 (NoMethodError)" errors that I traced back to the new rack-test code.

I noticed that https://github.com/brynary/rack-test/blob/master/lib/rack/test/cookie_jar.rb#L25 causes the URI path to be truncated.

irb(main):001:0> "/cookies/test".sub(/\/[^\/]*\Z/, "")
=> "/cookies"

I can have a go at fixing this (and related tests that may be broken) but I'm wondering if I'm missing something. What was the motivation behind the regexp?

I've got an app that uses hostnames to separate customers, and couldn't find an easy way to set the default_host for MockSession.

Googling showed that a number of other people have had the same problem, so I implemented a host-setter method, as well as added delegation for cookie_jar for easier cookie-related testing. The net result is two new methods for Rack::Test::Methods, and no changes to existing functionality.

I want to contribute the changes back, but I'm not sure how. I can just submit a patch, but it would be nice to include additional test cases covering the new functionality. Of course, I can fork the project, but I don't want to create Yet Another Copy Of Rack::Test For People To Choose From.

So, what should I do...?

The gemspec file lists that >rack-1.0.0 is required, but the spec_helper tries to load ~>1.0.0.

I've just ran the specs against rack 1.1.0 and from other bug reports and patches it seems that rack 1.1.0 is also supported, so probably the requirement in the spec_helper should be fixed? In any case it would be nice if both requirements matched.

I use rack-test and MiniTest religiously to functional and integration test my rack apps. For the most part it's a lot of parsing JSON and then comparing or inspecting hashes. But one thing I do in every test is check the response status (code). I'm constantly writing assert last_response.ok?, or since rack only provides helper methods for the most common response statuses, more often I'm comparing directly against integers, e.g. assert_equal last_response.status, 401. I got tired of the inconsistency of checking helpers and integers, so I eventually stopped asserting the helpers altogether. Then I moved to MiniTest's spec-style, but had to continue using the old/normal-style assertions mixed in with the spec-style, which I really didn't like. So I wrote my own little helpers gem brandonweiss/rack-minitest to add spec-style MiniTest matchers to rack-test for checking response statuses, like so.

This works fine, but I know so many people that do this themselves already, I'm wondering if there isn't some way to get official support for this, although I don't really know what that would look like, hence this thread.

Would it make sense for rack-test to be aware of MiniTest? I can certainly see arguments for and against. Or does it make sense to maintain a separate gem a bridge between the two that other people can use, like I've been doing?

While Sinatra will default the Content-Type to text/html, Rack::Test sets it as nil by default.

OAuth2 uses an authorisation header similar to digest authentication. The header is typically of the form:

Authorization: OAuth #{token}

You can see this was implemented in the OAuth2 gem in this commit.

If you use #header within Rack::Test::Methods to try to set the Authorize header authorisation won't work due to the key being rewritten. Perhaps #header could take an optional third argument, which explicitly preserves the key passed?

I am looking to use some of the env features now available in master, but not available in 0.6.2. Could you do a version bump for those features, added about a year ago now?

I'm including Rack::Test::Methods. After doing a request I do

rack_mock_session.cookie_jar.to_hash

But can't get the actual session that is set in my Sinatra app (it's sid). What I get is

{"rack.session"=>"BAh7BjoIc2lkSSI9RTlGWU5ndkw2M1psU01FcW1rQ2RWTnFsV1ppQkt6N3pt\nM05IRzBSSHIreVdLMDJJS2pSNFRRPT0GOg1lbmNvZGluZyINVVMtQVNDSUk=\n"}

Performing a base 64 decode I get

{"rack.session"=>"x04\b{\x06:\bsidI\"=E9FYNgvL63ZlSMEqmkCdVNqlWZiBKz7zm\x9C\xCD9!\x1C\xC1I!\xC8\xAD\xE5],\xC0\xC9%-\xA9H\xD1QD\xF4\xF4\x18\xE85\x95\xB9\x8D\xBD\x91\xA5\xB9\x9C\x885UL\xB5\x05M\r%$"}

I can see my sid cookie in there. Why is this only delivered like this? How can I get a hold of the actual session?

From the rfc at http://tools.ietf.org/html/rfc1945#section-11.1 :

"basic-cookie = <base64 [5] encoding of userid-password,
except not limited to 76 char/line>"

basic-cookie being the "userid:password" token of the "Authorization" header.

bottom line, the .pack('m*') in basic_authorize should read .pack('m0*') -- the 0 means that it shouldn't line-wrap, where the default behaviore of .pack('m') is to line wrap.

I'm using Rack::Test to test my app and need to test posting of data via AJAX. I've had no issues doing this with GET requests and AJAX but for POST I just can't get it to work.

My test looks like:

describe 'POST /user/' do
  include Rack::Test::Methods
  it 'must allow user registration with valid information' do
    post '/user', {
      username: 'test_reg',
      password: 'test_pass',
      email: '[email protected]'
    }.to_json, {"CONTENT_TYPE" => 'application/json', "HTTP_X_REQUESTED_WITH" => "XMLHttpRequest"}
    last_response.must_be :ok?
    last_response.body.must_match 'test_reg has been saved'
  end
end

Note I've also tried this by passing the hash and not the to_json string.

Is what I'm trying to do possible?

(FYI I have also posted this as a question on Stackoverflow).

I'm using digest authentication in my app and when writing test it seems I can't use get parameters because Rack::Auth::Digest::MD5#call does

if !auth.digest? || !auth.correct_uri? || !valid_qop?(auth)
  return bad_request
end

and auth.correct_uri is

def correct_uri?
  request.fullpath == uri
end

and when I call get '/users', 'limit' => '1' auth.request.fullpath is /users?limit=1 but auth.uri is /users so the request fail with an http 400, it seems to comes from Rack::Test::Session#process_request where we parse the uri but I'm not sure of it

see https://github.com/parisrb/httpcron/blob/test-params/test/admin_params_test.rb for a test that show the error

Not quite sure if it's an issue or me doing something wrong.
I define my Cucumber (Capybara + Rack::Test as web driver are used) step like this :

When /log in with (.+):(.+)$/ do |username, password|
  basic_authorize(username, password) #or just authorize(username, password)
end

In my controller before_filter method I log request header like this:

Rails.logger.info "#{request.headers['HTTP_AUTHORIZATION'].inspect}" 

Logger outputs nil.

If I change my step definition to:

When /log in with (.+):(.+)$/ do |username, password|
  page.driver.browser.basic_authorize(username, password)
end

Auth works fine.

Gem versions used:
capybara (1.1.1)
cucumber (1.1.0)
cucumber-rails (1.1.0)
rack (1.3.3)
rack-test (0.6.1)

Maybe it's cucumber/capybara issue and not a Rack::Test problem ?

Hi,

Thanks so much for Rack::Test; it's an incredibly useful library for me.

I think I might have uncovered a bug. I'd like to be able to specify an array of file uploads as params to a post request:

foo = Rack::Test::UploadedFile.new('foo.txt', 'text/plain')
bar = Rack::Test::UploadedFile.new('bar.txt', 'text/plain')

post "photos", :photo => { :file => [foo, bar]}

But in the controller, I receive the following params:

{"photo"=>{"file"=>[{"foo"=>"", "bar"=>""}]}, "action"=>"create", "controller"=>"photos"}

When posting a single file, the file is available in the params, as expected:

{"photo"=>{"file"=>#<File:/var/folders/QR/QR1Z8LNdGlaaW1vRT-ls2E+++TI/-Tmp-/RackMultipart20100322-9449-1o9a6i1-0>}, "action"=>"create", "controller"=>"photos"}

I'm happy to try a patch, but thought I'd ask whether I'm doing something stupid first!

Thanks very much.

Rails recently introduced a newline after opening <textarea> tag (rails/rails#4000). All browsers ignore this newline.

But rack-test does not. So if the test opens edit page with textarea, and clicks 'Save', then an additional trailing newline is added to the value, sent to the controller and saved to the DB.

The proof app is here: https://github.com/khustochka/newline_test

It has two integration tests, one driven with Selenium, another with rack-test. The first passes, the second fails.

I was using capybara to test my application and it was calling Rack::Test#get with http://www.doorkeeper.jp. As this did not include a trailing slash, cookies were not being set (as uri.path was nil, which doesn't match the regex), although the request otherwise responded normally. I don't want anyone else to go through the debugging headache, so it would be nice to fix this. Is RackTest the appropriate place to fix this? If so, what is your recommended approach (warning, exception, add the slash, etc)?

When attempting to test a view using Cucumber that contains the javascript_include_tag helper I get the following error,

Attempt to unlock a mutex which is not locked (ThreadError)

However after I comment out the helper all is well. Any suggestions as to what is causing this problem?

Brian

We'd really like to use the PATCH method that's on master but not released in any gems. It's been more than year since the last release and the additions that have been added seem relatively harmless.

Is there any reason to not release 0.6.2?

rack-test provides the header method that sets request headers, but there's no corresponding headers method to query what headers have been set. It would be really nice if such a method existed. Currently, we're doing current_session.instance_variable_get(:@headers) to get access to the headers, which feels like a brittle hack.

Please see related discussion and research in rack/rack#323.

RFC 1867 does not say to escape ASCII chars in filenames in multipart uploads using %XX (the URL-encoding scheme from RFC 1738). No browser I tried (see list in rack/rack#323) escapes ASCII filename characters in this way so Rack shouldn't either, with one exception: it could emulate either Firefox or Webkit by escaping only the double-quote character.

• Firefox escapes double-quote as \".
• Webkit escapes double-quote as %22.

All other ASCII characters--including single quote ('), percent (%), ampersand (&), question mark (?), backslash (\)--should be passed straight through without encoding.

Background: this bit me because rack-test escapes the entire filename using RFC 1738 encoding and rack unescapes the filename using the same method. I had a passing test in my app to verify that special characters were allowed in filenames and it was passing, but later I discovered that in a real browser with a % sign in the filename it raised an error in rack. The real browser did not escape the % sign but rack tried to unescape it. If rack-test were behaving more like a real browser and not escape percent signs, my test would have told me something was wrong.

The correct fix is for both rack-test and rack to be updated (which is why rack/rack#323 was filed along with this one).

Thanks!

cc @raggi

I was trying to use rack-test to perform a PUT with an api_key in the params and JSON in the input.

put "/data", {api_key: user.api_key}, {input: data.to_json, content_type: 'application/json'}

However, it appears the way env_for is implemented, it's not possible to mix params with :input.

I'm currently developing an app where many redirects occur.
The redirects are working properly, tests are working too for redirects which happens in "namespace"

map '/' 

But for tests where redirects occur in another namespace (eg.: map '/admin') the redirect is broken, last_response after follow_redirect! returns route not found

Created a repo where the error is reproduced: https://github.com/Benny1992/rack_redirect_bug

Run the tests with:

cd test
ruby admin_controller_test.rb
ruby root_controller_test.rb

currently, rack test does not honor the "SERVER_PORT" option.

Here's how the behavior should work:

describe "request" do
  it "allows SERVER_PORT to be set" do
    request "/", "SERVER_PORT" => "9000"
    last_request.port.should == 9000
    last_request.env["SERVER_PORT"].should == "9000"
  end

  it "honors a port set in HTTP_HOST" do
    request "/", "HTTP_HOST" => "example.org:9000"
    last_request.port.should == 9000
    last_request.env["HTTP_HOST"].should == "example.org:9000"
  end

  it "honors a port set via the Host header" do
    header "Host", "example.org:9000"
    request "/"
    last_request.port.should == 9000
    last_request.env["HTTP_HOST"].should == "example.org:9000"
  end
end

Hello @brynary. It will be nice to add some CI (for example Travis). I just fixed some warning in specs for MRI 2.0 and it will be nice to prove it in some CI links.

I have a monkyepatch:

# Monkeypatch fixing missing tempfile handler in rack-test
class Rack::Test::UploadedFile
  def tempfile
    @tempfile
  end
end

Because I call tempfile on an uploaded file in an application.

This also mignt indicate that there is a problem with the real uploaded file in Rails 3.1.1, since it does not accept a call to closed? and tempfile does.

Hi

Parameter values that are sent in a multipart body don't get escaped. This is causing an error when using rack-test to test file uploads with Rails 3 RC, the cause of the snowman character (http://github.com/rails/rails/commit/25215d7285db10e2c04d903f251b791342e4dd6a).

I've fix this issue in this commit of my fork:

http://github.com/alan/rack-test/commit/c5a4555545f706e53bb0ad96fb02345b92b0cb35

I am trying to emulate this behaviour:

curl -X POST "http://localhost:8080/test/uniqueuuid1337" \
     --data-binary "@a_small_image.jpg" \
     -v \

When I do this, I can actually read the image-file from the request body like this:

temp_file = Tempfile.new("pict")
request.body.rewind
temp_file.write(request.body.read)
temp_file.seek(0)
temp_file

But when I try to do the same with Rack-Test it fails in all kind of strange ways.

it "should upload an image" do
  file = File.open(File.expand_path("spec/images/a_small_image.jpg"))
  post "https://test.com/test/uniqueuuid1337", file
  # NoMethodError: undefined method `length' for #<File:0x007fb194a62578>
  # from /Users/kaspergrubbe/.rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/gems/rack-
  #1.5.2/lib/rack/mock.rb:135:in `env_for'
end

it "should upload an image" do
  file = File.open(File.expand_path("spec/images/a_small_image.jpg"))
  post "https://test.com/test/uniqueuuid1337", file.read
  # ArgumentError: invalid byte sequence in UTF-8
  # from /Users/kaspergrubbe/.rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/gems/rack-
  #1.5.2/lib/rack/utils.rb:104:in `normalize_params'
end

it "should upload an image" do
  file = File.open(File.expand_path("spec/images/a_small_image.jpg"))
  post "https://test.com/test/uniqueuuid1337", Rack::Test::UploadedFile.new("spec/images/a_small_image.jpg", "application/octet-stream")
  # ArgumentError: invalid byte sequence in UTF-8
  # from /Users/kaspergrubbe/.rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/gems/rack-
  #1.5.2/lib/rack/utils.rb:104:in `normalize_params'
end

Is there any way to send the same kind of request like curl does?

If a params hash is passed in with symbols for keys, the resulting params object inside the app can access the values using either the symbol or the string. Rack makes these values accessible through strings only.

The following test, therefore, passes when it should fail:

# in test
it "should send an email" do
  post '/email', to: '[email protected]'
  last_response.status.should == 204
end

# in sinatra app.rb
post '/email' do
  #snip
  begin
    @mailer.send params.clone.freeze
    halt 204
  rescue ArgumentError
    halt 422
  end
end

# in mailer.rb
def send(params)
  raise ArgumentError unless params.has_key? :to
  #snip
end

I have made a small fix to include the HTTP_HOST header. This is helpful when testing subdomains.

http://github.com/ihoka/rack-test/commit/2fb9949c990f6f97eba83bc9b19092641ce7864c

Getting this warning when using Rack::Test::UploadedFile with Rails 4.1:

Digest::Digest is deprecated; use Digest

I'm testing a Grape API through Rack::Test standalone (without Rails). It appears that when my Rack middleware consumes env['rack.input'] the grape API will see the params hash as being empty.

My midleware has code like this:

request_params = env['rack.input'].read

When I turn the same spec into a Rails request spec, it works just fine. I also know when calling the API from the command line (curl), I can get both my middleware and the request to process correctly, so there is some issue specific to use Rack::Test.

I've read through the Rack::Test code, but I'm not sure why consuming the rack.input seems to destroy it. Perhaps it's a different type of object in the mock request than it would normally be in real life, such that calling #read on env['rack.input'] causes the buffer to become empty?

Any insight would be appreciated. I am more than willing to put in some time doing a pull request if you could help me look in the right direction. Thanks!

Aaron Patterson's (@tenderlove) rewrite of ActionDispatch::Http::UploadedFile exposes tempfile as an attr_accessor.

Rack::Test::Uploaded file should do the same.

See: rails/rails@c52e2cf
http://marklunds.com/articles/one/433

When running capybara using rack-test, I'm getting this error:

undefined method parse' for Time:Class # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:58:inexpires'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:63:in expired?' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:82:inmatches?'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:167:in block in hash_for' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:166:ineach'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:166:in hash_for' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:143:infor'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/mock_session.rb:28:in request' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test.rb:207:inprocess_request'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test.rb:57:in get' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/rack-test-0.5.4/lib/rack/test.rb:154:infollow_redirect!'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/driver/rack_test_driver.rb:249:in block in follow_redirects!' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/driver/rack_test_driver.rb:248:intimes'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/driver/rack_test_driver.rb:248:in follow_redirects!' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/driver/rack_test_driver.rb:204:inprocess'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/driver/rack_test_driver.rb:85:in click' # /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/session.rb:46:inclick'
# /home/lunks/.rvm/gems/ruby-1.9.2-preview3@r3/gems/capybara-0.3.9/lib/capybara/dsl.rb:55:in click' # ./spec/integration/sign_in_guest_spec.rb:28:inblock (3 levels) in <top (required)>'

I realize it should be solved by requiring 'time' as it's being done now on cookie_jar.rb, but the error is still ocurring. Using Ruby 1.9.2 with Rails 3 and RSpec 2.

When building a multipart request, with multiple fields of the same name (specifically from Capybara) which has parameters formatted in the input hash like:

{'name' => 'John', 'language[]' => ['en', 'de']}

Rack-test incorrectly formulates the request with language[] as one field with the value set to the to_s value of the Ruby array (i.e.):

Content-Disposition: form-data; name="language[]"

["en", "de"]

instead of:

Content-Disposition: form-data; name="language[]"

en

Content-Disposition: form-data; name="language[]"

de

This causes Rails to parse the form inputs incorrectly.

I have created the following patch which fixes the issue on my side, but I am unsure of how this affects other functionality:

http://pastebin.com/jDPpFXp0

Is this possible? I need to test an API using JSON as the payload, but it looks like Rack::Test always assumes it's simulating a form submission. Do you know any way to make this work?

Bryan,

First thanks for all you awesome work on Rack::Test!

We are using Rack::Test to test the interplay of several rack apps. We've been using some custom headers, similar to X-Custom-Header. Our code expects that the request env should contain env['HTTP_X_CUSTOM_HEADER']. Looking through Rack::Test, we saw that headers are not renamed when Rack::Test::Session#header('X-Custom-Header','value') was used.

Is this intentional?

According to Google, all web servers send headers in this uppercase, underscorized form to CGI apps. Is this something Rack::Test should emulate or should we roll our own solution?

Thanks!

-Derek

When no port is specified as part of the request, the HTTP_HOST header has a trailing ':', without any port after it.

For instance, when doing
get '/' # => env['HTTP_HOST'] is "example.org:"

Fixed in ihoka/rack-test@814e3ed to skip the ':' when no port is specified.

I believe rack tests's cookie validation is too strict, in https://github.com/brynary/rack-test/blob/master/lib/rack/test/cookie_jar.rb#L77 it requires that the cookie's path attribute matches the request's path. I.e. it restricts the cookie path based on the requests's path and doesn't allow arbitrary paths. But I can't find any mention of that in the RFC (http://tools.ietf.org/id/draft-ietf-httpstate-cookie-23.txt).
I tested this also with Chrome, it allows to set cookies with arbitrary paths.
Thus I think the path check should be removed.

While testing a Rack middleware that proxies requests, I noticed that rack-test was giving me the following exception:

/home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:53:in `path': undefined method `strip' for #<Array:0x0000000377c4e8> (NoMethodError)
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:77:in `valid?'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:128:in `block in merge'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:126:in `each'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test/cookie_jar.rb:126:in `merge'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/mock_session.rb:35:in `request'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test.rb:207:in `process_request'
from /home/hal/.rvm/gems/ruby-1.9.2-p0/gems/rack-test-0.5.4/lib/rack/test.rb:57:in `get'
from /vault/1/code/ronin/ronin-web/spec/web/middleware/proxy_spec.rb:29:in `block (2 levels) in <top (required)>'

The problem stems from CookieJar#path assuming the path option will always be a String:

  # :api: private
  def path
    @options["path"].strip || "/"
  end

The path option is originally generated by Rack::Utils#parse_query, which will return an Array if query params are repeated:

include Rack::Utils

parse_query("csrf_id=8d0c781f207dfcc0e8db55bd467b01e0; path=/, _github_ses=BAh7BzoRbG9jYWxlX2d1ZXNzMCIKZmxhc2hJQzonQWN0aW9uQ29udHJvbGxlcjo6Rmxhc2g6OkZsYXNoSGFzaHsABjoKQHVzZWR7AA%3D%3D--e10506e0f6935897cafe4f56774e20aa35e579a5; path=/; expires=Wed, 01 Jan 2020 08:00:00 GMT; HttpOnly")['path']
# => ["/, _github_ses=BAh7BzoRbG9jYWxlX2d1ZXNzMCIKZmxhc2hJQzonQWN0aW9uQ29udHJvbGxlcjo6Rmxhc2g6OkZsYXNoSGFzaHsABjoKQHVzZWR7AA==--e10506e0f6935897cafe4f56774e20aa35e579a5", "/"]

Is it possible to add an option to make all the writes to env['rack.errors'] raise an exception?

This would make the code that searches for tests much easier to write and read.

I'm using rack-test in a rails app--both through cucumber/capybara and also in some standalone specs for the application's API. I tried upgrading to Rails 2.3.8 today (which includes an upgrade to rack 1.1), but I'm getting tons of errors like these:

undefined method `downcase' for nil:NilClass
/Users/myronmarston/code/Cuponice/.bundle/gems/activesupport-2.3.8/lib/active_support/whiny_nil.rb:52:in `method_missing'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:29:in `replaces?'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:130:in `<<'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:129:in `reject!'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:129:in `<<'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:124:in `merge'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:122:in `each'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test/cookie_jar.rb:122:in `merge'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/mock_session.rb:35:in `request'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test.rb:207:in `process_request'
/Users/myronmarston/code/Cuponice/.bundle/gems/rack-test-0.5.3/lib/rack/test.rb:57:in `get'

I've done some digging, and the reason for these errors is that CookieJar#merge is receiving a raw_cookies string like:

geography_id=4; path=/; expires=Mon, 25-May-2020 16:24:40 GMT

_cuponice_session=BAh7BjoPc2Vzc2lvbl9pZCIlZDg4M2Y4ZjYyNGI3ZjdkN2NjOTFkOTc3NDFmM2I2NzI%3D--208977a32d3a7b79904dd0f6144ea01da60c26e6; path=/; HttpOnly

This includes two newline characters in a row, and the method splits on newlines. I wind up with some blank cookies ("") and the code doesn't handle them well.

Is there a bug in rack 1.1 or rails 2.3.8 that's causing the extra newline? Or should rack-test be updated to handle the extra newlines?

The joining and splitting of the raw cookie string produces empty elements if the cookie string already contains a newline. The following patch removes these empty entries: http://gist.github.com/169106