Improve Puma::NullIO consistency with real IO

[casperisfine]

This isn't a big deal, but every small difference has a chance to let a bug slip as it's likely not every scenario is tested with both NullIO and a real IO.

[MSP-Greg]

Thanks for the PR. Five of the tests fail when only the test changes are added to master.

LGTM. Any thoughts on the below, which removes two asserts inside of assert_raises blocks, and changes ::Tempfile.new to ::Tempfile.create? Don't feel strongly about them...

diff --git a/test/test_null_io.rb b/test/test_null_io.rb
index a3b2a2b5..dd86ffc8 100644
--- a/test/test_null_io.rb
+++ b/test/test_null_io.rb
@@ -77,11 +77,11 @@ class TestNullIO < Minitest::Test
     assert_includes error.message, "no implicit conversion of Object into String"
 
     error = assert_raises TypeError do
-      assert_equal "", nio.read(0, Object.new)
+      nio.read(0, Object.new)
     end
 
     error = assert_raises TypeError do
-      assert_nil nio.read(1, Object.new)
+      nio.read(1, Object.new)
     end
     assert_includes error.message, "no implicit conversion of Object into String"
   end
@@ -133,7 +133,7 @@ end
 # ensure all the test behavior is accurate
 class TestNullIOConformance < TestNullIO
   def setup
-    self.nio = ::Tempfile.new
+    self.nio = ::Tempfile.create
     nio.sync = true
   end
 
@@ -141,4 +141,10 @@ class TestNullIOConformance < TestNullIO
     self.nio = StringIO.new
     super
   end
+
+  def teardown
+    return unless nio.is_a? ::File
+    nio.close
+    File.unlink nio.path
+  end
 end

[casperisfine]

removes two asserts inside of assert_raises blocks

Yup oversight on my part. Thanks for catching it.

[casperisfine]

PR updated.

[MSP-Greg]

@casperisfine Thanks again for the PR. As you stated "but every small difference has a chance to let a bug slip"

[MSP-Greg]

@casperisfine

Oversights. I always work locally with Ruby head builds, as I feel I should 'walk what I talk'.

A few tests check the response for a frozen IO, which should raise FrozenError. But FrozenError is not defined in Ruby 2.4.

My mistake. Note to self - always review the CI.

<Rant on> why are we supporting four EOL Ruby versions? <Rant off>

I'll fix...

[casperisfine]

Ah my bad, I should have paid attention to CI too.

FrozenError is not defined in Ruby 2.4.

Oh wow, crazy, I didn't remember FrozenError being such a novelty, but indeed:

[MSP-Greg]

I'm working on a fix, some non-MRI errors also. I'm not going to start on the EOL Ruby thing.

Oh wow, crazy, I didn't remember FrozenError being such a novelty, but indeed

I don't think it's fair to expect anyone to remember that. I certainly didn't.