[PATCH] deflater: remove "deflate" encoding support

48 views
Skip to first unread message

Eric Wong

unread,
Jul 26, 2016, 10:10:52 PM7/26/16
to rack-...@googlegroups.com
This will improve cache hit rates and reduce caching overhead at
small expense of increased header overhead for some user agents.
For reference, Varnish cache supports only gzip as well:

https://www.varnish-cache.org/docs/4.1/phk/gzip.html

In the past, "deflate" encoding was more likely to trigger
user-agent bugs, as noted in the comments removed with this
change as well as the Varnish documentation referenced above.
---
The following changes since commit 25a549883b85fb33970b4a1530a365c0c9e51f95:

bumping to 2.0.1 to work around Rack (2016-06-30 10:33:09 -0700)

are available in the git repository at:

git://80x24.org/rack.git no-deflate

for you to fetch changes up to d6380043a8953dca63743c947c8027f465d29a5d:

deflater: remove "deflate" encoding support (2016-07-26 23:14:55 +0000)

----------------------------------------------------------------
Eric Wong (1):
deflater: remove "deflate" encoding support

lib/rack/deflater.rb | 37 +------------------------------------
test/spec_deflater.rb | 38 ++++++++++++++++++++++++--------------
2 files changed, 25 insertions(+), 50 deletions(-)

diff --git a/lib/rack/deflater.rb b/lib/rack/deflater.rb
index 62a1124..79e39f5 100644
--- a/lib/rack/deflater.rb
+++ b/lib/rack/deflater.rb
@@ -8,7 +8,6 @@ module Rack
# Currently supported compression algorithms:
#
# * gzip
- # * deflate
# * identity (no transformation)
#
# The middleware automatically detects when compression is supported
@@ -41,7 +40,7 @@ module Rack

request = Request.new(env)

- encoding = Utils.select_best_encoding(%w(gzip deflate identity),
+ encoding = Utils.select_best_encoding(%w(gzip identity),
request.accept_encoding)

# Set the Vary HTTP header.
@@ -57,10 +56,6 @@ module Rack
mtime = headers.key?("Last-Modified") ?
Time.httpdate(headers["Last-Modified"]) : Time.now
[status, headers, GzipStream.new(body, mtime)]
- when "deflate"
- headers['Content-Encoding'] = "deflate"
- headers.delete(CONTENT_LENGTH)
- [status, headers, DeflateStream.new(body)]
when "identity"
[status, headers, body]
when nil
@@ -101,36 +96,6 @@ module Rack
end
end

- class DeflateStream
- DEFLATE_ARGS = [
- Zlib::DEFAULT_COMPRESSION,
- # drop the zlib header which causes both Safari and IE to choke
- -Zlib::MAX_WBITS,
- Zlib::DEF_MEM_LEVEL,
- Zlib::DEFAULT_STRATEGY
- ]
-
- def initialize(body)
- @body = body
- @closed = false
- end
-
- def each
- deflator = ::Zlib::Deflate.new(*DEFLATE_ARGS)
- @body.each { |part| yield deflator.deflate(part, Zlib::SYNC_FLUSH) }
- yield fin = deflator.finish
- ensure
- deflator.finish unless fin
- deflator.close
- end
-
- def close
- return if @closed
- @closed = true
- @body.close if @body.respond_to?(:close)
- end
- end
-
private

def should_deflate?(env, status, headers, body)
diff --git a/test/spec_deflater.rb b/test/spec_deflater.rb
index ba7ec5d..0f27c85 100644
--- a/test/spec_deflater.rb
+++ b/test/spec_deflater.rb
@@ -81,13 +81,22 @@ describe Rack::Deflater do
yield(status, headers, body) if block_given?
end

+ # automatic gzip detection (streamable)
+ def auto_inflater
+ Zlib::Inflate.new(32 + Zlib::MAX_WBITS)
+ end
+
+ def deflate_or_gzip
+ {'deflate, gzip' => 'gzip'}
+ end
+
it 'be able to deflate bodies that respond to each' do
app_body = Object.new
class << app_body; def each; yield('foo'); yield('bar'); end; end

- verify(200, 'foobar', 'deflate', { 'app_body' => app_body }) do |status, headers, body|
+ verify(200, 'foobar', deflate_or_gzip, { 'app_body' => app_body }) do |status, headers, body|
headers.must_equal({
- 'Content-Encoding' => 'deflate',
+ 'Content-Encoding' => 'gzip',
'Vary' => 'Accept-Encoding',
'Content-Type' => 'text/plain'
})
@@ -98,15 +107,15 @@ describe Rack::Deflater do
app_body = Object.new
class << app_body; def each; yield('foo'); yield('bar'); end; end

- verify(200, app_body, 'deflate', { 'skip_body_verify' => true }) do |status, headers, body|
+ verify(200, app_body, deflate_or_gzip, { 'skip_body_verify' => true }) do |status, headers, body|
headers.must_equal({
- 'Content-Encoding' => 'deflate',
+ 'Content-Encoding' => 'gzip',
'Vary' => 'Accept-Encoding',
'Content-Type' => 'text/plain'
})

buf = []
- inflater = Zlib::Inflate.new(-Zlib::MAX_WBITS)
+ inflater = auto_inflater
body.each { |part| buf << inflater.inflate(part) }
buf << inflater.finish

@@ -118,32 +127,33 @@ describe Rack::Deflater do
app_body = Object.new
class << app_body; def each; yield('foo'); yield('bar'); end; end
opts = { 'skip_body_verify' => true }
- verify(200, app_body, 'deflate', opts) do |status, headers, body|
+ verify(200, app_body, 'gzip', opts) do |status, headers, body|
headers.must_equal({
- 'Content-Encoding' => 'deflate',
+ 'Content-Encoding' => 'gzip',
'Vary' => 'Accept-Encoding',
'Content-Type' => 'text/plain'
})

buf = []
- inflater = Zlib::Inflate.new(-Zlib::MAX_WBITS)
+ inflater = auto_inflater
FakeDisconnect = Class.new(RuntimeError)
assert_raises(FakeDisconnect, "not Zlib::DataError not raised") do
body.each do |part|
- buf << inflater.inflate(part)
+ tmp = inflater.inflate(part)
+ buf << tmp if tmp.bytesize > 0
raise FakeDisconnect
end
end
- assert_raises(Zlib::BufError) { inflater.finish }
+ inflater.finish
buf.must_equal(%w(foo))
end
end

# TODO: This is really just a special case of the above...
it 'be able to deflate String bodies' do
- verify(200, 'Hello world!', 'deflate') do |status, headers, body|
+ verify(200, 'Hello world!', deflate_or_gzip) do |status, headers, body|
headers.must_equal({
- 'Content-Encoding' => 'deflate',
+ 'Content-Encoding' => 'gzip',
'Vary' => 'Accept-Encoding',
'Content-Type' => 'text/plain'
})
@@ -280,7 +290,7 @@ describe Rack::Deflater do
'Content-Encoding' => 'identity'
}
}
- verify(200, 'Hello World!', 'deflate', options)
+ verify(200, 'Hello World!', deflate_or_gzip, options)
end

it "deflate if content-type matches :include" do
@@ -334,7 +344,7 @@ describe Rack::Deflater do
:if => lambda { |env, status, headers, body| true }
}
}
- verify(200, 'Hello World!', 'deflate', options)
+ verify(200, 'Hello World!', deflate_or_gzip, options)
end

it "not deflate if :if lambda evaluates to false" do
--
EW

James Tucker

unread,
Jul 26, 2016, 10:44:33 PM7/26/16
to Rack Development

Can you explain the cache hit rate assertions you open with? Sorry for evil post, mobile client.


--

---
You received this message because you are subscribed to the Google Groups "Rack Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rack-devel+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Eric Wong

unread,
Jul 27, 2016, 2:40:29 AM7/27/16
to rack-...@googlegroups.com
James Tucker <jftu...@gmail.com> wrote:
> Can you explain the cache hit rate assertions you open with? Sorry for evil
> post, mobile client.

Sure. Some clients send "deflate, gzip", while others send
"gzip, deflate", just "gzip", or "gzip, bunch-of-other-things".
From informal observations dumping headers on live requests,
gzip support seems universal among clients which are capable
of decompressing anything at all; and most clients handle
compression nowadays.

When a (rare) client comes along preferring "deflate" over gzip,
the caching layer could either:

1) request & store and additional copy
2) reencode any already-cached gzip version (gunzip & deflate)

That's inefficient either way compared to just serving an
already-cached gzip version.

curl is one client which prefers deflate over gzip,
maybe there are others; but I haven't noticed them.


P.S. Honestly, my preference is not to quote at all on
publically-archived lists. It's the most space-efficient :>

Eric Wong

unread,
Jul 27, 2016, 4:46:05 AM7/27/16
to rack-...@googlegroups.com
I should add that any modern caching layer would likely normalize
"Accept-Encoding: gzip" the same way Varnish does; in which case
the deflate code path is going to be dead code.
(and more likely to trigger bugs in some clients or have bugs of
its own)

James Tucker

unread,
Aug 23, 2016, 3:40:36 PM8/23/16
to Rack Development
SGTM

--

---
You received this message because you are subscribed to the Google Groups "Rack Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rack-devel+unsubscribe@googlegroups.com.

Vy Nguyen

unread,
Apr 23, 2019, 4:27:09 AM4/23/19
to Rack Development
Vào 09:44:33 UTC+7 Thứ Tư, ngày 27 tháng 7 năm 2016, raggi đã viết:
> To unsubscribe from this group and stop receiving emails from it, send an email to rack-...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages