Reduce default compression to level 6; raise min compressible length.

Partially addresses #1095.

diff --git a/tornado/test/simple_httpclient_test.py b/tornado/test/simple_httpclient_test.py
index b301c7e6efd87906d0aa9768c326fd7942608519..dc4c865bb9e033e8e79f58e117f56f7b74ef34c0 100644 (file)
--- a/tornado/test/simple_httpclient_test.py
+++ b/tornado/test/simple_httpclient_test.py
@@ -205,6 +205,7 @@ class SimpleHTTPClientTestMixin(object):
         self.assertEqual(response.headers["Content-Encoding"], "gzip")
         self.assertNotEqual(response.body, b"asdfqwer")
         # Our test data gets bigger when gzipped.  Oops.  :)
+        # Chunked encoding bypasses the MIN_LENGTH check.
         self.assertEqual(len(response.body), 34)
         f = gzip.GzipFile(mode="r", fileobj=response.buffer)
         self.assertEqual(f.read(), b"asdfqwer")

diff --git a/tornado/test/web_test.py b/tornado/test/web_test.py
index 85c7552c9ebfe8950a55893e067cf9043c83bdf8..5938b7b683ccb5d4202c0fdcbb3de446ea977158 100644 (file)
--- a/tornado/test/web_test.py
+++ b/tornado/test/web_test.py
@@ -11,7 +11,7 @@ from tornado.template import DictLoader
 from tornado.testing import AsyncHTTPTestCase, ExpectLog, gen_test
 from tornado.test.util import unittest
 from tornado.util import u, ObjectDict, unicode_type, timedelta_to_seconds
-from tornado.web import RequestHandler, authenticated, Application, asynchronous, url, HTTPError, StaticFileHandler, _create_signature_v1, create_signed_value, decode_signed_value, ErrorHandler, UIModule, MissingArgumentError, stream_request_body, Finish, removeslash, addslash, RedirectHandler as WebRedirectHandler, get_signature_key_version
+from tornado.web import RequestHandler, authenticated, Application, asynchronous, url, HTTPError, StaticFileHandler, _create_signature_v1, create_signed_value, decode_signed_value, ErrorHandler, UIModule, MissingArgumentError, stream_request_body, Finish, removeslash, addslash, RedirectHandler as WebRedirectHandler, get_signature_key_version, GZipContentEncoding
 
 import binascii
 import contextlib
@@ -1483,7 +1483,8 @@ class GzipTestCase(SimpleHandlerTestCase):
         def get(self):
             if self.get_argument('vary', None):
                 self.set_header('Vary', self.get_argument('vary'))
-            self.write('hello world')
+            # Must write at least MIN_LENGTH bytes to activate compression.
+            self.write('hello world' + ('!' * GZipContentEncoding.MIN_LENGTH))
 
     def get_app_kwargs(self):
         return dict(

diff --git a/tornado/web.py b/tornado/web.py
index b2fa6086e6050281507b45f63e4bfe091ca34f21..b6197ce7e90d1dcf41d7d722ed79032010571e76 100644 (file)
--- a/tornado/web.py
+++ b/tornado/web.py
@@ -2658,7 +2658,16 @@ class GZipContentEncoding(OutputTransform):
     CONTENT_TYPES = set(["application/javascript", "application/x-javascript",
                          "application/xml", "application/atom+xml",
                          "application/json", "application/xhtml+xml"])
-    MIN_LENGTH = 5
+    # Python's GzipFile defaults to level 9, while most other gzip
+    # tools (including gzip itself) default to 6, which is probably a
+    # better CPU/size tradeoff.
+    GZIP_LEVEL = 6
+    # Responses that are too short are unlikely to benefit from gzipping
+    # after considering the "Content-Encoding: gzip" header and the header
+    # inside the gzip encoding.
+    # Note that responses written in multiple chunks will be compressed
+    # regardless of size.
+    MIN_LENGTH = 1024
 
     def __init__(self, request):
         self._gzipping = "gzip" in request.headers.get("Accept-Encoding", "")
@@ -2679,7 +2688,8 @@ class GZipContentEncoding(OutputTransform):
         if self._gzipping:
             headers["Content-Encoding"] = "gzip"
             self._gzip_value = BytesIO()
-            self._gzip_file = gzip.GzipFile(mode="w", fileobj=self._gzip_value)
+            self._gzip_file = gzip.GzipFile(mode="w", fileobj=self._gzip_value,
+                                            compresslevel=self.GZIP_LEVEL)
             chunk = self.transform_chunk(chunk, finishing)
             if "Content-Length" in headers:
                 # The original content length is no longer correct.

diff --git a/tornado/websocket.py b/tornado/websocket.py
index 2f57b99093ac76a9a545c676dd8550bbf66ae626..d688295ff97216b122c557bbc87913ad04ffde86 100644 (file)
--- a/tornado/websocket.py
+++ b/tornado/websocket.py
@@ -444,7 +444,8 @@ class _PerMessageDeflateCompressor(object):
             self._compressor = None
 
     def _create_compressor(self):
-        return zlib.compressobj(-1, zlib.DEFLATED, -self._max_wbits)
+        return zlib.compressobj(tornado.web.GZipContentEncoding.GZIP_LEVEL,
+                                zlib.DEFLATED, -self._max_wbits)
 
     def compress(self, data):
         compressor = self._compressor or self._create_compressor()