bpo-39389: gzip: fix compression level metadata (GH-18077)
As described in RFC 1952, section 2.3.1, the XFL (eXtra FLags) byte of a
gzip member header should indicate whether the DEFLATE algorithm was
tuned for speed or compression ratio. Prior to this patch, archives
emitted by the `gzip` module always indicated maximum compression.
diff --git a/Lib/gzip.py b/Lib/gzip.py
index e60d8ad..e422773 100644
--- a/Lib/gzip.py
+++ b/Lib/gzip.py
@@ -209,7 +209,7 @@
self.fileobj = fileobj
if self.mode == WRITE:
- self._write_gzip_header()
+ self._write_gzip_header(compresslevel)
@property
def filename(self):
@@ -236,7 +236,7 @@
self.bufsize = 0
self.offset = 0 # Current file offset for seek(), tell(), etc
- def _write_gzip_header(self):
+ def _write_gzip_header(self, compresslevel):
self.fileobj.write(b'\037\213') # magic header
self.fileobj.write(b'\010') # compression method
try:
@@ -257,7 +257,13 @@
if mtime is None:
mtime = time.time()
write32u(self.fileobj, int(mtime))
- self.fileobj.write(b'\002')
+ if compresslevel == _COMPRESS_LEVEL_BEST:
+ xfl = b'\002'
+ elif compresslevel == _COMPRESS_LEVEL_FAST:
+ xfl = b'\004'
+ else:
+ xfl = b'\000'
+ self.fileobj.write(xfl)
self.fileobj.write(b'\377')
if fname:
self.fileobj.write(fname + b'\000')
diff --git a/Lib/test/test_gzip.py b/Lib/test/test_gzip.py
index 57d851c..7833421 100644
--- a/Lib/test/test_gzip.py
+++ b/Lib/test/test_gzip.py
@@ -358,6 +358,26 @@
isizeBytes = fRead.read(4)
self.assertEqual(isizeBytes, struct.pack('<i', len(data1)))
+ def test_compresslevel_metadata(self):
+ # see RFC 1952: http://www.faqs.org/rfcs/rfc1952.html
+ # specifically, discussion of XFL in section 2.3.1
+ cases = [
+ ('fast', 1, b'\x04'),
+ ('best', 9, b'\x02'),
+ ('tradeoff', 6, b'\x00'),
+ ]
+ xflOffset = 8
+
+ for (name, level, expectedXflByte) in cases:
+ with self.subTest(name):
+ fWrite = gzip.GzipFile(self.filename, 'w', compresslevel=level)
+ with fWrite:
+ fWrite.write(data1)
+ with open(self.filename, 'rb') as fRead:
+ fRead.seek(xflOffset)
+ xflByte = fRead.read(1)
+ self.assertEqual(xflByte, expectedXflByte)
+
def test_with_open(self):
# GzipFile supports the context management protocol
with gzip.GzipFile(self.filename, "wb") as f: