mirror of
https://github.com/python-pillow/Pillow.git
synced 2024-11-11 04:07:21 +03:00
d50445ff30
Similar to the recent adoption of Black. isort is a Python utility to sort imports alphabetically and automatically separate into sections. By using isort, contributors can quickly and automatically conform to the projects style without thinking. Just let the tool do it. Uses the configuration recommended by the Black to avoid conflicts of style. Rewrite TestImageQt.test_deprecated to no rely on import order.
678 lines
22 KiB
Python
678 lines
22 KiB
Python
import sys
|
|
import zlib
|
|
from io import BytesIO
|
|
|
|
from PIL import Image, ImageFile, PngImagePlugin
|
|
from PIL._util import py3
|
|
|
|
from .helper import PillowLeakTestCase, PillowTestCase, hopper, unittest
|
|
|
|
try:
|
|
from PIL import _webp
|
|
|
|
HAVE_WEBP = True
|
|
except ImportError:
|
|
HAVE_WEBP = False
|
|
|
|
codecs = dir(Image.core)
|
|
|
|
|
|
# sample png stream
|
|
|
|
TEST_PNG_FILE = "Tests/images/hopper.png"
|
|
|
|
# stuff to create inline PNG images
|
|
|
|
MAGIC = PngImagePlugin._MAGIC
|
|
|
|
|
|
def chunk(cid, *data):
|
|
test_file = BytesIO()
|
|
PngImagePlugin.putchunk(*(test_file, cid) + data)
|
|
return test_file.getvalue()
|
|
|
|
|
|
o32 = PngImagePlugin.o32
|
|
|
|
IHDR = chunk(b"IHDR", o32(1), o32(1), b"\x08\x02", b"\0\0\0")
|
|
IDAT = chunk(b"IDAT")
|
|
IEND = chunk(b"IEND")
|
|
|
|
HEAD = MAGIC + IHDR
|
|
TAIL = IDAT + IEND
|
|
|
|
|
|
def load(data):
|
|
return Image.open(BytesIO(data))
|
|
|
|
|
|
def roundtrip(im, **options):
|
|
out = BytesIO()
|
|
im.save(out, "PNG", **options)
|
|
out.seek(0)
|
|
return Image.open(out)
|
|
|
|
|
|
class TestFilePng(PillowTestCase):
|
|
def setUp(self):
|
|
if "zip_encoder" not in codecs or "zip_decoder" not in codecs:
|
|
self.skipTest("zip/deflate support not available")
|
|
|
|
def get_chunks(self, filename):
|
|
chunks = []
|
|
with open(filename, "rb") as fp:
|
|
fp.read(8)
|
|
with PngImagePlugin.PngStream(fp) as png:
|
|
while True:
|
|
cid, pos, length = png.read()
|
|
chunks.append(cid)
|
|
try:
|
|
s = png.call(cid, pos, length)
|
|
except EOFError:
|
|
break
|
|
png.crc(cid, s)
|
|
return chunks
|
|
|
|
def test_sanity(self):
|
|
|
|
# internal version number
|
|
self.assertRegex(Image.core.zlib_version, r"\d+\.\d+\.\d+(\.\d+)?$")
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
|
|
hopper("RGB").save(test_file)
|
|
|
|
im = Image.open(test_file)
|
|
im.load()
|
|
self.assertEqual(im.mode, "RGB")
|
|
self.assertEqual(im.size, (128, 128))
|
|
self.assertEqual(im.format, "PNG")
|
|
self.assertEqual(im.get_format_mimetype(), "image/png")
|
|
|
|
for mode in ["1", "L", "P", "RGB", "I", "I;16"]:
|
|
im = hopper(mode)
|
|
im.save(test_file)
|
|
reloaded = Image.open(test_file)
|
|
if mode == "I;16":
|
|
reloaded = reloaded.convert(mode)
|
|
self.assert_image_equal(reloaded, im)
|
|
|
|
def test_invalid_file(self):
|
|
invalid_file = "Tests/images/flower.jpg"
|
|
|
|
self.assertRaises(SyntaxError, PngImagePlugin.PngImageFile, invalid_file)
|
|
|
|
def test_broken(self):
|
|
# Check reading of totally broken files. In this case, the test
|
|
# file was checked into Subversion as a text file.
|
|
|
|
test_file = "Tests/images/broken.png"
|
|
self.assertRaises(IOError, Image.open, test_file)
|
|
|
|
def test_bad_text(self):
|
|
# Make sure PIL can read malformed tEXt chunks (@PIL152)
|
|
|
|
im = load(HEAD + chunk(b"tEXt") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"tEXt", b"spam") + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"tEXt", b"spam\0") + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"tEXt", b"spam\0egg") + TAIL)
|
|
self.assertEqual(im.info, {"spam": "egg"})
|
|
|
|
im = load(HEAD + chunk(b"tEXt", b"spam\0egg\0") + TAIL)
|
|
self.assertEqual(im.info, {"spam": "egg\x00"})
|
|
|
|
def test_bad_ztxt(self):
|
|
# Test reading malformed zTXt chunks (python-pillow/Pillow#318)
|
|
|
|
im = load(HEAD + chunk(b"zTXt") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"zTXt", b"spam") + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"zTXt", b"spam\0") + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"zTXt", b"spam\0\0") + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"zTXt", b"spam\0\0" + zlib.compress(b"egg")[:1]) + TAIL)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(HEAD + chunk(b"zTXt", b"spam\0\0" + zlib.compress(b"egg")) + TAIL)
|
|
self.assertEqual(im.info, {"spam": "egg"})
|
|
|
|
def test_bad_itxt(self):
|
|
|
|
im = load(HEAD + chunk(b"iTXt") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"iTXt", b"spam") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"iTXt", b"spam\0") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"iTXt", b"spam\0\x02") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"iTXt", b"spam\0\0\0foo\0") + TAIL)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(HEAD + chunk(b"iTXt", b"spam\0\0\0en\0Spam\0egg") + TAIL)
|
|
self.assertEqual(im.info, {"spam": "egg"})
|
|
self.assertEqual(im.info["spam"].lang, "en")
|
|
self.assertEqual(im.info["spam"].tkey, "Spam")
|
|
|
|
im = load(
|
|
HEAD
|
|
+ chunk(b"iTXt", b"spam\0\1\0en\0Spam\0" + zlib.compress(b"egg")[:1])
|
|
+ TAIL
|
|
)
|
|
self.assertEqual(im.info, {"spam": ""})
|
|
|
|
im = load(
|
|
HEAD
|
|
+ chunk(b"iTXt", b"spam\0\1\1en\0Spam\0" + zlib.compress(b"egg"))
|
|
+ TAIL
|
|
)
|
|
self.assertEqual(im.info, {})
|
|
|
|
im = load(
|
|
HEAD
|
|
+ chunk(b"iTXt", b"spam\0\1\0en\0Spam\0" + zlib.compress(b"egg"))
|
|
+ TAIL
|
|
)
|
|
self.assertEqual(im.info, {"spam": "egg"})
|
|
self.assertEqual(im.info["spam"].lang, "en")
|
|
self.assertEqual(im.info["spam"].tkey, "Spam")
|
|
|
|
def test_interlace(self):
|
|
|
|
test_file = "Tests/images/pil123p.png"
|
|
im = Image.open(test_file)
|
|
|
|
self.assert_image(im, "P", (162, 150))
|
|
self.assertTrue(im.info.get("interlace"))
|
|
|
|
im.load()
|
|
|
|
test_file = "Tests/images/pil123rgba.png"
|
|
im = Image.open(test_file)
|
|
|
|
self.assert_image(im, "RGBA", (162, 150))
|
|
self.assertTrue(im.info.get("interlace"))
|
|
|
|
im.load()
|
|
|
|
def test_load_transparent_p(self):
|
|
test_file = "Tests/images/pil123p.png"
|
|
im = Image.open(test_file)
|
|
|
|
self.assert_image(im, "P", (162, 150))
|
|
im = im.convert("RGBA")
|
|
self.assert_image(im, "RGBA", (162, 150))
|
|
|
|
# image has 124 unique alpha values
|
|
self.assertEqual(len(im.getchannel("A").getcolors()), 124)
|
|
|
|
def test_load_transparent_rgb(self):
|
|
test_file = "Tests/images/rgb_trns.png"
|
|
im = Image.open(test_file)
|
|
self.assertEqual(im.info["transparency"], (0, 255, 52))
|
|
|
|
self.assert_image(im, "RGB", (64, 64))
|
|
im = im.convert("RGBA")
|
|
self.assert_image(im, "RGBA", (64, 64))
|
|
|
|
# image has 876 transparent pixels
|
|
self.assertEqual(im.getchannel("A").getcolors()[0][0], 876)
|
|
|
|
def test_save_p_transparent_palette(self):
|
|
in_file = "Tests/images/pil123p.png"
|
|
im = Image.open(in_file)
|
|
|
|
# 'transparency' contains a byte string with the opacity for
|
|
# each palette entry
|
|
self.assertEqual(len(im.info["transparency"]), 256)
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
# check if saved image contains same transparency
|
|
im = Image.open(test_file)
|
|
self.assertEqual(len(im.info["transparency"]), 256)
|
|
|
|
self.assert_image(im, "P", (162, 150))
|
|
im = im.convert("RGBA")
|
|
self.assert_image(im, "RGBA", (162, 150))
|
|
|
|
# image has 124 unique alpha values
|
|
self.assertEqual(len(im.getchannel("A").getcolors()), 124)
|
|
|
|
def test_save_p_single_transparency(self):
|
|
in_file = "Tests/images/p_trns_single.png"
|
|
im = Image.open(in_file)
|
|
|
|
# pixel value 164 is full transparent
|
|
self.assertEqual(im.info["transparency"], 164)
|
|
self.assertEqual(im.getpixel((31, 31)), 164)
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
# check if saved image contains same transparency
|
|
im = Image.open(test_file)
|
|
self.assertEqual(im.info["transparency"], 164)
|
|
self.assertEqual(im.getpixel((31, 31)), 164)
|
|
self.assert_image(im, "P", (64, 64))
|
|
im = im.convert("RGBA")
|
|
self.assert_image(im, "RGBA", (64, 64))
|
|
|
|
self.assertEqual(im.getpixel((31, 31)), (0, 255, 52, 0))
|
|
|
|
# image has 876 transparent pixels
|
|
self.assertEqual(im.getchannel("A").getcolors()[0][0], 876)
|
|
|
|
def test_save_p_transparent_black(self):
|
|
# check if solid black image with full transparency
|
|
# is supported (check for #1838)
|
|
im = Image.new("RGBA", (10, 10), (0, 0, 0, 0))
|
|
self.assertEqual(im.getcolors(), [(100, (0, 0, 0, 0))])
|
|
|
|
im = im.convert("P")
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
# check if saved image contains same transparency
|
|
im = Image.open(test_file)
|
|
self.assertEqual(len(im.info["transparency"]), 256)
|
|
self.assert_image(im, "P", (10, 10))
|
|
im = im.convert("RGBA")
|
|
self.assert_image(im, "RGBA", (10, 10))
|
|
self.assertEqual(im.getcolors(), [(100, (0, 0, 0, 0))])
|
|
|
|
def test_save_greyscale_transparency(self):
|
|
for mode, num_transparent in {"1": 1994, "L": 559, "I": 559}.items():
|
|
in_file = "Tests/images/" + mode.lower() + "_trns.png"
|
|
im = Image.open(in_file)
|
|
self.assertEqual(im.mode, mode)
|
|
self.assertEqual(im.info["transparency"], 255)
|
|
|
|
im_rgba = im.convert("RGBA")
|
|
self.assertEqual(im_rgba.getchannel("A").getcolors()[0][0], num_transparent)
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
test_im = Image.open(test_file)
|
|
self.assertEqual(test_im.mode, mode)
|
|
self.assertEqual(test_im.info["transparency"], 255)
|
|
self.assert_image_equal(im, test_im)
|
|
|
|
test_im_rgba = test_im.convert("RGBA")
|
|
self.assertEqual(
|
|
test_im_rgba.getchannel("A").getcolors()[0][0], num_transparent
|
|
)
|
|
|
|
def test_save_rgb_single_transparency(self):
|
|
in_file = "Tests/images/caption_6_33_22.png"
|
|
im = Image.open(in_file)
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
def test_load_verify(self):
|
|
# Check open/load/verify exception (@PIL150)
|
|
|
|
im = Image.open(TEST_PNG_FILE)
|
|
|
|
# Assert that there is no unclosed file warning
|
|
self.assert_warning(None, im.verify)
|
|
|
|
im = Image.open(TEST_PNG_FILE)
|
|
im.load()
|
|
self.assertRaises(RuntimeError, im.verify)
|
|
|
|
def test_verify_struct_error(self):
|
|
# Check open/load/verify exception (#1755)
|
|
|
|
# offsets to test, -10: breaks in i32() in read. (IOError)
|
|
# -13: breaks in crc, txt chunk.
|
|
# -14: malformed chunk
|
|
|
|
for offset in (-10, -13, -14):
|
|
with open(TEST_PNG_FILE, "rb") as f:
|
|
test_file = f.read()[:offset]
|
|
|
|
im = Image.open(BytesIO(test_file))
|
|
self.assertIsNotNone(im.fp)
|
|
self.assertRaises((IOError, SyntaxError), im.verify)
|
|
|
|
def test_verify_ignores_crc_error(self):
|
|
# check ignores crc errors in ancillary chunks
|
|
|
|
chunk_data = chunk(b"tEXt", b"spam")
|
|
broken_crc_chunk_data = chunk_data[:-1] + b"q" # break CRC
|
|
|
|
image_data = HEAD + broken_crc_chunk_data + TAIL
|
|
self.assertRaises(SyntaxError, PngImagePlugin.PngImageFile, BytesIO(image_data))
|
|
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = True
|
|
try:
|
|
im = load(image_data)
|
|
self.assertIsNotNone(im)
|
|
finally:
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = False
|
|
|
|
def test_verify_not_ignores_crc_error_in_required_chunk(self):
|
|
# check does not ignore crc errors in required chunks
|
|
|
|
image_data = MAGIC + IHDR[:-1] + b"q" + TAIL
|
|
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = True
|
|
try:
|
|
self.assertRaises(
|
|
SyntaxError, PngImagePlugin.PngImageFile, BytesIO(image_data)
|
|
)
|
|
finally:
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = False
|
|
|
|
def test_roundtrip_dpi(self):
|
|
# Check dpi roundtripping
|
|
|
|
im = Image.open(TEST_PNG_FILE)
|
|
|
|
im = roundtrip(im, dpi=(100, 100))
|
|
self.assertEqual(im.info["dpi"], (100, 100))
|
|
|
|
def test_load_dpi_rounding(self):
|
|
# Round up
|
|
im = Image.open(TEST_PNG_FILE)
|
|
self.assertEqual(im.info["dpi"], (96, 96))
|
|
|
|
# Round down
|
|
im = Image.open("Tests/images/icc_profile_none.png")
|
|
self.assertEqual(im.info["dpi"], (72, 72))
|
|
|
|
def test_save_dpi_rounding(self):
|
|
im = Image.open(TEST_PNG_FILE)
|
|
|
|
im = roundtrip(im, dpi=(72.2, 72.2))
|
|
self.assertEqual(im.info["dpi"], (72, 72))
|
|
|
|
im = roundtrip(im, dpi=(72.8, 72.8))
|
|
self.assertEqual(im.info["dpi"], (73, 73))
|
|
|
|
def test_roundtrip_text(self):
|
|
# Check text roundtripping
|
|
|
|
im = Image.open(TEST_PNG_FILE)
|
|
|
|
info = PngImagePlugin.PngInfo()
|
|
info.add_text("TXT", "VALUE")
|
|
info.add_text("ZIP", "VALUE", zip=True)
|
|
|
|
im = roundtrip(im, pnginfo=info)
|
|
self.assertEqual(im.info, {"TXT": "VALUE", "ZIP": "VALUE"})
|
|
self.assertEqual(im.text, {"TXT": "VALUE", "ZIP": "VALUE"})
|
|
|
|
def test_roundtrip_itxt(self):
|
|
# Check iTXt roundtripping
|
|
|
|
im = Image.new("RGB", (32, 32))
|
|
info = PngImagePlugin.PngInfo()
|
|
info.add_itxt("spam", "Eggs", "en", "Spam")
|
|
info.add_text("eggs", PngImagePlugin.iTXt("Spam", "en", "Eggs"), zip=True)
|
|
|
|
im = roundtrip(im, pnginfo=info)
|
|
self.assertEqual(im.info, {"spam": "Eggs", "eggs": "Spam"})
|
|
self.assertEqual(im.text, {"spam": "Eggs", "eggs": "Spam"})
|
|
self.assertEqual(im.text["spam"].lang, "en")
|
|
self.assertEqual(im.text["spam"].tkey, "Spam")
|
|
self.assertEqual(im.text["eggs"].lang, "en")
|
|
self.assertEqual(im.text["eggs"].tkey, "Eggs")
|
|
|
|
def test_nonunicode_text(self):
|
|
# Check so that non-Unicode text is saved as a tEXt rather than iTXt
|
|
|
|
im = Image.new("RGB", (32, 32))
|
|
info = PngImagePlugin.PngInfo()
|
|
info.add_text("Text", "Ascii")
|
|
im = roundtrip(im, pnginfo=info)
|
|
self.assertIsInstance(im.info["Text"], str)
|
|
|
|
def test_unicode_text(self):
|
|
# Check preservation of non-ASCII characters on Python 3
|
|
# This cannot really be meaningfully tested on Python 2,
|
|
# since it didn't preserve charsets to begin with.
|
|
|
|
def rt_text(value):
|
|
im = Image.new("RGB", (32, 32))
|
|
info = PngImagePlugin.PngInfo()
|
|
info.add_text("Text", value)
|
|
im = roundtrip(im, pnginfo=info)
|
|
self.assertEqual(im.info, {"Text": value})
|
|
|
|
if py3:
|
|
rt_text(" Aa" + chr(0xA0) + chr(0xC4) + chr(0xFF)) # Latin1
|
|
rt_text(chr(0x400) + chr(0x472) + chr(0x4FF)) # Cyrillic
|
|
# CJK:
|
|
rt_text(chr(0x4E00) + chr(0x66F0) + chr(0x9FBA) + chr(0x3042) + chr(0xAC00))
|
|
rt_text("A" + chr(0xC4) + chr(0x472) + chr(0x3042)) # Combined
|
|
|
|
def test_scary(self):
|
|
# Check reading of evil PNG file. For information, see:
|
|
# http://scary.beasts.org/security/CESA-2004-001.txt
|
|
# The first byte is removed from pngtest_bad.png
|
|
# to avoid classification as malware.
|
|
|
|
with open("Tests/images/pngtest_bad.png.bin", "rb") as fd:
|
|
data = b"\x89" + fd.read()
|
|
|
|
pngfile = BytesIO(data)
|
|
self.assertRaises(IOError, Image.open, pngfile)
|
|
|
|
def test_trns_rgb(self):
|
|
# Check writing and reading of tRNS chunks for RGB images.
|
|
# Independent file sample provided by Sebastian Spaeth.
|
|
|
|
test_file = "Tests/images/caption_6_33_22.png"
|
|
im = Image.open(test_file)
|
|
self.assertEqual(im.info["transparency"], (248, 248, 248))
|
|
|
|
# check saving transparency by default
|
|
im = roundtrip(im)
|
|
self.assertEqual(im.info["transparency"], (248, 248, 248))
|
|
|
|
im = roundtrip(im, transparency=(0, 1, 2))
|
|
self.assertEqual(im.info["transparency"], (0, 1, 2))
|
|
|
|
def test_trns_p(self):
|
|
# Check writing a transparency of 0, issue #528
|
|
im = hopper("P")
|
|
im.info["transparency"] = 0
|
|
|
|
f = self.tempfile("temp.png")
|
|
im.save(f)
|
|
|
|
im2 = Image.open(f)
|
|
self.assertIn("transparency", im2.info)
|
|
|
|
self.assert_image_equal(im2.convert("RGBA"), im.convert("RGBA"))
|
|
|
|
def test_trns_null(self):
|
|
# Check reading images with null tRNS value, issue #1239
|
|
test_file = "Tests/images/tRNS_null_1x1.png"
|
|
im = Image.open(test_file)
|
|
|
|
self.assertEqual(im.info["transparency"], 0)
|
|
|
|
def test_save_icc_profile(self):
|
|
im = Image.open("Tests/images/icc_profile_none.png")
|
|
self.assertIsNone(im.info["icc_profile"])
|
|
|
|
with_icc = Image.open("Tests/images/icc_profile.png")
|
|
expected_icc = with_icc.info["icc_profile"]
|
|
|
|
im = roundtrip(im, icc_profile=expected_icc)
|
|
self.assertEqual(im.info["icc_profile"], expected_icc)
|
|
|
|
def test_discard_icc_profile(self):
|
|
im = Image.open("Tests/images/icc_profile.png")
|
|
|
|
im = roundtrip(im, icc_profile=None)
|
|
self.assertNotIn("icc_profile", im.info)
|
|
|
|
def test_roundtrip_icc_profile(self):
|
|
im = Image.open("Tests/images/icc_profile.png")
|
|
expected_icc = im.info["icc_profile"]
|
|
|
|
im = roundtrip(im)
|
|
self.assertEqual(im.info["icc_profile"], expected_icc)
|
|
|
|
def test_roundtrip_no_icc_profile(self):
|
|
im = Image.open("Tests/images/icc_profile_none.png")
|
|
self.assertIsNone(im.info["icc_profile"])
|
|
|
|
im = roundtrip(im)
|
|
self.assertNotIn("icc_profile", im.info)
|
|
|
|
def test_repr_png(self):
|
|
im = hopper()
|
|
|
|
repr_png = Image.open(BytesIO(im._repr_png_()))
|
|
self.assertEqual(repr_png.format, "PNG")
|
|
self.assert_image_equal(im, repr_png)
|
|
|
|
def test_chunk_order(self):
|
|
im = Image.open("Tests/images/icc_profile.png")
|
|
test_file = self.tempfile("temp.png")
|
|
im.convert("P").save(test_file, dpi=(100, 100))
|
|
|
|
chunks = self.get_chunks(test_file)
|
|
|
|
# https://www.w3.org/TR/PNG/#5ChunkOrdering
|
|
# IHDR - shall be first
|
|
self.assertEqual(chunks.index(b"IHDR"), 0)
|
|
# PLTE - before first IDAT
|
|
self.assertLess(chunks.index(b"PLTE"), chunks.index(b"IDAT"))
|
|
# iCCP - before PLTE and IDAT
|
|
self.assertLess(chunks.index(b"iCCP"), chunks.index(b"PLTE"))
|
|
self.assertLess(chunks.index(b"iCCP"), chunks.index(b"IDAT"))
|
|
# tRNS - after PLTE, before IDAT
|
|
self.assertGreater(chunks.index(b"tRNS"), chunks.index(b"PLTE"))
|
|
self.assertLess(chunks.index(b"tRNS"), chunks.index(b"IDAT"))
|
|
# pHYs - before IDAT
|
|
self.assertLess(chunks.index(b"pHYs"), chunks.index(b"IDAT"))
|
|
|
|
def test_getchunks(self):
|
|
im = hopper()
|
|
|
|
chunks = PngImagePlugin.getchunks(im)
|
|
self.assertEqual(len(chunks), 3)
|
|
|
|
def test_textual_chunks_after_idat(self):
|
|
im = Image.open("Tests/images/hopper.png")
|
|
self.assertIn("comment", im.text.keys())
|
|
for k, v in {
|
|
"date:create": "2014-09-04T09:37:08+03:00",
|
|
"date:modify": "2014-09-04T09:37:08+03:00",
|
|
}.items():
|
|
self.assertEqual(im.text[k], v)
|
|
|
|
# Raises a SyntaxError in load_end
|
|
im = Image.open("Tests/images/broken_data_stream.png")
|
|
with self.assertRaises(IOError):
|
|
self.assertIsInstance(im.text, dict)
|
|
|
|
# Raises a UnicodeDecodeError in load_end
|
|
im = Image.open("Tests/images/truncated_image.png")
|
|
# The file is truncated
|
|
self.assertRaises(IOError, lambda: im.text)
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = True
|
|
self.assertIsInstance(im.text, dict)
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = False
|
|
|
|
# Raises an EOFError in load_end
|
|
im = Image.open("Tests/images/hopper_idat_after_image_end.png")
|
|
self.assertEqual(im.text, {"TXT": "VALUE", "ZIP": "VALUE"})
|
|
|
|
def test_exif(self):
|
|
im = Image.open("Tests/images/exif.png")
|
|
exif = im._getexif()
|
|
self.assertEqual(exif[274], 1)
|
|
|
|
def test_exif_save(self):
|
|
im = Image.open("Tests/images/exif.png")
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
reloaded = Image.open(test_file)
|
|
exif = reloaded._getexif()
|
|
self.assertEqual(exif[274], 1)
|
|
|
|
def test_exif_from_jpg(self):
|
|
im = Image.open("Tests/images/pil_sample_rgb.jpg")
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file)
|
|
|
|
reloaded = Image.open(test_file)
|
|
exif = reloaded._getexif()
|
|
self.assertEqual(exif[305], "Adobe Photoshop CS Macintosh")
|
|
|
|
def test_exif_argument(self):
|
|
im = Image.open(TEST_PNG_FILE)
|
|
|
|
test_file = self.tempfile("temp.png")
|
|
im.save(test_file, exif=b"exifstring")
|
|
|
|
reloaded = Image.open(test_file)
|
|
self.assertEqual(reloaded.info["exif"], b"Exif\x00\x00exifstring")
|
|
|
|
@unittest.skipUnless(
|
|
HAVE_WEBP and _webp.HAVE_WEBPANIM, "WebP support not installed with animation"
|
|
)
|
|
def test_apng(self):
|
|
im = Image.open("Tests/images/iss634.apng")
|
|
self.assertEqual(im.get_format_mimetype(), "image/apng")
|
|
|
|
# This also tests reading unknown PNG chunks (fcTL and fdAT) in load_end
|
|
expected = Image.open("Tests/images/iss634.webp")
|
|
self.assert_image_similar(im, expected, 0.23)
|
|
|
|
|
|
@unittest.skipIf(sys.platform.startswith("win32"), "requires Unix or macOS")
|
|
class TestTruncatedPngPLeaks(PillowLeakTestCase):
|
|
mem_limit = 2 * 1024 # max increase in K
|
|
iterations = 100 # Leak is 56k/iteration, this will leak 5.6megs
|
|
|
|
def setUp(self):
|
|
if "zip_encoder" not in codecs or "zip_decoder" not in codecs:
|
|
self.skipTest("zip/deflate support not available")
|
|
|
|
def test_leak_load(self):
|
|
with open("Tests/images/hopper.png", "rb") as f:
|
|
DATA = BytesIO(f.read(16 * 1024))
|
|
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = True
|
|
with Image.open(DATA) as im:
|
|
im.load()
|
|
|
|
def core():
|
|
with Image.open(DATA) as im:
|
|
im.load()
|
|
|
|
try:
|
|
self._test_leak(core)
|
|
finally:
|
|
ImageFile.LOAD_TRUNCATED_IMAGES = False
|