def test_typesize_is_set_correctly_with_custom_blosc_args():
a = np.array([1, 2, 3], dtype='uint8')
sink = CompressedMemorySink()
input_args = BloscArgs(clevel=9)
pack_ndarray(a, sink, blosc_args=input_args)
expected_args = BloscArgs(clevel=9, typesize=1)
nt.assert_equal(expected_args, sink.blosc_args)
def test_decode_blosc_header_deactivate_shuffle():
array_ = np.ones(16000, dtype=np.uint8)
blosc_args = BloscArgs()
blosc_args.shuffle = False
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {'versionlz': 1,
'version': 2,
'flags': 0, # no shuffle flag
'nbytes': len(array_),
'typesize': blosc_args.typesize}
header_slice = dict((k, header[k]) for k in expected.keys())
nt.assert_equal(expected, header_slice)
def test_decode_blosc_header_deactivate_shuffle():
array_ = np.ones(16000, dtype=np.uint8)
blosc_args = BloscArgs()
blosc_args.shuffle = False
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'version': 2,
'flags': 0, # no shuffle flag
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
header_slice = dict((k, header[k]) for k in expected.keys())
nt.assert_equal(expected, header_slice)
def test_decode_blosc_header_uncompressible_data():
array_ = np.asarray(np.random.randn(255),
dtype=np.float32).tostring()
blosc_args = BloscArgs()
blosc_args.shuffle = True
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {'versionlz': 1,
'blocksize': 1016,
'ctbytes': len(array_) + 16, # original + 16 header bytes
'version': 2,
'flags': 0x13, # 1 for shuffle 2 for non-compressed 4 for small blocksize
'nbytes': len(array_),
'typesize': blosc_args.typesize}
nt.assert_equal(expected, header)
def test_decode_blosc_header_uncompressible_data_dont_split_false():
array_ = np.asarray(np.random.randn(256), dtype=np.float32).tostring()
blosc_args = BloscArgs()
blosc_args.shuffle = True
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'version': 2,
'blocksize': 1024,
'ctbytes': len(array_) + 16, # original + 16 header bytes
'flags': 0x3, # 1 for shuffle 2 for non-compressed
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
nt.assert_equal(expected, header)
def test_invalid_format():
blosc_args = BloscArgs()
with create_tmp_files() as (tdir, in_file, out_file, dcmp_file):
create_array(1, in_file)
pack_file_to_file(in_file, out_file, blosc_args=blosc_args)
nt.assert_raises(FormatVersionMismatch, unpack_file_from_file,
out_file, dcmp_file)
def test_append_mix_shuffle():
orig, new, new_size, dcmp = prep_array_for_append()
# use the typesize from the file
# deactivate shuffle
# crank up the clevel to ensure compression happens, otherwise the flags
# will be screwed later on
blosc_args = BloscArgs(typesize=None, shuffle=False, clevel=9)
reset_append_fp(orig, new, new_size, blosc_args=blosc_args)
source = CompressedFPSource(orig)
sink = PlainFPSink(dcmp)
unpack(source, sink)
orig.seek(0)
dcmp.seek(0)
new.seek(0)
new_str = new.read()
dcmp_str = dcmp.read()
nt.assert_equal(len(dcmp_str), len(new_str * 2))
nt.assert_equal(dcmp_str, new_str * 2)
# now get the first and the last chunk and check that the shuffle doesn't
# match
bloscpack_header, offsets = reset_read_beginning(orig)[0:4:3]
orig.seek(offsets[0])
checksum_impl = CHECKSUMS_LOOKUP[bloscpack_header['checksum']]
compressed_zero, blosc_header_zero, digest = \
_read_compressed_chunk_fp(orig, checksum_impl)
decompressed_zero = blosc.decompress(compressed_zero)
orig.seek(offsets[-1])
compressed_last, blosc_header_last, digest = \
_read_compressed_chunk_fp(orig, checksum_impl)
decompressed_last = blosc.decompress(compressed_last)
# first chunk has shuffle active
nt.assert_equal(blosc_header_zero['flags'], 1)
# last chunk doesn't
nt.assert_equal(blosc_header_last['flags'], 0)
def test_mixing_clevel():
# the first set of chunks has max compression
blosc_args = BloscArgs(clevel=9)
orig, new, new_size, dcmp = prep_array_for_append()
# get the original size
orig.seek(0, 2)
orig_size = orig.tell()
orig.seek(0)
# get a backup of the settings
bloscpack_header, metadata, metadata_header, offsets = \
reset_read_beginning(orig)
# compressed size of the last chunk, including checksum
last_chunk_compressed_size = orig_size - offsets[-1]
# do append
# use the typesize from the file and
# make the second set of chunks have no compression
blosc_args = BloscArgs(typesize=None, clevel=0)
nchunks = append_fp(orig, new, new_size, blosc_args=blosc_args)
# get the final size
orig.seek(0, 2)
final_size = orig.tell()
orig.seek(0)
# the original file minus the compressed size of the last chunk
discounted_orig_size = orig_size - last_chunk_compressed_size
# size of the appended data
# * raw new size, since we have no compression
# * uncompressed size of the last chunk
# * nchunks + 1 times the blosc and checksum overhead
appended_size = new_size + bloscpack_header['last_chunk'] + (nchunks +
1) * (16 + 4)
# final size should be original plus appended data
nt.assert_equal(final_size, appended_size + discounted_orig_size)
# check by unpacking
source = CompressedFPSource(orig)
sink = PlainFPSink(dcmp)
unpack(source, sink)
dcmp.seek(0)
new.seek(0)
new_str = new.read()
dcmp_str = dcmp.read()
nt.assert_equal(len(dcmp_str), len(new_str * 2))
nt.assert_equal(dcmp_str, new_str * 2)
def test_decode_blosc_header():
array_ = np.linspace(0, 100, 2e4).tostring()
# basic test case
blosc_args = BloscArgs()
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'blocksize': 160000,
'ctbytes': len(compressed),
'version': 2,
'flags': 1,
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
nt.assert_equal(expected, header)
# deactivate shuffle
blosc_args.shuffle = False
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'blocksize': 160000,
'ctbytes': len(compressed),
'version': 2,
'flags': 0, # no shuffle flag
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
nt.assert_equal(expected, header)
# uncompressible data
array_ = np.asarray(np.random.randn(23), dtype=np.float32).tostring()
blosc_args.shuffle = True
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'blocksize': 88,
'ctbytes': len(array_) + 16, # original + 16 header bytes
'version': 2,
'flags': 3, # 1 for shuffle 2 for non-compressed
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
nt.assert_equal(expected, header)
def test_decode_blosc_header():
array_ = np.linspace(0, 100, 2e4).tostring()
# basic test case
blosc_args = BloscArgs()
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {'versionlz': 1,
'blocksize': 160000,
'ctbytes': len(compressed),
'version': 2,
'flags': 1,
'nbytes': len(array_),
'typesize': blosc_args.typesize}
nt.assert_equal(expected, header)
# deactivate shuffle
blosc_args.shuffle = False
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {'versionlz': 1,
'blocksize': 160000,
'ctbytes': len(compressed),
'version': 2,
'flags': 0, # no shuffle flag
'nbytes': len(array_),
'typesize': blosc_args.typesize}
nt.assert_equal(expected, header)
# uncompressible data
array_ = np.asarray(np.random.randn(23),
dtype=np.float32).tostring()
blosc_args.shuffle = True
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {'versionlz': 1,
'blocksize': 88,
'ctbytes': len(array_) + 16, # original + 16 header bytes
'version': 2,
'flags': 3, # 1 for shuffle 2 for non-compressed
'nbytes': len(array_),
'typesize': blosc_args.typesize}
nt.assert_equal(expected, header)
def test_append_mix_shuffle():
orig, new, new_size, dcmp = prep_array_for_append()
# use the typesize from the file
# deactivate shuffle
# crank up the clevel to ensure compression happens, otherwise the flags
# will be screwed later on
blosc_args = BloscArgs(typesize=None, shuffle=False, clevel=9)
# need to create something that will be compressible even without shuffle,
# the linspace used in 'new' doesn't work anymore as of python-blosc 1.6.1
to_append = np.zeros(int(2e6))
to_append_fp = StringIO()
to_append_fp.write(to_append.tostring())
to_append_fp_size = to_append_fp.tell()
to_append_fp.seek(0)
# now do the append
reset_append_fp(orig,
to_append_fp,
to_append_fp_size,
blosc_args=blosc_args)
# decompress 'orig' so that we can examine it
source = CompressedFPSource(orig)
sink = PlainFPSink(dcmp)
unpack(source, sink)
orig.seek(0)
dcmp.seek(0)
new.seek(0)
new_str = new.read()
dcmp_str = dcmp.read()
# now sanity check the length and content of the decompressed
nt.assert_equal(len(dcmp_str), len(new_str) + to_append_fp_size)
nt.assert_equal(dcmp_str, new_str + to_append.tostring())
# now get the first and the last chunk and check that the shuffle doesn't
# match
bloscpack_header, offsets = reset_read_beginning(orig)[0:4:3]
orig.seek(offsets[0])
checksum_impl = CHECKSUMS_LOOKUP[bloscpack_header['checksum']]
compressed_zero, blosc_header_zero, digest = \
_read_compressed_chunk_fp(orig, checksum_impl)
decompressed_zero = blosc.decompress(compressed_zero)
orig.seek(offsets[-1])
compressed_last, blosc_header_last, digest = \
_read_compressed_chunk_fp(orig, checksum_impl)
decompressed_last = blosc.decompress(compressed_last)
# first chunk has shuffle active
nt.assert_equal(blosc_header_zero['flags'], 1)
# last chunk doesn't
nt.assert_equal(blosc_header_last['flags'], 0)
def test_alternate_cname():
for cname, int_id in [
('blosclz', 0),
('lz4', 1),
('lz4hc', 1),
('snappy', 2),
('zlib', 3),
]:
blosc_args = BloscArgs(cname=cname)
array_ = np.linspace(0, 1, 2e6)
sink = CompressedMemorySink()
pack_ndarray(array_, sink, blosc_args=blosc_args)
blosc_header = decode_blosc_header(sink.chunks[0])
yield nt.assert_equal, blosc_header['flags'] >> 5, int_id
def test_decode_blosc_header_basic():
array_ = np.linspace(0, 100, int(2e4)).tostring()
blosc_args = BloscArgs()
compressed = blosc.compress(array_, **blosc_args)
header = decode_blosc_header(compressed)
expected = {
'versionlz': 1,
'version': 2,
'flags': 1,
'nbytes': len(array_),
'typesize': blosc_args.typesize
}
header_slice = dict((k, header[k]) for k in expected.keys())
nt.assert_equal(expected, header_slice)
def prep_array_for_append(blosc_args=BloscArgs(),
bloscpack_args=BloscpackArgs()):
orig, new, dcmp = StringIO(), StringIO(), StringIO()
create_array_fp(1, new)
new_size = new.tell()
new.seek(0)
chunking = calculate_nchunks(new_size)
source = PlainFPSource(new)
sink = CompressedFPSink(orig)
pack(source, sink, *chunking,
blosc_args=blosc_args,
bloscpack_args=bloscpack_args)
orig.seek(0)
new.seek(0)
return orig, new, new_size, dcmp
def test_typesize_is_set_correctly_with_default_blosc_args():
a = np.array([1, 2, 3], dtype='uint8')
sink = CompressedMemorySink()
pack_ndarray(a, sink)
expected_args = BloscArgs(typesize=1)
nt.assert_equal(expected_args, sink.blosc_args)
def test_init(self):
blosc_args = BloscArgs()
self.assertEqual(DEFAULT_TYPESIZE, blosc_args.typesize)
self.assertEqual(DEFAULT_CLEVEL, blosc_args.clevel)
self.assertEqual(DEFAULT_SHUFFLE, blosc_args.shuffle)
self.assertEqual(DEFAULT_CNAME, blosc_args.cname)