Updated to include the verbose flag from test_support
This commit is contained in:
parent
549ab711aa
commit
7a61f871a7
|
@ -1,209 +1,254 @@
|
|||
# Test audioop.
|
||||
import audioop
|
||||
from test_support import verbose
|
||||
|
||||
def gendata1():
|
||||
return '\0\1\2'
|
||||
return '\0\1\2'
|
||||
|
||||
def gendata2():
|
||||
if audioop.getsample('\0\1', 2, 0) == 1:
|
||||
return '\0\0\0\1\0\2'
|
||||
else:
|
||||
return '\0\0\1\0\2\0'
|
||||
if verbose:
|
||||
print 'getsample'
|
||||
if audioop.getsample('\0\1', 2, 0) == 1:
|
||||
return '\0\0\0\1\0\2'
|
||||
else:
|
||||
return '\0\0\1\0\2\0'
|
||||
|
||||
def gendata4():
|
||||
if audioop.getsample('\0\0\0\1', 4, 0) == 1:
|
||||
return '\0\0\0\0\0\0\0\1\0\0\0\2'
|
||||
else:
|
||||
return '\0\0\0\0\1\0\0\0\2\0\0\0'
|
||||
if verbose:
|
||||
print 'getsample'
|
||||
if audioop.getsample('\0\0\0\1', 4, 0) == 1:
|
||||
return '\0\0\0\0\0\0\0\1\0\0\0\2'
|
||||
else:
|
||||
return '\0\0\0\0\1\0\0\0\2\0\0\0'
|
||||
|
||||
def testmax(data):
|
||||
if audioop.max(data[0], 1) <> 2 or \
|
||||
audioop.max(data[1], 2) <> 2 or \
|
||||
audioop.max(data[2], 4) <> 2:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'max'
|
||||
if audioop.max(data[0], 1) <> 2 or \
|
||||
audioop.max(data[1], 2) <> 2 or \
|
||||
audioop.max(data[2], 4) <> 2:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testminmax(data):
|
||||
if audioop.minmax(data[0], 1) <> (0, 2) or \
|
||||
audioop.minmax(data[1], 2) <> (0, 2) or \
|
||||
audioop.minmax(data[2], 4) <> (0, 2):
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'minmax'
|
||||
if audioop.minmax(data[0], 1) <> (0, 2) or \
|
||||
audioop.minmax(data[1], 2) <> (0, 2) or \
|
||||
audioop.minmax(data[2], 4) <> (0, 2):
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testmaxpp(data):
|
||||
if audioop.maxpp(data[0], 1) <> 0 or \
|
||||
audioop.maxpp(data[1], 2) <> 0 or \
|
||||
audioop.maxpp(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'maxpp'
|
||||
if audioop.maxpp(data[0], 1) <> 0 or \
|
||||
audioop.maxpp(data[1], 2) <> 0 or \
|
||||
audioop.maxpp(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testavg(data):
|
||||
if audioop.avg(data[0], 1) <> 1 or \
|
||||
audioop.avg(data[1], 2) <> 1 or \
|
||||
audioop.avg(data[2], 4) <> 1:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'avg'
|
||||
if audioop.avg(data[0], 1) <> 1 or \
|
||||
audioop.avg(data[1], 2) <> 1 or \
|
||||
audioop.avg(data[2], 4) <> 1:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testavgpp(data):
|
||||
if audioop.avgpp(data[0], 1) <> 0 or \
|
||||
audioop.avgpp(data[1], 2) <> 0 or \
|
||||
audioop.avgpp(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'avgpp'
|
||||
if audioop.avgpp(data[0], 1) <> 0 or \
|
||||
audioop.avgpp(data[1], 2) <> 0 or \
|
||||
audioop.avgpp(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testrms(data):
|
||||
if audioop.rms(data[0], 1) <> 1 or \
|
||||
audioop.rms(data[1], 2) <> 1 or \
|
||||
audioop.rms(data[2], 4) <> 1:
|
||||
return 0
|
||||
return 1
|
||||
if audioop.rms(data[0], 1) <> 1 or \
|
||||
audioop.rms(data[1], 2) <> 1 or \
|
||||
audioop.rms(data[2], 4) <> 1:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testcross(data):
|
||||
if audioop.cross(data[0], 1) <> 0 or \
|
||||
audioop.cross(data[1], 2) <> 0 or \
|
||||
audioop.cross(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'cross'
|
||||
if audioop.cross(data[0], 1) <> 0 or \
|
||||
audioop.cross(data[1], 2) <> 0 or \
|
||||
audioop.cross(data[2], 4) <> 0:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testadd(data):
|
||||
data2 = []
|
||||
for d in data:
|
||||
str = ''
|
||||
for s in d:
|
||||
str = str + chr(ord(s)*2)
|
||||
data2.append(str)
|
||||
if audioop.add(data[0], data[0], 1) <> data2[0] or \
|
||||
audioop.add(data[1], data[1], 2) <> data2[1] or \
|
||||
audioop.add(data[2], data[2], 4) <> data2[2]:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'add'
|
||||
data2 = []
|
||||
for d in data:
|
||||
str = ''
|
||||
for s in d:
|
||||
str = str + chr(ord(s)*2)
|
||||
data2.append(str)
|
||||
if audioop.add(data[0], data[0], 1) <> data2[0] or \
|
||||
audioop.add(data[1], data[1], 2) <> data2[1] or \
|
||||
audioop.add(data[2], data[2], 4) <> data2[2]:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testbias(data):
|
||||
# Note: this test assumes that avg() works
|
||||
d1 = audioop.bias(data[0], 1, 100)
|
||||
d2 = audioop.bias(data[1], 2, 100)
|
||||
d4 = audioop.bias(data[2], 4, 100)
|
||||
if audioop.avg(d1, 1) <> 101 or \
|
||||
audioop.avg(d2, 2) <> 101 or \
|
||||
audioop.avg(d4, 4) <> 101:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'bias'
|
||||
# Note: this test assumes that avg() works
|
||||
d1 = audioop.bias(data[0], 1, 100)
|
||||
d2 = audioop.bias(data[1], 2, 100)
|
||||
d4 = audioop.bias(data[2], 4, 100)
|
||||
if audioop.avg(d1, 1) <> 101 or \
|
||||
audioop.avg(d2, 2) <> 101 or \
|
||||
audioop.avg(d4, 4) <> 101:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testlin2lin(data):
|
||||
# too simple: we test only the size
|
||||
for d1 in data:
|
||||
for d2 in data:
|
||||
got = len(d1)/3
|
||||
wtd = len(d2)/3
|
||||
if len(audioop.lin2lin(d1, got, wtd)) <> len(d2):
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'lin2lin'
|
||||
# too simple: we test only the size
|
||||
for d1 in data:
|
||||
for d2 in data:
|
||||
got = len(d1)/3
|
||||
wtd = len(d2)/3
|
||||
if len(audioop.lin2lin(d1, got, wtd)) <> len(d2):
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testadpcm2lin(data):
|
||||
# Very cursory test
|
||||
if audioop.adpcm2lin('\0\0', 1, None) <> ('\0\0\0\0', (0,0)):
|
||||
return 0
|
||||
return 1
|
||||
# Very cursory test
|
||||
if audioop.adpcm2lin('\0\0', 1, None) <> ('\0\0\0\0', (0,0)):
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testlin2adpcm(data):
|
||||
# Very cursory test
|
||||
if audioop.lin2adpcm('\0\0\0\0', 1, None) <> ('\0\0', (0,0)):
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'lin2adpcm'
|
||||
# Very cursory test
|
||||
if audioop.lin2adpcm('\0\0\0\0', 1, None) <> ('\0\0', (0,0)):
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testlin2ulaw(data):
|
||||
if audioop.lin2ulaw(data[0], 1) <> '\377\347\333' or \
|
||||
audioop.lin2ulaw(data[1], 2) <> '\377\377\377' or \
|
||||
audioop.lin2ulaw(data[2], 4) <> '\377\377\377':
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'lin2ulaw'
|
||||
if audioop.lin2ulaw(data[0], 1) <> '\377\347\333' or \
|
||||
audioop.lin2ulaw(data[1], 2) <> '\377\377\377' or \
|
||||
audioop.lin2ulaw(data[2], 4) <> '\377\377\377':
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testulaw2lin(data):
|
||||
# Cursory
|
||||
d = audioop.lin2ulaw(data[0], 1)
|
||||
if audioop.ulaw2lin(d, 1) <> data[0]:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'ulaw2lin'
|
||||
# Cursory
|
||||
d = audioop.lin2ulaw(data[0], 1)
|
||||
if audioop.ulaw2lin(d, 1) <> data[0]:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testmul(data):
|
||||
data2 = []
|
||||
for d in data:
|
||||
str = ''
|
||||
for s in d:
|
||||
str = str + chr(ord(s)*2)
|
||||
data2.append(str)
|
||||
if audioop.mul(data[0], 1, 2) <> data2[0] or \
|
||||
audioop.mul(data[1],2, 2) <> data2[1] or \
|
||||
audioop.mul(data[2], 4, 2) <> data2[2]:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'mul'
|
||||
data2 = []
|
||||
for d in data:
|
||||
str = ''
|
||||
for s in d:
|
||||
str = str + chr(ord(s)*2)
|
||||
data2.append(str)
|
||||
if audioop.mul(data[0], 1, 2) <> data2[0] or \
|
||||
audioop.mul(data[1],2, 2) <> data2[1] or \
|
||||
audioop.mul(data[2], 4, 2) <> data2[2]:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testreverse(data):
|
||||
if audioop.reverse(data[0], 1) <> '\2\1\0':
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'reverse'
|
||||
if audioop.reverse(data[0], 1) <> '\2\1\0':
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testtomono(data):
|
||||
data2 = ''
|
||||
for d in data[0]:
|
||||
data2 = data2 + d + d
|
||||
if audioop.tomono(data2, 1, 0.5, 0.5) <> data[0]:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'tomono'
|
||||
data2 = ''
|
||||
for d in data[0]:
|
||||
data2 = data2 + d + d
|
||||
if audioop.tomono(data2, 1, 0.5, 0.5) <> data[0]:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testtostereo(data):
|
||||
data2 = ''
|
||||
for d in data[0]:
|
||||
data2 = data2 + d + d
|
||||
if audioop.tostereo(data[0], 1, 1, 1) <> data2:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'tostereo'
|
||||
data2 = ''
|
||||
for d in data[0]:
|
||||
data2 = data2 + d + d
|
||||
if audioop.tostereo(data[0], 1, 1, 1) <> data2:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testfindfactor(data):
|
||||
if audioop.findfactor(data[1], data[1]) <> 1.0:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'findfactor'
|
||||
if audioop.findfactor(data[1], data[1]) <> 1.0:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testfindfit(data):
|
||||
if audioop.findfit(data[1], data[1]) <> (0, 1.0):
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'findfit'
|
||||
if audioop.findfit(data[1], data[1]) <> (0, 1.0):
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testfindmax(data):
|
||||
if audioop.findmax(data[1], 1) <> 2:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'findmax'
|
||||
if audioop.findmax(data[1], 1) <> 2:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testgetsample(data):
|
||||
for i in range(3):
|
||||
if audioop.getsample(data[0], 1, i) <> i or \
|
||||
audioop.getsample(data[1], 2, i) <> i or \
|
||||
audioop.getsample(data[2], 4, i) <> i:
|
||||
return 0
|
||||
return 1
|
||||
if verbose:
|
||||
print 'getsample'
|
||||
for i in range(3):
|
||||
if audioop.getsample(data[0], 1, i) <> i or \
|
||||
audioop.getsample(data[1], 2, i) <> i or \
|
||||
audioop.getsample(data[2], 4, i) <> i:
|
||||
return 0
|
||||
return 1
|
||||
|
||||
def testone(name, data):
|
||||
try:
|
||||
func = eval('test'+name)
|
||||
except NameError:
|
||||
print 'No test found for audioop.'+name+'()'
|
||||
return
|
||||
try:
|
||||
rv = func(data)
|
||||
except 'xx':
|
||||
print 'Test FAILED for audioop.'+name+'() (with an exception)'
|
||||
return
|
||||
if not rv:
|
||||
print 'Test FAILED for audioop.'+name+'()'
|
||||
try:
|
||||
func = eval('test'+name)
|
||||
except NameError:
|
||||
print 'No test found for audioop.'+name+'()'
|
||||
return
|
||||
try:
|
||||
rv = func(data)
|
||||
except 'xx':
|
||||
print 'Test FAILED for audioop.'+name+'() (with an exception)'
|
||||
return
|
||||
if not rv:
|
||||
print 'Test FAILED for audioop.'+name+'()'
|
||||
|
||||
def testall():
|
||||
data = [gendata1(), gendata2(), gendata4()]
|
||||
names = dir(audioop)
|
||||
# We know there is a routine 'add'
|
||||
routines = []
|
||||
for n in names:
|
||||
if type(eval('audioop.'+n)) == type(audioop.add):
|
||||
routines.append(n)
|
||||
for n in routines:
|
||||
testone(n, data)
|
||||
data = [gendata1(), gendata2(), gendata4()]
|
||||
names = dir(audioop)
|
||||
# We know there is a routine 'add'
|
||||
routines = []
|
||||
for n in names:
|
||||
if type(eval('audioop.'+n)) == type(audioop.add):
|
||||
routines.append(n)
|
||||
for n in routines:
|
||||
testone(n, data)
|
||||
testall()
|
||||
|
|
Loading…
Reference in New Issue