mirror of
https://gitlab.gnome.org/GNOME/libxml2.git
synced 2024-12-27 03:21:26 +03:00
3403adde32
* regressions.py, regressions.xml: new files for running regression tests under Python. Not yet complete, but should provide good testing under both Linux and Windows. * testHTML.c, testSAX.c, xmllint.c: changed the 'fopen' used for --push testing to include the 'rb' param when compiled under Windows.
351 lines
12 KiB
Python
Executable File
351 lines
12 KiB
Python
Executable File
#!/usr/bin/python -u
|
|
import glob, os, string, sys, thread, time
|
|
# import difflib
|
|
import libxml2
|
|
|
|
###
|
|
#
|
|
# This is a "Work in Progress" attempt at a python script to run the
|
|
# various regression tests. The rationale for this is that it should be
|
|
# possible to run this on most major platforms, including those (such as
|
|
# Windows) which don't support gnu Make.
|
|
#
|
|
# The script is driven by a parameter file which defines the various tests
|
|
# to be run, together with the unique settings for each of these tests. A
|
|
# script for Linux is included (regressions.xml), with comments indicating
|
|
# the significance of the various parameters. To run the tests under Windows,
|
|
# edit regressions.xml and remove the comment around the default parameter
|
|
# "<execpath>" (i.e. make it point to the location of the binary executables).
|
|
#
|
|
# Note that this current version requires the Python bindings for libxml2 to
|
|
# have been previously installed and accessible
|
|
#
|
|
# See Copyright for the status of this software.
|
|
# William Brack (wbrack@mmm.com.hk)
|
|
#
|
|
###
|
|
defaultParams = {} # will be used as a dictionary to hold the parsed params
|
|
|
|
# This routine is used for comparing the expected stdout / stdin with the results.
|
|
# The expected data has already been read in; the result is a file descriptor.
|
|
# Within the two sets of data, lines may begin with a path string. If so, the
|
|
# code "relativises" it by removing the path component. The first argument is a
|
|
# list already read in by a separate thread; the second is a file descriptor.
|
|
# The two 'base' arguments are to let me "relativise" the results files, allowing
|
|
# the script to be run from any directory.
|
|
def compFiles(res, expected, base1, base2):
|
|
l1 = len(base1)
|
|
exp = expected.readlines()
|
|
expected.close()
|
|
# the "relativisation" is done here
|
|
for i in range(len(res)):
|
|
j = string.find(res[i],base1)
|
|
if (j == 0) or ((j == 2) and (res[i][0:2] == './')):
|
|
col = string.find(res[i],':')
|
|
if col > 0:
|
|
start = string.rfind(res[i][:col], '/')
|
|
if start > 0:
|
|
res[i] = res[i][start+1:]
|
|
|
|
for i in range(len(exp)):
|
|
j = string.find(exp[i],base2)
|
|
if (j == 0) or ((j == 2) and (exp[i][0:2] == './')):
|
|
col = string.find(exp[i],':')
|
|
if col > 0:
|
|
start = string.rfind(exp[i][:col], '/')
|
|
if start > 0:
|
|
exp[i] = exp[i][start+1:]
|
|
|
|
ret = 0
|
|
# ideally we would like to use difflib functions here to do a
|
|
# nice comparison of the two sets. Unfortunately, during testing
|
|
# (using python 2.3.3 and 2.3.4) the following code went into
|
|
# a dead loop under windows. I'll pursue this later.
|
|
# diff = difflib.ndiff(res, exp)
|
|
# diff = list(diff)
|
|
# for line in diff:
|
|
# if line[:2] != ' ':
|
|
# print string.strip(line)
|
|
# ret = -1
|
|
|
|
# the following simple compare is fine for when the two data sets
|
|
# (actual result vs. expected result) are equal, which should be true for
|
|
# us. Unfortunately, if the test fails it's not nice at all.
|
|
rl = len(res)
|
|
el = len(exp)
|
|
if el != rl:
|
|
print 'Length of expected is %d, result is %d' % (el, rl)
|
|
ret = -1
|
|
for i in range(min(el, rl)):
|
|
if string.strip(res[i]) != string.strip(exp[i]):
|
|
print '+:%s-:%s' % (res[i], exp[i])
|
|
ret = -1
|
|
if el > rl:
|
|
for i in range(rl, el):
|
|
print '-:%s' % exp[i]
|
|
ret = -1
|
|
elif rl > el:
|
|
for i in range (el, rl):
|
|
print '+:%s' % res[i]
|
|
ret = -1
|
|
return ret
|
|
|
|
# Separate threads to handle stdout and stderr are created to run this function
|
|
def readPfile(file, list, flag):
|
|
data = file.readlines() # no call by reference, so I cheat
|
|
for l in data:
|
|
list.append(l)
|
|
file.close()
|
|
flag.append('ok')
|
|
|
|
# This routine runs the test program (e.g. xmllint)
|
|
def runOneTest(testDescription, filename, inbase, errbase):
|
|
if 'execpath' in testDescription:
|
|
dir = testDescription['execpath'] + '/'
|
|
else:
|
|
dir = ''
|
|
cmd = os.path.abspath(dir + testDescription['testprog'])
|
|
if 'flag' in testDescription:
|
|
for f in string.split(testDescription['flag']):
|
|
cmd += ' ' + f
|
|
if 'stdin' not in testDescription:
|
|
cmd += ' ' + inbase + filename
|
|
if 'extarg' in testDescription:
|
|
cmd += ' ' + testDescription['extarg']
|
|
|
|
noResult = 0
|
|
expout = None
|
|
if 'resext' in testDescription:
|
|
if testDescription['resext'] == 'None':
|
|
noResult = 1
|
|
else:
|
|
ext = '.' + testDescription['resext']
|
|
else:
|
|
ext = ''
|
|
if not noResult:
|
|
try:
|
|
fname = errbase + filename + ext
|
|
expout = open(fname, 'rt')
|
|
except:
|
|
print "Can't open result file %s - bypassing test" % fname
|
|
return
|
|
|
|
noErrors = 0
|
|
if 'reserrext' in testDescription:
|
|
if testDescription['reserrext'] == 'None':
|
|
noErrors = 1
|
|
else:
|
|
if len(testDescription['reserrext'])>0:
|
|
ext = '.' + testDescription['reserrext']
|
|
else:
|
|
ext = ''
|
|
else:
|
|
ext = ''
|
|
if not noErrors:
|
|
try:
|
|
fname = errbase + filename + ext
|
|
experr = open(fname, 'rt')
|
|
except:
|
|
experr = None
|
|
else:
|
|
experr = None
|
|
|
|
pin, pout, perr = os.popen3(cmd)
|
|
if 'stdin' in testDescription:
|
|
infile = open(inbase + filename, 'rt')
|
|
pin.writelines(infile.readlines())
|
|
infile.close()
|
|
pin.close()
|
|
|
|
# popen is great fun, but can lead to the old "deadly embrace", because
|
|
# synchronizing the writing (by the task being run) of stdout and stderr
|
|
# with respect to the reading (by this task) is basically impossible. I
|
|
# tried several ways to cheat, but the only way I have found which works
|
|
# is to do a *very* elementary multi-threading approach. We can only hope
|
|
# that Python threads are implemented on the target system (it's okay for
|
|
# Linux and Windows)
|
|
|
|
th1Flag = [] # flags to show when threads finish
|
|
th2Flag = []
|
|
outfile = [] # lists to contain the pipe data
|
|
errfile = []
|
|
th1 = thread.start_new_thread(readPfile, (pout, outfile, th1Flag))
|
|
th2 = thread.start_new_thread(readPfile, (perr, errfile, th2Flag))
|
|
while (len(th1Flag)==0) or (len(th2Flag)==0):
|
|
time.sleep(0.001)
|
|
if not noResult:
|
|
ret = compFiles(outfile, expout, inbase, 'test/')
|
|
if ret != 0:
|
|
print 'trouble with %s' % cmd
|
|
else:
|
|
if len(outfile) != 0:
|
|
for l in outfile:
|
|
print l
|
|
print 'trouble with %s' % cmd
|
|
if experr != None:
|
|
ret = compFiles(errfile, experr, inbase, 'test/')
|
|
if ret != 0:
|
|
print 'trouble with %s' % cmd
|
|
else:
|
|
if not noErrors:
|
|
if len(errfile) != 0:
|
|
for l in errfile:
|
|
print l
|
|
print 'trouble with %s' % cmd
|
|
|
|
if 'stdin' not in testDescription:
|
|
pin.close()
|
|
|
|
# This routine is called by the parameter decoding routine whenever the end of a
|
|
# 'test' section is encountered. Depending upon file globbing, a large number of
|
|
# individual tests may be run.
|
|
def runTest(description):
|
|
testDescription = defaultParams.copy() # set defaults
|
|
testDescription.update(description) # override with current ent
|
|
if 'testname' in testDescription:
|
|
print "## %s" % testDescription['testname']
|
|
if not 'file' in testDescription:
|
|
print "No file specified - can't run this test!"
|
|
return
|
|
# Set up the source and results directory paths from the decoded params
|
|
dir = ''
|
|
if 'srcdir' in testDescription:
|
|
dir += testDescription['srcdir'] + '/'
|
|
if 'srcsub' in testDescription:
|
|
dir += testDescription['srcsub'] + '/'
|
|
|
|
rdir = ''
|
|
if 'resdir' in testDescription:
|
|
rdir += testDescription['resdir'] + '/'
|
|
if 'ressub' in testDescription:
|
|
rdir += testDescription['ressub'] + '/'
|
|
|
|
testFiles = glob.glob(os.path.abspath(dir + testDescription['file']))
|
|
if testFiles == []:
|
|
print "No files result from '%s'" % testDescription['file']
|
|
return
|
|
|
|
# Some test programs just don't work (yet). For now we exclude them.
|
|
count = 0
|
|
excl = []
|
|
if 'exclfile' in testDescription:
|
|
for f in string.split(testDescription['exclfile']):
|
|
glb = glob.glob(dir + f)
|
|
for g in glb:
|
|
excl.append(os.path.abspath(g))
|
|
|
|
# Run the specified test program
|
|
for f in testFiles:
|
|
if not os.path.isdir(f):
|
|
if f not in excl:
|
|
count = count + 1
|
|
runOneTest(testDescription, os.path.basename(f), dir, rdir)
|
|
|
|
#
|
|
# The following classes are used with the xmlreader interface to interpret the
|
|
# parameter file. Once a test section has been identified, runTest is called
|
|
# with a dictionary containing the parsed results of the interpretation.
|
|
#
|
|
|
|
class testDefaults:
|
|
curText = '' # accumulates text content of parameter
|
|
|
|
def addToDict(self, key):
|
|
txt = string.strip(self.curText)
|
|
# if txt == '':
|
|
# return
|
|
if key not in defaultParams:
|
|
defaultParams[key] = txt
|
|
else:
|
|
defaultParams[key] += ' ' + txt
|
|
|
|
def processNode(self, reader, curClass):
|
|
if reader.Depth() == 2:
|
|
if reader.NodeType() == 1:
|
|
self.curText = '' # clear the working variable
|
|
elif reader.NodeType() == 15:
|
|
if (reader.Name() != '#text') and (reader.Name() != '#comment'):
|
|
self.addToDict(reader.Name())
|
|
elif reader.Depth() == 3:
|
|
if reader.Name() == '#text':
|
|
self.curText += reader.Value()
|
|
|
|
elif reader.NodeType() == 15: # end of element
|
|
print "Defaults have been set to:"
|
|
for k in defaultParams.keys():
|
|
print " %s : '%s'" % (k, defaultParams[k])
|
|
curClass = rootClass()
|
|
return curClass
|
|
|
|
|
|
class testClass:
|
|
def __init__(self):
|
|
self.testParams = {} # start with an empty set of params
|
|
self.curText = '' # and empty text
|
|
|
|
def addToDict(self, key):
|
|
data = string.strip(self.curText)
|
|
if key not in self.testParams:
|
|
self.testParams[key] = data
|
|
else:
|
|
if self.testParams[key] != '':
|
|
data = ' ' + data
|
|
self.testParams[key] += data
|
|
|
|
def processNode(self, reader, curClass):
|
|
if reader.Depth() == 2:
|
|
if reader.NodeType() == 1:
|
|
self.curText = '' # clear the working variable
|
|
if reader.Name() not in self.testParams:
|
|
self.testParams[reader.Name()] = ''
|
|
elif reader.NodeType() == 15:
|
|
if (reader.Name() != '#text') and (reader.Name() != '#comment'):
|
|
self.addToDict(reader.Name())
|
|
elif reader.Depth() == 3:
|
|
if reader.Name() == '#text':
|
|
self.curText += reader.Value()
|
|
|
|
elif reader.NodeType() == 15: # end of element
|
|
runTest(self.testParams)
|
|
curClass = rootClass()
|
|
return curClass
|
|
|
|
|
|
class rootClass:
|
|
def processNode(self, reader, curClass):
|
|
if reader.Depth() == 0:
|
|
return curClass
|
|
if reader.Depth() != 1:
|
|
print "Unexpected junk: Level %d, type %d, name %s" % (
|
|
reader.Depth(), reader.NodeType(), reader.Name())
|
|
return curClass
|
|
if reader.Name() == 'test':
|
|
curClass = testClass()
|
|
curClass.testParams = {}
|
|
elif reader.Name() == 'defaults':
|
|
curClass = testDefaults()
|
|
return curClass
|
|
|
|
def streamFile(filename):
|
|
try:
|
|
reader = libxml2.newTextReaderFilename(filename)
|
|
except:
|
|
print "unable to open %s" % (filename)
|
|
return
|
|
|
|
curClass = rootClass()
|
|
ret = reader.Read()
|
|
while ret == 1:
|
|
curClass = curClass.processNode(reader, curClass)
|
|
ret = reader.Read()
|
|
|
|
if ret != 0:
|
|
print "%s : failed to parse" % (filename)
|
|
|
|
# OK, we're finished with all the routines. Now for the main program:-
|
|
if len(sys.argv) != 2:
|
|
print "Usage: maketest {filename}"
|
|
sys.exit(-1)
|
|
|
|
streamFile(sys.argv[1])
|