DAMASK_EICMD/lib/damask/test/test.py

509 lines
19 KiB
Python

# -*- coding: UTF-8 no BOM -*-
# $Id$
import os, sys, shlex, inspect
import subprocess,shutil,string
import logging, logging.config
import damask
from optparse import OptionParser
class Test():
'''
General class for testing.
Is sub-classed by the individual tests.
'''
variants = []
def __init__(self,test_description):
logger = logging.getLogger()
logger.setLevel(0)
fh = logging.FileHandler('test.log') # create file handler which logs even debug messages
fh.setLevel(logging.DEBUG)
full = logging.Formatter('%(asctime)s - %(levelname)s: \n%(message)s')
fh.setFormatter(full)
ch = logging.StreamHandler(stream=sys.stdout) # create console handler with a higher log level
ch.setLevel(logging.INFO)
# create formatter and add it to the handlers
plain = logging.Formatter('%(message)s')
ch.setFormatter(plain)
# add the handlers to the logger
logger.addHandler(fh)
logger.addHandler(ch)
logging.info('!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n\n' \
+'----------------------------------------------------------------\n' \
+'| '+test_description+'\n' \
+'----------------------------------------------------------------')
self.dirBase = os.path.dirname(os.path.realpath(sys.modules[self.__class__.__module__].__file__))
self.parser = OptionParser(
description = test_description+' (using class: $Id$)',
usage='./test.py [options]')
self.updateRequested = False
def execute(self):
'''
Run all variants and report first failure.
'''
if not self.testPossible(): return -1
self.clean()
self.prepareAll()
for variant in xrange(len(self.variants)):
try:
self.prepare(variant)
self.run(variant)
self.postprocess(variant)
if self.updateRequested: # update requested
self.update(variant)
elif not self.compare(variant): # no update, do comparison
return variant+1 # return culprit
except Exception as e :
logging.critical('\nWARNING:\n %s\n'%e)
return variant+1 # return culprit
return 0
def testPossible(self):
'''
Check if test is possible or not (e.g. no license available).
'''
return True
def clean(self):
'''
Delete directory tree containing current results.
'''
status = True
try:
shutil.rmtree(self.dirCurrent())
except:
logging.warning('removal of directory "%s" not possible...'%(self.dirCurrent()))
status = status and False
try:
os.mkdir(self.dirCurrent())
except:
logging.critical('creation of directory "%s" failed...'%(self.dirCurrent()))
status = status and False
return status
def prepareAll(self):
'''
Do all necessary preparations for the whole test
'''
return True
def prepare(self,variant):
'''
Do all necessary preparations for the run of each test variant
'''
return True
def run(self,variant):
'''
Execute the requested test variant.
'''
return True
def postprocess(self,variant):
'''
Perform post-processing of generated results for this test variant.
'''
return True
def compare(self,variant):
'''
Compare reference to current results.
'''
return True
def update(self,variant):
'''
Update reference with current results.
'''
logging.debug('Update not necessary')
return True
def dirReference(self):
'''
Directory containing reference results of the test.
'''
return os.path.normpath(os.path.join(self.dirBase,'reference/'))
def dirCurrent(self):
'''
Directory containing current results of the test.
'''
return os.path.normpath(os.path.join(self.dirBase,'current/'))
def dirProof(self):
'''
Directory containing human readable proof of correctness for the test.
'''
return os.path.normpath(os.path.join(self.dirBase,'proof/'))
def fileInRoot(self,dir,file):
'''
Path to a file in the root directory of DAMASK.
'''
return os.path.join(damask.Environment().rootDir(),dir,file)
def fileInReference(self,file):
'''
Path to a file in the refrence directory for the test.
'''
return os.path.join(self.dirReference(),file)
def fileInCurrent(self,file):
'''
Path to a file in the current results directory for the test.
'''
return os.path.join(self.dirCurrent(),file)
def fileInProof(self,file):
'''
Path to a file in the proof directory for the test.
'''
return os.path.join(self.dirProof(),file)
def copy(self, mapA, mapB,
A = [], B = []):
'''
copy list of files from (mapped) source to target.
mapA/B is one of self.fileInX.
'''
if not B or len(B) == 0: B = A
for source,target in zip(map(mapA,A),map(mapB,B)):
try:
shutil.copy2(source,target)
except:
logging.critical('error copying {} to {}'.format(source,target))
def copy_Reference2Current(self,sourcefiles=[],targetfiles=[]):
if len(targetfiles) == 0: targetfiles = sourcefiles
for i,file in enumerate(sourcefiles):
try:
shutil.copy2(self.fileInReference(file),self.fileInCurrent(targetfiles[i]))
except:
logging.critical('Reference2Current: Unable to copy file %s'%file)
def copy_Base2Current(self,sourceDir,sourcefiles=[],targetfiles=[]):
source=os.path.normpath(os.path.join(self.dirBase,'../../'+sourceDir))
if len(targetfiles) == 0: targetfiles = sourcefiles
for i,file in enumerate(sourcefiles):
try:
shutil.copy2(os.path.join(source,file),self.fileInCurrent(targetfiles[i]))
except:
logging.error(os.path.join(source,file))
logging.critical('Base2Current: Unable to copy file %s'%file)
def copy_Current2Reference(self,sourcefiles=[],targetfiles=[]):
if len(targetfiles) == 0: targetfiles = sourcefiles
for i,file in enumerate(sourcefiles):
try:
shutil.copy2(self.fileInCurrent(file),self.fileInReference(targetfiles[i]))
except:
logging.critical('Current2Reference: Unable to copy file %s'%file)
def copy_Proof2Current(self,sourcefiles=[],targetfiles=[]):
if len(targetfiles) == 0: targetfiles = sourcefiles
for i,file in enumerate(sourcefiles):
try:
shutil.copy2(self.fileInProof(file),self.fileInCurrent(targetfiles[i]))
except:
logging.critical('Proof2Current: Unable to copy file %s'%file)
def copy_Current2Current(self,sourcefiles=[],targetfiles=[]):
for i,file in enumerate(sourcefiles):
try:
shutil.copy2(self.fileInReference(file),self.fileInCurrent(targetfiles[i]))
except:
logging.critical('Current2Current: Unable to copy file %s'%file)
def execute_inCurrentDir(self,cmd,streamIn=None):
logging.info(cmd)
out,error = damask.util.execute(cmd,streamIn,self.dirCurrent())
logging.info(error)
logging.debug(out)
return out,error
def compare_Array(self,File1,File2):
import numpy as np
logging.info('comparing\n '+File1+'\n '+File2)
table1 = damask.ASCIItable(name=File1,readonly=True)
table1.head_read()
len1=len(table1.info)+2
table2 = damask.ASCIItable(name=File2,readonly=True)
table2.head_read()
len2=len(table2.info)+2
refArray = np.nan_to_num(np.genfromtxt(File1,missing_values='n/a',skip_header = len1,autostrip=True))
curArray = np.nan_to_num(np.genfromtxt(File2,missing_values='n/a',skip_header = len2,autostrip=True))
if len(curArray) == len(refArray):
refArrayNonZero = refArray[refArray.nonzero()]
curArray = curArray[refArray.nonzero()]
max_err=np.max(abs(refArrayNonZero[curArray.nonzero()]/curArray[curArray.nonzero()]-1.))
max_loc=np.argmax(abs(refArrayNonZero[curArray.nonzero()]/curArray[curArray.nonzero()]-1.))
refArrayNonZero = refArrayNonZero[curArray.nonzero()]
curArray = curArray[curArray.nonzero()]
print(' ********\n * maximum relative error %e for %e and %e\n ********'
%(max_err, refArrayNonZero[max_loc],curArray[max_loc]))
return max_err
else:
raise Exception('mismatch in array size to compare')
def compare_ArrayRefCur(self,ref,cur=''):
if cur =='': cur = ref
refName = self.fileInReference(ref)
curName = self.fileInCurrent(cur)
return self.compare_Array(refName,curName)
def compare_ArrayCurCur(self,cur0,cur1):
cur0Name = self.fileInCurrent(cur0)
cur1Name = self.fileInCurrent(cur1)
return self.compare_Array(cur0Name,cur1Name)
def compare_Table(self,headings0,file0,headings1,file1,normHeadings='',normType=None,
absoluteTolerance=False,perLine=False,skipLines=[]):
import numpy as np
logging.info('comparing ASCII Tables\n %s \n %s'%(file0,file1))
if normHeadings == '': normHeadings = headings0
if len(headings0) == len(headings1) == len(normHeadings): #check if comparison is possible and determine lenght of columns
dataLength = len(headings0)
length = [1 for i in xrange(dataLength)]
shape = [[] for i in xrange(dataLength)]
data = [[] for i in xrange(dataLength)]
maxError = [0.0 for i in xrange(dataLength)]
absTol = [absoluteTolerance for i in xrange(dataLength)]
column = [[1 for i in xrange(dataLength)] for j in xrange(2)]
norm = [[] for i in xrange(dataLength)]
normLength = [1 for i in xrange(dataLength)]
normShape = [[] for i in xrange(dataLength)]
normColumn = [1 for i in xrange(dataLength)]
for i in xrange(dataLength):
if headings0[i]['shape'] != headings1[i]['shape']:
raise Exception('shape mismatch when comparing %s with %s '%(headings0[i]['label'],headings1[i]['label']))
shape[i] = headings0[i]['shape']
for j in xrange(np.shape(shape[i])[0]):
length[i] *= shape[i][j]
normShape[i] = normHeadings[i]['shape']
for j in xrange(np.shape(normShape[i])[0]):
normLength[i] *= normShape[i][j]
else:
raise Exception('trying to compare %i with %i normed by %i data sets'%(len(headings0),len(headings1),len(normHeadings)))
table0 = damask.ASCIItable(name=file0,readonly=True)
table0.head_read()
table1 = damask.ASCIItable(name=file1,readonly=True)
table1.head_read()
for i in xrange(dataLength):
key0 = {True :'1_%s',
False:'%s' }[length[i]>1]%headings0[i]['label']
key1 = {True :'1_%s',
False:'%s' }[length[i]>1]%headings1[i]['label']
normKey = {True :'1_%s',
False:'%s' }[normLength[i]>1]%normHeadings[i]['label']
if key0 not in table0.labels:
raise Exception('column %s not found in 1. table...\n'%key0)
elif key1 not in table1.labels:
raise Exception('column %s not found in 2. table...\n'%key1)
elif normKey not in table0.labels:
raise Exception('column %s not found in 1. table...\n'%normKey)
else:
column[0][i] = table0.labels.index(key0) # remember columns of requested data
column[1][i] = table1.labels.index(key1) # remember columns of requested data in second column
normColumn[i] = table0.labels.index(normKey) # remember columns of requested data in second column
line0 = 0
while table0.data_read(): # read next data line of ASCII table
if line0 not in skipLines:
for i in xrange(dataLength):
myData = np.array(map(float,table0.data[column[0][i]:\
column[0][i]+length[i]]),'d')
normData = np.array(map(float,table0.data[normColumn[i]:\
normColumn[i]+normLength[i]]),'d')
data[i] = np.append(data[i],np.reshape(myData,shape[i]))
if normType == 'pInf':
norm[i] = np.append(norm[i],np.max(np.abs(normData)))
else:
norm[i] = np.append(norm[i],np.linalg.norm(np.reshape(normData,normShape[i]),normType))
line0 +=1
for i in xrange(dataLength):
if not perLine: norm[i] = [np.max(norm[i]) for j in xrange(line0-len(skipLines))]
data[i] = np.reshape(data[i],[line0-len(skipLines),length[i]])
if any(norm[i]) == 0.0 or absTol[i]:
norm[i] = [1.0 for j in xrange(line0-len(skipLines))]
absTol[i] = True
if perLine:
logging.warning('At least one norm of %s in 1. table is 0.0, using absolute tolerance'%headings0[i]['label'])
else:
logging.warning('Maximum norm of %s in 1. table is 0.0, using absolute tolerance'%headings0[i]['label'])
line1 = 0
while table1.data_read(): # read next data line of ASCII table
if line1 not in skipLines:
for i in xrange(dataLength):
myData = np.array(map(float,table1.data[column[1][i]:\
column[1][i]+length[i]]),'d')
maxError[i] = max(maxError[i],np.linalg.norm(np.reshape(myData-data[i][line1-len(skipLines),:],shape[i]))/
norm[i][line1-len(skipLines)])
line1 +=1
if (line0 != line1): raise Exception('found %s lines in 1. table and %s in 2. table'%(line0,line1))
logging.info(' ********')
for i in xrange(dataLength):
if absTol[i]:
logging.info(' * maximum absolute error %e for %s and %s'%(maxError[i],headings0[i]['label'],headings1[i]['label']))
else:
logging.info(' * maximum relative error %e for %s and %s'%(maxError[i],headings0[i]['label'],headings1[i]['label']))
logging.info(' ********')
return maxError
def compare_Tables(self,
files = [None,None], # list of file names
columns = [None], # list of list of column labels (per file)
rtol = 1e-5,
atol = 1e-8,
threshold = -1.0,
debug = False):
'''
compare tables with np.allclose
threshold can be used to ignore small values (a negative number disables this feature)
'''
import numpy as np
from collections import Iterable
if not (isinstance(files, Iterable) and not isinstance(files, str)): # check whether list of files is requested
files = [str(files)]
tables = [damask.ASCIItable(name = filename,readonly = True) for filename in files]
for table in tables:
table.head_read()
columns += [columns[0]]*(len(files)-len(columns)) # extend to same length as files
columns = columns[:len(files)] # truncate to same length as files
for i,column in enumerate(columns):
if column is None: columns[i] = tables[i].labels # if no column is given, read all
logging.info('comparing ASCIItables')
for i in xrange(len(columns)):
columns[i] = columns[0] if not columns[i] else \
([columns[i]] if not (isinstance(columns[i], Iterable) and not isinstance(columns[i], str)) else \
columns[i]
)
logging.info(files[i]+':'+','.join(columns[i]))
if len(files) < 2: return True # single table is always close to itself...
maximum = np.zeros(len(columns[0]),dtype='f')
data = []
for table,labels in zip(tables,columns):
table.data_readArray(labels)
data.append(table.data)
maximum += np.abs(table.data).max(axis=0)
table.close()
maximum /= len(tables)
maximum = np.where(maximum >0.0, maximum, 1)
for i in xrange(len(data)):
data[i] /= maximum
mask = np.zeros_like(table.data,dtype='bool')
for table in data:
mask |= np.where(np.abs(table)<threshold,True,False) # mask out (all) tiny values
allclose = True # start optimistic
for i in xrange(1,len(data)):
allclose &= np.allclose(np.where(mask,0.0,data[i-1]),
np.where(mask,0.0,data[i ]),rtol,atol) # accumulate "pessimism"
return allclose
def compare_TableRefCur(self,headingsRef,ref,headingsCur='',cur='',normHeadings='',normType=None,\
absoluteTolerance=False,perLine=False,skipLines=[]):
if cur == '': cur = ref
if headingsCur == '': headingsCur = headingsRef
refName = self.fileInReference(ref)
curName = self.fileInCurrent(cur)
return self.compare_Table(headingsRef,refName,headingsCur,curName,normHeadings,normType,
absoluteTolerance,perLine,skipLines)
def compare_TableCurCur(self,headingsCur0,Cur0,Cur1,headingsCur1='',normHeadings='',normType=None,\
absoluteTolerance=False,perLine=False,skipLines=[]):
if headingsCur1 == '': headingsCur1 = headingsCur0
cur0Name = self.fileInCurrent(Cur0)
cur1Name = self.fileInCurrent(Cur1)
return self.compare_Table(headingsCur0,cur0Name,headingsCur1,cur1Name,normHeadings,normType,
absoluteTolerance,perLine,skipLines)
def report_Success(self,culprit):
if culprit == 0:
logging.critical('%s passed.'%({False: 'The test',
True: 'All %i tests'%(len(self.variants))}[len(self.variants) > 1]))
logging.critical('\n!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n')
return 0
if culprit == -1:
logging.warning('Warning: Could not start test')
return 0
else:
logging.critical(' ********\n * Test %i failed...\n ********'%(culprit))
logging.critical('\n!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n')
return culprit