can now deal with arbitrarily long lines for microstructure definition. useful for 2D ASCII representation of grain morphology in the geom file.

additional internal code reworking.
This commit is contained in:
Philip Eisenlohr 2011-10-11 17:25:22 +00:00
parent 8edb59d50c
commit ffc6442098
1 changed files with 72 additions and 51 deletions

View File

@ -1,7 +1,7 @@
#!/usr/bin/env python
# -*- coding: UTF-8 no BOM -*-
import os,sys,string,numpy
import os,sys,string,re,numpy
from optparse import OptionParser, OptionGroup, Option, SUPPRESS_HELP
# -----------------------------
@ -65,7 +65,7 @@ def vtk_writeASCII_mesh(dim,res,data):
cmds = [\
'# vtk DataFile Version 3.1',
string.replace('powered by $Id$','\n','\\n')
string.replace('powered by $Id$','\n','\\n'),
'ASCII',
'DATASET RECTILINEAR_GRID',
'DIMENSIONS %i %i %i'%(res[0]+1,res[1]+1,res[2]+1),
@ -91,68 +91,89 @@ def vtk_writeASCII_mesh(dim,res,data):
# ----------------------- MAIN -------------------------------
mapping = {
'resolution': {'a':0,'b':1,'c':2},
'dimension': {'x':0,'y':1,'z':2},
identifiers = {
'resolution': ['a','b','c'],
'dimension': ['x','y','z'],
}
mappings = {
'resolution': lambda x: int(x),
'dimension': lambda x: float(x),
}
parser = OptionParser(option_class=extendedOption, usage='%prog geomfile[s]', description = """
Produce VTK point file from geom data
parser = OptionParser(option_class=extendedOption, usage='%prog [geomfile[s]]', description = """
Produce VTK rectilinear mesh of texture data from geom description
""" + string.replace('$Id$','\n','\\n')
)
(options, args) = parser.parse_args()
(options, filenames) = parser.parse_args()
for filename in args:
if not os.path.exists(filename):
continue
# ------------------------------------------ setup file handles ---------------------------------------
print filename
file = open(filename)
files = []
if filenames == []:
files.append({'name':'STDIN', 'input':sys.stdin})
else:
for name in filenames:
if os.path.exists(name):
files.append({'name':name, 'input':open(name)})
res = [0,0,0]
resolutionData = file.readline().split()
if resolutionData[0].lower() != 'resolution':
sys.stderr.write('no resolution info found...\n')
continue
# ------------------------------------------ loop over input files ---------------------------------------
for file in files:
if file['name'] != 'STDIN': print file['name']
# get labels by either read the first row, or - if keyword header is present - the last line of the header
firstline = file['input'].readline()
m = re.search('(\d+)\s*head', firstline.lower())
if m:
headerlines = int(m.group(1))
headers = [file['input'].readline() for i in range(headerlines)]
else:
for i in range(1,6,2):
if resolutionData[i] not in mapping['resolution']:
sys.stderr.write('corrupt resolution info...\n')
continue
else:
res[mapping['resolution'][resolutionData[i]]] = int(resolutionData[i+1])
N = res[0]*res[1]*res[2]
headerlines = 1
headers = firstline
dim = [0.0,0.0,0.0]
dimensionData = file.readline().split()
if dimensionData[0].lower() != 'dimension':
sys.stderr.write('no dimension info found...\n')
continue
else:
for i in range(1,6,2):
if dimensionData[i] not in mapping['dimension']:
sys.stderr.write('corrupt dimension info...\n')
continue
else:
dim[mapping['dimension'][dimensionData[i]]] = float(dimensionData[i+1])
content = file['input'].readlines()
file['input'].close()
print 'dimension',dim
resolution = [0,0,0]
dimension = [0.0,0.0,0.0]
for header in headers:
headitems = header.split()
if headitems[0] == 'resolution': # located resolution entry
for i in xrange(3):
resolution[i] = mappings['resolution'](headitems[headitems.index(identifiers['resolution'][i])+1])
if headitems[0] == 'dimension': # located dimension entry
for i in xrange(3):
dimension[i] = mappings['dimension'](headitems[headitems.index(identifiers['dimension'][i])+1])
file.readline() # skip homogenization
if resolution == [0,0,0]:
print 'no resolution info found.'
sys.exit(1)
if dimension == [0.0,0.0,0.0]:
print 'no dimension info found.'
sys.exit(1)
print 'resolution: %s'%(' x '.join(map(str,resolution)))
print 'dimension: %s'%(' x '.join(map(str,dimension)))
data = {'scalar':{'texture':numpy.zeros(resolution,'i')}}
i = 0
for line in content:
for item in map(int,line.split()):
data['scalar']['texture'][i%resolution[0],(i/resolution[0])%resolution[1],i/resolution[0]/resolution[1]] = item
i += 1
data = {'scalar':{'texture':numpy.zeros(N,'i').reshape(res[2],res[1],res[0]).T}}
for k in range(res[2]):
for j in range(res[1]):
for i in range(res[0]):
data['scalar']['texture'][i,j,k] = int(file.readline().split()[0])
out = {}
out['mesh'] = vtk_writeASCII_mesh(dim,res,data)
out['mesh'] = vtk_writeASCII_mesh(dimension,resolution,data)
for what in out.keys():
(head,tail) = os.path.split(filename)
vtk = open(os.path.join(head,'%s_'%what+os.path.splitext(tail)[0]+'.vtk'), 'w')
if file['name'] == 'STDIN':
output(out[what],{},'Stdout')
else:
(head,tail) = os.path.split(file['name'])
vtk = open(os.path.join(head,what+'_'+os.path.splitext(tail)[0]+'.vtk'), 'w')
output(out[what],{'filepointer':vtk},'File')
vtk.close()