appypod-rattail/appy/shared/csv_parser.py

296 lines
13 KiB
Python
Raw Normal View History

# ------------------------------------------------------------------------------
# Appy is a framework for building applications in the Python language.
# Copyright (C) 2007 Gaetan Delannay
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; either version 2
# of the License, or (at your option) any later version.
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,USA.
# ------------------------------------------------------------------------------
from appy import Object
from appy.shared.utils import sequenceTypes
# ------------------------------------------------------------------------------
WRONG_LINE = 'Line number %d in file %s does not have the right number of ' \
'fields.'
class CsvParser:
'''This class reads a CSV file and creates a list of Python objects from it.
The first line of the CSV file must declare the format of the following
lines, which are 'data' lines. For example, if the first line of the file
is
id,roles*,password
Then subsequent lines in the CSV need to conform to this syntax. Field
separator will be the comma. Result of method 'parse' will be a list of
Python objects, each one having attributes id, roles and password.
Attributes declared with a star (like 'roles') are lists. An empty value
will produce an empty list in the resulting object; several values need
to be separated with the '+' sign. Here are some examples of valid 'data'
lines for the first line above:
gdy,,
gdy,MeetingManager,abc
gdy,MeetingManager+MeetingMember,abc
In the first (and subsequent) line(s), you may choose among the following
separators: , : ; |
'''
separators = [',', ':', ';', '|']
typeLetters = {'i': int, 'f': float, 's': str, 'b': bool}
def __init__(self, fileName, references={}, klass=None):
self.fileName = fileName
self.res = [] # The resulting list of Python objects.
self.sep = None
self.attributes = None # The list of attributes corresponding to
# CSV columns.
self.attributesFlags = None # Here we now if every attribute is a list
# (True) of not (False).
self.attributesTypes = None # Here we now the type of the attribute (if
# the attribute is a list it denotes the type of every item in the
# list): string, integer, float, boolean.
self.references = references
self.klass = klass # If a klass is given here, instead of creating
# Object instances we will create instances of this class.
# But be careful: we will not call the constructor of this class. We
# will simply create instances of Object and dynamically
# change the class of created instances to this class.
def identifySeparator(self, line):
'''What is the separator used in this file?'''
maxLength = 0
res = None
for sep in self.separators:
newLength = len(line.split(sep))
if newLength > maxLength:
maxLength = newLength
res = sep
self.sep = res
def identifyAttributes(self, line):
self.attributes = line.split(self.sep)
self.attributesFlags = [False] * len(self.attributes)
self.attributesTypes = [str] * len(self.attributes)
i = -1
for attr in self.attributes:
i += 1
# Is this attribute mono- or multi-valued?
if attr.endswith('*'):
self.attributesFlags[i] = True
attrNoFlag = attr.strip('*')
attrInfo = attrNoFlag.split('-')
# What is the type of value(s) for this attribute ?
if (len(attrInfo) == 2) and (attrInfo[1] in self.typeLetters):
self.attributesTypes[i] = self.typeLetters[attrInfo[1]]
# Remove trailing stars
self.attributes = [a.strip('*').split('-')[0] for a in self.attributes]
def resolveReference(self, attrName, refId):
'''Finds, in self.reference, the object having p_refId.'''
refObjects, refAttrName = self.references[attrName]
res = None
for refObject in refObjects:
if getattr(refObject, refAttrName) == refId:
res = refObject
break
return res
def convertValue(self, value, basicType):
'''Converts the atomic p_value which is a string into some other atomic
Python type specified in p_basicType (int, float, ...).'''
if (basicType != str) and (basicType != str):
try:
exec('res = %s' % str(value))
except SyntaxError as se:
res = None
else:
try:
exec('res = """%s"""' % str(value))
except SyntaxError as se:
try:
exec("res = '''%s'''" % str(value))
except SyntaxError as se:
res = None
return res
def parse(self):
'''Parses the CSV file named self.fileName and creates a list of
corresponding Python objects (Object instances). Among object fields,
some may be references. If it is the case, you may specify in
p_references a dict of referred objects. The parser will then
replace string values of some fields (which are supposed to be
ids of referred objects) with corresponding objects in p_references.
How does this work? p_references must be a dictionary:
- keys correspond to field names of the current object;
- values are 2-tuples:
* 1st value is the list of available referred objects;
* 2nd value is the name of the attribute on those objects that
stores their ID.
'''
# The first pass parses the file and creates the Python object
f = file(self.fileName)
firstLine = True
lineNb = 0
for line in f:
lineNb += 1
line = line.strip()
if not line: continue
if firstLine:
# The first line declares the structure of the following 'data'
# lines.
self.identifySeparator(line)
self.identifyAttributes(line)
firstLine = False
else:
# Add an object corresponding to this line.
lineObject = Object()
if self.klass:
lineObject.__class__ = self.klass
i = -1
# Do we get the right number of field values on this line ?
attrValues = line.split(self.sep)
if len(attrValues) != len(self.attributes):
raise WRONG_LINE % (lineNb, self.fileName)
for attrValue in line.split(self.sep):
i += 1
theValue = attrValue
vType = self.attributesTypes[i]
if self.attributesFlags[i]:
# The attribute is multi-valued
if not attrValue:
theValue = []
elif '+' in theValue:
theValue = [self.convertValue(v, vType) \
for v in attrValue.split('+')]
else:
theValue = [self.convertValue(theValue, vType)]
else:
# The attribute is mono-valued
theValue = self.convertValue(theValue, vType)
setattr(lineObject, self.attributes[i], theValue)
self.res.append(lineObject)
f.close()
# The second pass resolves the p_references if any
for attrName, refInfo in self.references.items():
if attrName in self.attributes:
# Replace ID with real object from p_references
for obj in self.res:
attrValue = getattr(obj, attrName)
if isinstance(attrValue, list) or \
isinstance(attrValue, tuple):
# Multiple values to resolve
newValue = []
for v in attrValue:
newValue.append(self.resolveReference(attrName,v))
else:
# Only one value to resolve
newValue = self.resolveReference(attrName, attrValue)
setattr(obj, attrName, newValue)
return self.res
# ------------------------------------------------------------------------------
class CsvMarshaller:
'''This class is responsible for producing a string, CSV-ready, line of data
from a Appy object.'''
undumpable = ('File', 'Action', 'Info', 'Pod')
def __init__(self, at=None, sep=';', subSep=',', wrap='"',
includeHeaders=True, include=None, exclude=None):
# If specified, p_at is an opened file handler to the CSV file to fill
self.at = at
# The CSV field separator
self.sep = sep
# The sub-separator for multi-valued fields
self.subSep = subSep
# The "wrap" char will wrap any value that contains self.sep.
self.wrap = wrap
# Must we put field names as first line in the CSV?
self.includeHeaders = includeHeaders
# If p_include is given, it lists names of fields that will be included
self.include = include
# If p_exclude is given, it lists names of fields that will be excluded
self.exclude = exclude
def marshallString(self, value):
'''Produces a version of p_value that can be put in the CSV file.'''
return value.replace('\r\n', ' ').replace('\n', ' ')
def marshallValue(self, field, value):
'''Produces a version of p_value that can be dumped in a CSV file.'''
if isinstance(value, str):
# Format the string as a one-line CSV-ready value
res = self.marshallString(value)
elif type(value) in sequenceTypes:
# Create a list of values, separated by a sub-separator.
res = []
for v in value:
res.append(self.marshallValue(field, v))
res = self.subSep.join(res)
elif hasattr(value, 'klass') and hasattr(value, 'title'):
# This is a reference to another object. Dump only its title.
res = value.title
elif value == None:
# Empty string is more beautiful than 'None'
res = ''
else:
res = str(value)
# If self.sep is found among this value, we must wrap it with self.wrap
if self.sep in res:
# Double any wrapper char if present
res = res.replace(self.wrap, '%s%s' % (self.wrap, self.wrap))
# Wrap the value
res = '%s%s%s' % (self.wrap, res, self.wrap)
return res
def includeField(self, field):
'''Must p_field be included in the result ?'''
# Check self.include and self.exclude
if self.include and field.name not in self.include: return False
if self.exclude and field.name in self.exclude: return False
# Check field type
if field.type in self.undumpable: return False
# Don't dump password fields
if (field.type == 'String') and (field.format == 3): return False
if (field.type == 'Ref') and field.isBack: return False
if (field.type == 'Computed') and not field.plainText: return False
return True
def marshall(self, obj):
'''Creates the CSV line representing p_obj and dumps it in self.at if
specified, or return it else.'''
obj = obj.appy()
res = []
# Dump the header line if required, and if there is still no line
# dumped in self.at.
headers = []
if self.includeHeaders and self.at and (self.at.tell() == 0):
for field in obj.fields:
if not self.includeField(field): continue
headers.append(field.name)
self.at.write(self.sep.join(headers))
self.at.write('\n')
# Dump the data line.
for field in obj.fields:
if not self.includeField(field): continue
# Get the field value
value = field.getValue(obj.o)
value = self.marshallValue(field, value)
res.append(value)
res = self.sep.join(res)
if self.at:
self.at.write(res)
self.at.write('\n')
else: return res
# ------------------------------------------------------------------------------