Commit c78e1c97 authored by Jérome Perrin's avatar Jérome Perrin

core: make OOo import compatible with python3

parent 97fcf0ff
......@@ -25,6 +25,7 @@ The returned mapping has the following structure:
This scripts guarantees that the list of category info is sorted in such a
way that parent always precedes their children.
"""
import six
from Products.ERP5Type.Message import translateString
from Products.ERP5OOo.OOoUtils import OOoParser
parser = OOoParser()
......@@ -41,41 +42,7 @@ if invalid_spreadsheet_error_handler is None:
property_id_set = portal.portal_types.Category.getInstancePropertySet()
property_id_set.update(getattr(portal.portal_types, 'Base Category').getInstancePropertySet())
def getIDFromString(string=None):
"""
This function transform a string to a safe and beautiful ID.
It is used here to create a safe category ID from a string.
But the code is not really clever...
"""
if string is None:
return None
clean_id = ''
translation_map = { 'a' : [u'\xe0', u'\xe3']
, 'e' : [u'\xe9', u'\xe8']
, 'i' : [u'\xed']
, 'u' : [u'\xf9']
, '_' : [' ', '+']
, '-' : ['-', u'\u2013']
, 'and': ['&']
}
# Replace odd chars by safe ascii
string = string.lower()
string = string.strip()
for (safe_char, char_list) in translation_map.items():
for char in char_list:
string = string.replace(char, safe_char)
# Exclude all non alphanumeric chars
for char in string:
if char.isalnum() or char in translation_map.keys():
clean_id += char
# Delete leading and trailing char which are not alpha-numerics
# This prevent having IDs with starting underscores
while len(clean_id) > 0 and not clean_id[0].isalnum():
clean_id = clean_id[1:]
while len(clean_id) > 0 and not clean_id[-1].isalnum():
clean_id = clean_id[:-1]
return clean_id
getIDFromString = portal.Base_getSafeIdFromString
# if the file is not an open office format, try to convert it using oood
# FIXME: use portal_transforms
......@@ -129,7 +96,7 @@ for table_name in spreadsheet_list.keys():
else:
# If there is a new column with a header and the path definition has
# started, that seems the path definition has ended
property_map[column_index] = column_id.encode('utf8')
property_map[column_index] = column_id.encode('utf8') if six.PY2 else column_id
column_index += 1
# Construct categories data (with absolute path) from table lines
......@@ -137,9 +104,9 @@ for table_name in spreadsheet_list.keys():
# 1 table = 1 base category
base_category_name = table_name
base_category_id = getIDFromString(base_category_name)
if same_type(base_category_name, u''):
if six.PY2 and isinstance(base_category_name, unicode):
base_category_name = base_category_name.encode('utf8')
if same_type(base_category_id, u''):
if six.PY2 and isinstance(base_category_id, unicode):
base_category_id = base_category_id.encode('utf8')
category_list = category_list_spreadsheet_mapping.setdefault(base_category_id, [])
category_list.append({ 'path' : base_category_id
......@@ -198,7 +165,7 @@ for table_name in spreadsheet_list.keys():
if cell_id not in ('', None):
# Handle normal properties
if not property_id.startswith('path_'):
if same_type(cell_data, u''):
if six.PY2 and same_type(cell_data, u''):
cell_data = cell_data.encode('utf8')
category_property_list[property_id] = cell_data
# Handle 'path' property
......@@ -218,7 +185,7 @@ for table_name in spreadsheet_list.keys():
# Get the next depth
break
path = '/'.join([base_category_id,] + absolute_path_element_list[::-1])
if same_type(path, u''):
if six.PY2 and same_type(path, u''):
path = path.encode('utf8')
category_property_list['path'] = path
......
......@@ -2,27 +2,32 @@
This function transform a string to a safe id.
It is used here to create a safe category id from a string.
"""
translation_map = { "a": ['\xe0']
, "e": ['\xe9', '\xe8']
}
clean_id = ''
if s is None:
return None
clean_id = ''
translation_map = { 'a' : [u'\xe0', u'\xe3']
, 'e' : [u'\xe9', u'\xe8']
, 'i' : [u'\xed']
, 'u' : [u'\xf9']
, '_' : [' ', '+']
, '-' : ['-', u'\u2013']
, 'and': ['&']
}
# Replace odd chars by safe ascii
s = s.lower()
s = s.strip()
# oocalc inserts some strange chars when you press - key in a text cell.
# Following line is a workaround for this, because \u2013 does not exist in latin1
s = s.replace(u'\u2013', '-')
for char in s.encode('iso8859_1'):
if char == '_' or char.isalnum():
for (safe_char, char_list) in translation_map.items():
for char in char_list:
s = s.replace(char, safe_char)
# Exclude all non alphanumeric chars
for char in s:
if char.isalnum() or char in translation_map.keys():
clean_id += char
elif char.isspace() or char in ('+', '-'):
clean_id += '_'
else:
for (safe_char, char_list) in translation_map.items():
if char in char_list:
clean_id += safe_char
break
# Delete leading and trailing char which are not alpha-numerics
# This prevent having IDs with starting underscores
while len(clean_id) > 0 and not clean_id[0].isalnum():
clean_id = clean_id[1:]
while len(clean_id) > 0 and not clean_id[-1].isalnum():
clean_id = clean_id[:-1]
return clean_id
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment