Mercurial > pyarq-presupuestos
view Generic/fiebdc.py @ 12:29a4044a95ff
svg to png
author | Miguel Ángel Bárcena Rodríguez <miguelangel@obraencurso.es> |
---|---|
date | Thu, 30 Dec 2010 12:28:33 +0100 |
parents | 0359329a1c26 |
children | a7b9f7e7dfa4 |
line wrap: on
line source
#!/usr/bin/python # -*- coding: utf-8 -*- ## File fiebdc.py ## This file is part of pyArq-Presupuestos. ## ## Copyright (C) 2010 Miguel Ángel Bárcena Rodríguez ## <miguelangel@obraencurso.es> ## ## pyArq-Presupuestos 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 3 of the License, or ## (at your option) any later version. ## ## pyArq-Presupuestos 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, see <http://www.gnu.org/licenses/>. # specifications in http://www.fiebdc.org # Modules import time import re import calendar import os.path # pyArq-Presupuestos modules import base from Generic import utils from Generic import globalVars class Read(object): """fiebdc.Read: Description: Reads and parses a fiebdc file Constructor: fiebdc.Read(filename=None, budget=None) Ancestry: +-- object +-- Read Atributes: "__budget": budget ("base.Budget" object) "__file_format": File format of the fiebdc file "__format_list": List of file format that can be readed "__character_sets_dict": Dictionary with the character sets supported "__character_set": character_set of the file "__generator": program which the file is created "__cancel": Boolean value, True mean that the read process must stop "__filename": The filename of the fiebdc file that is readed "__pattern": re compiled pattern dict Methods: __init__(self, filename=None, budget=None) cancel(self) eraseControlCharacters(self, string) validateCode(self, code) parseDate(self, date) parseRecord(self,record) _parseV(self, field_list) _parseC(self, field_list) _parseDY(self, field_list) _parseMN(self, field_list) _parseT(self, field_list) _parseK(self, field_list) _parseW(self, field_list) _parseL(self, field_list) _parseQ(self, field_list) _parseJ(self, field_list) _parseG(self, field_list) _parseE(self, field_list) _parseX(self, field_list) _parseF(self, field_list) readFile(self, budget=None, filename=None) """ def __init__(self, filename=None, budget=None): """def __init__(self, filename=None, budget=None) Sets the instance attributes """ self.__budget = budget self.__filename = filename if not self.__budget is None: self.__budget.filename = self.__filename self.__cancel = False self.__format_list = ["FIEBDC-3/95", "FIEBDC-3/98", "FIEBDC-3/2002", "FIEBDC-3/2004", "FIEBDC-3/2007"] # ANSI->¿"ISO-8859-15" or "latin1 ISO-8859-1" or "cp1252 windows-1252"? # 850 -> IBM850 -> cp850 # 437 -> IBM437 -> cp437 self.__character_sets_dict = {"ANSI" : "cp1252", "850" : "850", "437" : "cp437"} self.__file_format = "FIEBDC-3/2007" self.__generator = globalVars.version self.__character_set = "850" self.__pattern = { "control_tilde" : re.compile("((\r\n)| |\t)+~"), "control_vbar" : re.compile("((\r\n)| |\t)+\|"), "control_backslash" : re.compile(r"((\r\n)| |\t)+\\"), "valid_code" : re.compile("[^A-Za-z0-9ñÑ.$#%&_]"), "special_char": re.compile("[#%&]"), "no_float": re.compile("[^0-9.]"), "formula" : re.compile(".*[^0123456789\.()\+\-\*/\^abcdp ].*"), "comment": re.compile("#.*\r\n"), "empty_line": re.compile(r"(\r\n) *\r\n"), "space_before_backslash" : re.compile(r"( )+\\"), "space_after_backslash" : re.compile(r"\\( )+"), "start_noend_backslash" : re.compile("(\r\n\\\.*[^\\\])\r\n"), "end_oper": re.compile("(\+|-|\*|/|/^|@|&|<|>|<=|>=|=|!) *\r\n"), "matricial_var" : re.compile("(\r\n *[%|\$][A-ZÑ].*=.*,) *\r\n"), "descomposition" : re.compile("^([^:]+):(.*)$"), "var" : re.compile("^([$%][A-ZÑ][()0-9, ]*)=(.*)$"), "after_first_tilde" : re.compile("^[^~]*~"), "end_control" : re.compile("((\r\n)| |\t)+$"), } def cancel(self): """def cancel(self) Sets the "__cancel" attribute to True, It stops the read process. """ self.__cancel = True def eraseControlCharacters(self, string): """eraseControlCharacters(self,string) Return a copy of the string with the blank characters (32), tabs (9) and end of line (13 and 10) before of the separators '~', '|' erased. Before separator \ not deleted because it affects the reading of the record ~P """ # "control_tilde" : "((\r\n)| |\t)+~" string = self.__pattern["control_tilde"].sub("~",string) # "control_vbar" : "((\r\n)| |\t)+\|" string = self.__pattern["control_vbar"].sub("|",string) # "control_backslash" : r"((\r\n)| |\t)+\\" #string = self.__pattern["control_backslash"].sub(r"\\",string) return string def validateCode(self, code): """validateCode(self, code) Test if the code have invalid characters and try to erase it, if it is posible return a valid code else return a empty string. """ if not isinstance(code, str): print _("Invalid code, it must be a string") return "" # Valid chararcter: A-Z a-z 0-9 ñ Ñ . $ # % & _ # "valid_code" : "[^A-Za-z0-9ñÑ.$#%&_]" _code = self.__pattern["valid_code"].sub("", code) if _code != code: print utils.mapping(_("The code '$1' have invalid characters."), (code,)) code = _code # the lasts characters can not be <#> or <##> # <##> -> root record in FIEFDC-3 # <#> -> chapter record in FIEFDC-3 if len(code) > 0: while code[-1] == "#": code = code[:-1] if len(code) > 20: code = code[:20] # only one charecter # % or & if sum([code.count(c) for c in '#%&']) > 1: print utils.mapping(_("The code '$1' contains special "\ "characters repeated."),(code,)) _i = min([code.find(c) for c in '#%&']) code = code[:_i+1] + \ self.__pattern["special_char"].sub("", code[_i+1:]) return code def parseDate(self, date): """parseDate(self, date) date: in the format: uneven len: add a Leading 0 len = 8 DDMMYYYY len <= 6 DDMMYY “80/20”. >80 -> >1980 <80 -> <2080 len < 5 MMYY len < 3 YY Test date string and return a tuple (YYYY, MM, DD) or None if the date format is invalid """ # All characters must be numbers, len <= 8 and not empty string if not date.isdigit() or len(date) > 8 or date == "": return None else: if len(date)%2 == 1: # uneven len: add a leading 0 date = "0" + date if len(date) == 8: _d = int(date[:2]) _m = int(date[2:4]) _y = int(date[4:8]) elif len(date) <= 6: _y = int(date[-2:]) if _y < 80: _y = 2000 + _y else: _y = 1900 + _y if len(date) == 6: _d = int(date[:2]) _m = int(date[2:4]) elif len(date) == 4: _d = 0 _m = int(date[:2]) elif len(date) == 2: _d = 0 _m = 0 if not _d in range(1,31): _d = 0 if not _m in range(1,12): _m = 0 if _m == 0: _d = 0 if _m != 0 and _d != 0: if calendar.monthrange(_y, _m)[1] < _d: _d = 0 return (_y, _m, _d) def parseRecord(self,record): """parseRecord(self,record) record: the record line readed from the file whith the format: type|field|field|subfield\subfield|... [a] nothing or "a" {a} zero or more #-#twice#-# "a" <a> one or more #-#twice#-# "a" Types: V C D Y M N T K L Q J G E X B F A V: Property and Version 1- [File_Owner] 2- Format_Version[\DDMMYYYY] 3- [Program_Generator] 4- [Header]\{Title\} 5- [Chaters_set] 6- [Comment] C: Record: 1- Code{\Code} 2- [Unit] 3- [Summary] 4- {Price\} 5- {Date\} 6- [Type] D or Y: DECOMPOSITION or ADD DECOMPOSITION 1- Parent Code 2- <Child Code\ [Factor]\ [Yield]> M or N: MEASURE or ADD MEASURE 1- [Parent Code\]Child Code 2- {Path\} 3- TOTAL MEASURE 4- {Type\Comment\Unit\Length\Width\Height\} 5- [Label] T: Text 1- Code 2- Description text K: Coefficients 1- { DN \ DD \ DS \ DR \ DI \ DP \ DC \ DM \ DIVISA \ } 2- CI \ GG \ BI \ BAJA \ IVA 3- { DRC \ DC \ DRO \ DFS \ DRS \ DFO \ DUO \ DI \ DES \ DN \ DD \ DS \ DIVISA \ } 4- [ n ] L: Sheet of Conditions 1 A) 1- Empty 2- {Section Code\Section Title} B) 1- Record Code 2- {Section Code\Section Text} 3- {Section Code\RTF file} 4- {Section Code\HTM file} Q: Sheet of Conditions 2 1- Record Code 2- {Section Code\Paragraph key\{Field key;}\}| J: Sheet of Conditions 3 1- Paragraph code 2- [Paragraph text] 3- [RTF file] 4- [HTML file] G: Grafic info 1- <grafic_file.ext\> E: Company 1- company Code 2 [ summary ] 3- [ name ] 4- { [ type ] \ [ subname ] \ [ address ] \ [ postal_code ] \ [ town ] \ [ province ] \ [ country ] \ { phone; } \ { fax; } \ {contact_person; } \ } 5- [ cif ] \ [ web ] \ [ email ] \ X: Tecnical information A) 1- Empty 2- < TI_Code \ TI_Descitption \ TI_Unit > B) 1- Record_code 2- < TI_Code \ TI_value > F: #-#Adjunto#-# File 1- Record code 2- { Type \ { Filenames; } \ [Description] } B: Change code 1- Record Code 2- New code A: Labels 1- Record Code 2- <Label\> """ # TODO: ~L ~J RTF and HTML files # TODO: test ~Q ~J ~G # TODO: ~P. Registro tipo Descripción Paramétrica. # TODO: ~O. Registro tipo Relación Comercial. # TODO: test records _field_list = record.split("|") self._record_number = self._record_number +1 _budget = self.__budget if _field_list[0] == "V": self._record_V_number += 1 self._parseV(_field_list) elif _field_list[0] == "C": self._record_C_number += 1 self._parseC(_field_list) elif _field_list[0] == "D": self._record_D_number += 1 self._parseDY(_field_list) elif _field_list[0] == "Y": self._record_Y_number += 1 self._parseDY(_field_list) elif _field_list[0] == "M": self._record_M_number += 1 self._parseMN(_field_list) elif _field_list[0] == "N": self._record_N_number += 1 self._parseMN(_field_list) elif _field_list[0] == "T": self._record_T_number += 1 self._parseT(_field_list) elif _field_list[0] == "K": self._record_K_number += 1 self._parseK(_field_list) elif _field_list[0] == "W": self._record_W_number += 1 self._parseW(_field_list) elif _field_list[0] == "L": self._record_L_number += 1 self._parseL(_field_list) elif _field_list[0] == "Q": self._record_Q_number += 1 self._parseQ(_field_list) elif _field_list[0] == "J": self._record_J_number += 1 self._parseJ(_field_list) elif _field_list[0] == "G": self._record_G_number += 1 self._parseG(_field_list) elif _field_list[0] == "E": self._record_E_number += 1 self._parseE(_field_list) elif _field_list[0] == "O": self._record_O_number += 1 elif _field_list[0] == "P": self._record_P_number += 1 self._parseP(_field_list) elif _field_list[0] == "X": self._record_X_number += 1 self._parseX(_field_list) elif _field_list[0] == "B": self._record_B_number += 1 self._parseB(_field_list) elif _field_list[0] == "F": self._record_F_number += 1 self._parseF(_field_list) elif _field_list[0] == "A": self._record_A_number += 1 self._parseA(_field_list) else: self._record_Unknow_number += 1 def _parseV(self, field_list): """_parseV(self, field_list) field_list: field list of the record 0- V :Property and Version 1- [File_Owner] 2- Format_Version[\DDMMYYYY] 3- [Program_Generator] 4- [Header]\{Title\} 5- [Chaters_set] 6- [Comment] 7- [Data type] 8- [Number budget certificate] 9- [Date budget certificate] """ if self._record_number != 1: print utils.mapping(_("The 'V' record (Property and Version) "\ "must be the first record in the file but it is the "\ "number: $1"), (self._record_number,)) print _("The default values were taken and this V record is "\ "ignored") return # _____number of fields_____ # Any INFORMATION after last field separator is ignored if len(field_list) > 10: field_list = field_list[:10] # If there are no sufficient fields, the fields are added # with empty value:"" else: field_list = field_list + [""]*(10-len(field_list)) # control character are erased: end of line, tab, space # only leading and trailing whitespace in owner, generator, comment # _____Fields_____ _record_type = self.delete_control_space(field_list[0]) _owner = field_list[1].strip() _owner = self.delete_control(_owner) _version_date = self.delete_control_space(field_list[2]) _generator = field_list[3].strip() _generator = self.delete_control(_generator) _header_title = field_list[4].strip() _header_title = self.delete_control(_header_title) _character_set = self.delete_control_space(field_list[5]) _comment = field_list[6].strip("\t \n\r") _data_type = self.delete_control_space(field_list[7]) _number_certificate = self.delete_control_space(field_list[8]) __date_certificate = self.delete_control_space(field_list[9]) # _____Owner_____ self.__budget.setOwner(_owner) # _____Version-Date_____ _version_date = _version_date.split("\\") _file_format = _version_date[0] if _file_format in self.__format_list: self.__file_format = _file_format print _("FIEBDC format: %s" % _file_format) if len(_version_date) > 1: _date = _version_date[1] if _date != "": _parsed_date = self.parseDate(_date) if _parsed_date is not None: self.__budget.setDate(_parsed_date) # _____Generator_____ # ignored field print _("FIEBDC file generated by %s" % _generator) # _____Header_Title_____ _header_title = _header_title.split("\\") _header_title = [_title.strip() for _title in _header_title] _header = _header_title.pop(0) _title = [ ] for _title_index in _header_title: if _title_index != "": _title.append(_title_index) if _header != "": self.__budget.setTitleList([ _header, _title ]) # _____Characters_set_____ # field parsed in readFile method # _____Comment_____ if _comment != "": self.__budget.setComment(_comment) # _____Data type_____ # 1 -> Base data. # 2 -> Budget. # 3 -> Budget certificate. # 4 -> Base date update. try: _data_type = int(_data_type) except ValueError: _data_type = "" if _data_type == 3: # _____Number budget certificate_____ try: _number_certificate = int(_number_certificate) except ValueError: _number_certificate = "" # _____Date budget certificate_____ if _date_certificate != "": _parsed_date_certificate = self.parseDate(_date_certificate) if _parsed_date_certificate is None: _date_certificate = "" else: _date_certificate = _parsed_date_certificate self.__budget.setBudgetype(_data_type) self.__budget.setCertificateOrder(_number_certificate) self.__budget.setCertificateDate(_parsed_date_cerfificate) elif _data_type != "": self.__budget.setBudgeType(_data_type) self.num_valid_record = self.num_valid_record + 1 def _parseK(self, field_list): """_parseK(self, field_list) field_list: field list of the record 0- K: Coefficients 1- { DN \ DD \ DS \ DR \ DI \ DP \ DC \ DM \ DIVISA \ } 2- CI \ GG \ BI \ BAJA \ IVA 3- A){ DRC \ DC \ DRO \ DFS \ DRS \ DFO \ DUO \ DI \ DES \ DN \ DD \ DS \ DIVISA \ } B){ DRC \ DC \ \ DFS \ DRS \ \ DUO \ DI \ DES \ DN \ DD \ DS \ DSP\ DEC\ DIVISA \ } 4- [ n ] """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 3 fields # The last field is ignored, pyArq hate dll's if len(field_list) > 4: field_list = field_list[1:4] # The record must have 3 fields else: field_list = field_list[1:] + [""]*(4-len(field_list)) # control character are erased: end of line, tab, space # _____Fields_____ _field0 = self.delete_control_space(field_list[0]) _field1 = self.delete_control_space(field_list[1]) _field2 = self.delete_control_space(field_list[2]) # _____Field 1_____ if len(_field1) > 0 and _field1[-1] == "\\": _field1 = _field1[:-1] # if there are a \ character at the end it must be erased _percentages = _field1.split("\\") if len(_percentages) > 5: _percentages = _percentages[:5] # If there are no sufficient subfields, the subfields are added # with empty value:"" else: _percentages = _percentages + [""]*(5-len(_percentages)) _percentage_titles = [ "CI", "GG", "BI", "BAJA", "IVA" ] _percentage_dict = {} for _percentage_index in range(len(_percentages)): try: _percentage = int(_percentages[_percentage_index]) except ValueError: _percentage = "" _percentage_dict[_percentage_titles[_percentage_index]] = \ _percentage self.__budget.setPercentages(_percentage_dict) # _____Field 0 and 1_____ # Default number of decimal places # Number of titles in ~V record _title_num = len(self.__budget.getTitleList()[1]) if _title_num == 0: _title_num = 1 # If the field 2 is empty, the field 0 is readed if _field2 == "": # _____Field 0_____ if _field0[-1] == "\\": _field0 = _field0[:-1] # if there are a \ character at the end it must be erased _decimal_list = _field0.split("\\") _decimal_index = 0 if len(_decimal_list)%9 != 0: # if it is not multiple of 9, empty subfield are added _decimal_list = _decimal_list + [""]*(9 - \ len(_decimal_list)%9) # The number of decimal values is the same as the numbers of # titles in the V record if len(_decimal_list)//9 > _title_num: _decimal_list = _decimal_list[:_title_num*9] elif len(_decimal_list)//9 < _title_num: _decimal_list = _decimal_list + _decimal_list[-9:] * \ (_title_num-(len(_decimal_list)//9)) while _decimal_index <= len(_decimal_list)-9: _decimals = _decimal_list[_decimal_index:(_decimal_index + 9)] _forlist = range(len(_decimals)-1) for _index in range(len(_decimals)): try: #TODO: test this _decimals[_index] = int(_decimals[_index]) except ValueError: _decimals[_index] = "" _DN = _decimals[0] _DD = _decimals[1] _DS = _decimals[2] _DR = _decimals[3] _DI = _decimals[4] _DP = _decimals[5] _DC = _decimals[6] _DM = _decimals[7] _DIVISA = _decimals[8] _percentage_dict = {"DN" : _DN, "DD" : _DD, "DSP" : _DS, "DS" : _DS, "DFC" : _DR, "DFPU" : _DR, "DFUO" : _DR, "DFA" : _DR, "DRC" : _DR, "DRPU" : _DR, "DRUO" : _DR, "DRA" : _DR, "DP" : _DC, "DC" : _DC, "DPU" : _DC, "DUO" : _DC, "DEA" : _DC, "DES" : _DC, "DIR" : _DI, "DIRC" : _DI, "DCD" : _DP, "DIVISA": _DIVISA } _decimal_index = _decimal_index + 9 self.__budget.setDecimals(_percentage_dict, (_decimal_index//9)) else: # _____Field 3_____ if _field2[-1] == "\\": _field2 = _field2[:-1] # if there are a \ character at the end it must be erased _decimal_list = _field2.split("\\") # test if the Divisa subfield is 12 or 14 position # Divisa is the only Alphanumeric subfield # "no_float": "[^0-9.]" if len(_decimal_list) >= 13 and \ self.__pattern["no_float"].search(_decimal_list[12]): _multiple = 13 elif len(_decimal_list) >= 15 and \ self.__pattern["no_float"].search(_decimal_list[14]): _multiple = 15 else: if self.__file_format == "FIEBDC-3/2002": _multiple = 13 elif self.__file_format == "FIEBDC-3/2004": _multiple = 13 elif self.__file_format == "FIEBDC-3/2007": _multiple = 15 else: _multiple = 15 _decimal_index = 0 if len(_decimal_list)%_multiple != 0 : # if it is not multiple of _multiple, empty subfield are added _decimal_list = _decimal_list + \ [""]*(_multiple-len(_decimal_list)%_multiple) # The number of decimal values is the same as the numbers of # titles in the V record if len(_decimal_list)//_multiple > _title_num: _decimal_list = _decimal_list[:_title_num*_multiple] elif len(_decimal_list)//_multiple < _title_num: _decimal_list = _decimal_list + [_decimal_list[-_multiple:]]*\ (_title_num-(len(_decimal_list)//_multiple)) while _decimal_index <= len(_decimal_list)-_multiple: _decimals = _decimal_list[_decimal_index:(_decimal_index +\ _multiple)] for _index in range(len(_decimals)-1): try: _decimals[_index] = int(_decimals[_index]) except: _decimals[_index] = "" if _multiple == 13: _DRC = _decimals[0] _DC = _decimals[1] _DRO = _decimals[2] _DFS = _decimals[3] _DRS = _decimals[4] _DFO = _decimals[5] _DUO = _decimals[6] _DI = _decimals[7] _DES = _decimals[8] _DN = _decimals[9] _DD = _decimals[10] _DS = _decimals[11] _DIVISA = _decimals[12] _percentage_dict = { "DN" : _DN, "DD" : _DD, "DSP" : _DS, "DS" : _DS, "DFC" : _DFS, "DFPU" : _DRC, "DFUO" : _DFS, "DFA" : _DFS, "DRC" : _DRS, "DRPU" : _DRC, "DRUO" : _DRS, "DRA" : _DRS, "DP" : _DC, "DC" : _DC, "DPU" : _DC, "DUO" : _DUO, "DEA" : _DES, "DES" : _DES, "DIR" : _DI, "DIRC" : _DC, "DCD" : _DI, "DIVISA": _DIVISA, } else: # _multiple == 15: _DRC = _decimals[0] _DC = _decimals[1] _DRO = _decimals[2] _DFS = _decimals[3] _DRS = _decimals[4] _DFO = _decimals[5] _DUO = _decimals[6] _DI = _decimals[7] _DES = _decimals[8] _DN = _decimals[9] _DD = _decimals[10] _DS = _decimals[11] _DSP = _decimals[12] _DEC = _decimals[13] _DIVISA = _decimals[14] _percentage_dict = { "DN" : _DN, "DD" : _DD, "DSP" : _DSP, "DS" : _DS, "DFC" : _DFS, "DFPU" : _DRC, "DFUO" : _DFS, "DFA" : _DFS, "DRC" : _DRS, "DRPU" : _DRC, "DRUO" : _DRS, "DRA" : _DRS, "DP" : _DC, "DC" : _DC, "DPU" : _DC, "DUO" : _DUO, "DEA" : _DEC, "DES" : _DES, "DIR" : _DI, "DIRC" : _DC, "DCD" : _DI, "DIVISA": _DIVISA} _decimal_index = _decimal_index + 13 self.__budget.setDecimals(_percentage_dict, (_decimal_index//13)) self.num_valid_record = self.num_valid_record +1 def _parseC(self, field_list): """_parseC(self, field_list) field_list: field list of the record 0- C: Record 1- Code{\Code} 2- [Unit] 3- [Summary] 4- {Price\} 5- {Date\} 6- [Type] """ # _____number of fields_____ # Any INFORMATION after last field separator is ignored if len(field_list) > 7: field_list = field_list[:7] # If there are no sufficient fields, the fields are added # with empty value:"" else: field_list = field_list + [""]*(7-len(field_list)) # control character are erased: en of line, tab, space # _____Fields_____ _record_type = field_list[0] _codes = self.delete_control_space(field_list[1]) _unit = self.delete_control_space(field_list[2]) _summary = self.delete_control(field_list[3]) _prices = self.delete_control_space(field_list[4]) _dates = self.delete_control_space(field_list[5]) _type = self.delete_control_space(field_list[6]) # _____Code_____ _codes = _codes.split("\\") if len(_codes) > 0: # parse the hierarchy of the first code # hierarchy: 0->root, 1->Chapter/subchapter, 2->other if len(_codes[0]) > 2 and _codes[0][-2:] == "##": _hierarchy = 0 elif len(_codes[0]) > 1 and _codes[0][-1:] == "#": _hierarchy = 1 else: _hierarchy = 2 # "#" and "##" characters at the end of the code are erased # invalid characters are also erased # maximun len 20 characters _codes = [self.validateCode(_code) for _code in _codes] # empty codes are ignored while "" in _codes: _codes.remove("") if len(_codes) > 0: #TODO: test this _code = _codes[0] _synonyms = _codes else: print _("Record C without a valid code") return # _____Unit_____ # nothing to do # _____Summary_____ # nothing to do # _____Price_____ and _____Dates_____ # last \ is erased if len(_dates) > 0 and _dates[-1] == "\\": _dates = _dates[:-1] if len(_prices) > 0 and _prices[-1] == "\\": _prices = _prices[:-1] _dates = _dates.split("\\") _prices = _prices.split("\\") # number of prices = number of titles in "V" line # if there are no sufficient prices it takes the last price defined _title_num = len(self.__budget.getTitleList()[1]) if _title_num == 0: _title_num = 1 if len(_prices) > _title_num: _prices = _prices[:_title_num] elif len(_prices) < _title_num: _prices = _prices + [_prices[-1]]*(_title_num-len(_prices)) # number of dates = number of prices # if there are no sufficient dates it takes the last date defined if len(_dates) > len(_prices): _dates = _dates[:len(_prices)] elif len(_dates) < len(_prices): _dates = _dates + [_dates[-1]]*(len(_prices)-len(_dates)) for _index in range(len(_prices)): # TODO: lack to specify the number of decimals of the price try: _prices[_index] = float(_prices[_index]) except: _prices[_index] = 0.0 _parsed_date = self.parseDate(_dates[_index]) if _parsed_date is None: _dates[_index] = "" else: _dates[_index] = _parsed_date # _____Type_____ # 0 Without classifying # EA Auxiliary element # EU Unitary element # EC Complex element # EF Functional element # OB Construction site # PA Cost overrun # PU Unitary budget # 1 Labourforce # H Labourforce # 2 Machinery and auxiliary equipment # Q Machinery # % Auxiliary equipment # 3 Building materials # MC Cement # MCr Ceramic # MM Wood # MS Iron and steel # ME Energy # MCu Copper # MAl Aluminium # ML Bonding agents # M Others materials # Hierarchy type subtype # 0->root -> 0 -> None,OB # 1->[sub]chapter -> 0 -> None,PU # 2->Other -> 0 -> None,EA,EU,EC,EF,PA # 1 -> None,H # 2 -> None,Q,% # 3 -> None,MC,MCr,MM,MS,ME,MCu,Mal,ML,M if _hierarchy == 0: if _type == "OB": _subtype = _type _type = 0 elif _type == "0" or _type == "": _subtype = "" _type = 0 else: print utils.mapping(_("Incorrect type ($1) in the code $2"), (str(_type), _code)) _type = 0 _subtype = "" elif _hierarchy == 1: if _type == "PU": _subtype = _type _type = 0 elif _type == "0" or _type == "": _subtype = "" _type = 0 else: print utils.mapping(_("Incorrect type ($1) in the code $2"), (str(_type), _code)) _type = 0 _subtype = "" else: if _type == "EA" or _type == "EU" or _type == "EC" or \ _type == "EF" or _type == "PA": _subtype = _type _type = 0 elif _type == "H": _subtype = _type _type = 1 elif _type == "Q" or _type == "%": _subtype = _type _type = 2 elif _type == "MC" or _type == "MCr" or _type == "MM" or \ _type == "MS" or _type == "ME" or _type == "MCu" or \ _type == "Mal" or _type == "ML" or _type == "M": _subtype = _type _type = 3 elif _type == "0" or _type == "1" or _type == "2" or \ _type == "3": _subtype = "" _type = int(_type) elif _type == "": _subtype = "" _type = 0 else: print utils.mapping(_("Incorrect type ($1) in the code $2"), (str(_type), _code)) _type = 0 _subtype = "" self.__budget.setRecord(_code, _synonyms, _hierarchy, _unit, _summary, _prices, _dates, _type, _subtype) self.num_valid_record = self.num_valid_record + 1 def _parseDY(self, field_list): """_parseDY(self, field_list) field_list: field list of the record 0- D or Y: DECOMPOSITION or ADD DECOMPOSITION 1- Parent Code 2- <Child Code\ [Factor]\ [Yield]> """ # _____number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[:3] # If there are no sufficient fields, the fields are added # with empty value:"" else: field_list = field_list + [""]*(3-len(field_list)) # control character are erased: end of line, tab, space # _____Fields_____ _record_type = field_list[0] _code = self.delete_control_space(field_list[1]) _children = self.delete_control_space(field_list[2]) # _____Code_____ # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _code = self.validateCode(_code) # _____children_____ # TODO: test the number of decimals in factor an yield values _children = _children.split( "\\" ) _children_list = [ ] _child_index = 0 while _child_index < len(_children)-3: # _____subfields_____ _child_code = _children[_child_index] _factor = _children[_child_index+1] _yield = _children[_child_index+2] # _____child_code_____ _child_code = self.validateCode(_child_code) # _____factor_____ if _factor != "": try: _factor = float(_factor) except ValueError: print utils.mapping(_("ValueError loadig the "\ "descomposition of the record $1, the factor "\ "of the child $2 must be a float number and "\ "can not be $3, seted default value 1.0"), (_code, _child_code, _factor)) _factor = 1.0 #____yield___ if _yield != "": try: _yield = float(_yield) except ValueError: print utils.mapping(_("ValueError loading the "\ "descomposition of the record $1, the yield of "\ "the child $2, must be a float number and can"\ "not be $3, seted default value 1.0"), (_code, _child_code, _factor)) _yield = 1.0 if _child_code != "" and _code != "": _children_list.append([_child_code, _factor, _yield ]) if _record_type == "D": _position = _child_index / 3 else: #_record_type == "Y" _position = -1 self.__budget.setTree(_code, _child_code, _position, _factor, _yield, "", "", "", "") _child_index = _child_index + 3 self.num_valid_record = self.num_valid_record +1 def _parseT(self, field_list): """_parseT(self, field_list) field_list: field list of the record 0- T: Text 1- Record code 2- Description text """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] if len(field_list) != 2: return # control character are erased: end of line, tab, space # _____Fields_____ _code = self.delete_control_space(field_list[0]) _text = field_list[1] # _____Code_____ # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _code = self.validateCode(_code) # _____Text_____ self.__budget.setText(_code, _text) self.num_valid_record = self.num_valid_record + 1 def _parseMN(self, field_list): """_parseMN(self, field_list) field_list: field list of the record 0- M or N: MEASURE or ADD MEASURE 1- [Parent Code\]Child Code 2- {Path\} 3- TOTAL MEASURE 4- {Type\Comment\Unit\Length\Width\Height\} 5- [Label] """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 6 fields if len(field_list) > 6: field_list = field_list[:6] # If there are no sufficient fields, the fields are added # with empty value:"" else: field_list = field_list + [""]*(6-len(field_list)) # control character are erased: end of line, tab, space # _____Fields_____ _record_type = field_list[0] _codes = self.delete_control_space(field_list[1]) _path = self.delete_control_space(field_list[2]) _total = self.delete_control_space(field_list[3]) _lines = self.delete_control(field_list[4]) _label = self.delete_control_space(field_list[5]) # _____Codes_____ _code_list = _codes.split( "\\" ) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased if len(_code_list) == 2: _parent_code = self.validateCode(_code_list[0]) if _parent_code == "": _parent_code = None _child_code = self.validateCode(_code_list[1]) elif len(_code_list) == 1: _child_code = self.validateCode(_code_list[0]) _parent_code = None else: print utils.mapping(_("Invalid codes in $1 record, codes $2"), (_record_type, _codes)) return if _child_code == "": print utils.mapping(_("Empty child code in $1 record, codes: "\ "$2"), (_record_type, _codes)) return # _____Path_____ # TODO: path=0, no-estructured measures _path_list = _path.split( "\\" ) if len(_path_list) > 0: while _path_list[-1] == "": _path_list = _path_list[:-1] _path = _path_list[-1] try: _path = int(_path) except ValueError: print utils.mapping(_("Invalid path in $1 record, "\ "codes $2"), (_record_type, _codes)) return if _path > 0: _path -= 1 else: _path = 0 # _____Total_____ try: _total = float(_total) except ValueError: print utils.mapping(_("Invalid Total Measure value in $1 "\ "record, codes $2"), (_record_type, _codes)) return # _____Measure lines_____ _lines = _lines.split( "\\" ) _line_index = 0 _line_list = [ ] while _line_index < len(_lines)-6: _linetype = _lines[_line_index] if _linetype == "": _linetype = 0 elif _linetype == "1" or _linetype == "2" or \ _linetype == "3": _linetype = int(_linetype) else: _linetype = 0 _comment= _lines[_line_index + 1] if _linetype == 3: # "formula": ".*[^0123456789\.()\+\-\*/\^abcdp ].*" if self.__pattern["formula"].match(_comment): print utils.mapping(_("The comment is not a formula or "\ "its have invalid characters, in the $1 record, "\ "codes $2"), (_record_type, _codes)) return else: _formula = _comment _comment = "" else: _formula = "" _units = _lines[_line_index + 2] _length = _lines[_line_index + 3] _width = _lines[_line_index + 4] _height = _lines[_line_index + 5] try: if _units != "": _units = float(_units) if _length != "": _length = float(_length) if _width != "": _width = float(_width) if _height != "": _height = float(_height) except ValueError: print utils.mapping("The measure values are not float "\ "numbers, code $1", (_codes,)) return _line_list.append([_linetype, _comment, _units, _length, _width, _height, _formula]) _line_index = _line_index + 6 self.__budget.setTree(_parent_code, _child_code, _path, "", "", _total, _line_list, _label, _record_type) self.num_valid_record = self.num_valid_record + 1 def _parseW(self, field_list): """_parseW(self, field_list) field_list: field list of the record 0- W: Geografical field 1- Field Code 2- Field """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 2 fields if len(field_list) >= 2: field_list = field_list[1:2] else: return # control character are erased: end of line, tab, space # _____Fields_____ _code_fields = field_list[0] # last \ is erased if len(_code_fields) and _code_fields[-1] == "\\": _code_fields = _code_fields[:-1] _code_fields = _code_fields.split("\\") _field_dict = {} _field_index = 0 while _field_index < len(_code_fields)-1: # _____subfields_____ _field_code = _code_fields[_field_index] _field_title = _code_fields[_field_index+1] # control character are erased: end of line, tab, space # _____section_code_____ #"control": "[\t \n\r]" _field_code = self.delete_control_space(_field_code) # _____section_title_____ if _field_code != "": _field_dict[_field_code] = _field_title _field_index = _field_index + 2 self.__budget.setSheetFields(_field_dict) self.num_valid_record = self.num_valid_record +1 def _parseL(self, field_list): """_parseL(self, field_list) field_list: field list of the record 0- L: Sheet of Conditions 1 A: 1- Empty 2- {Section Code\Section Title} B: 1- Record Code 2- {Section Code\Section Text} 3- {Section Code\RTF file} 4- {Section Code\HTM file} """ # _____Number of fields_____ # The record must have at least 3 fields if len(field_list) < 3: return _code = field_list[1] if _code == "": # A: Section Titles # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:3] # _____Fields_____ _section_codes = field_list[1] # last \ is erased if len(_section_codes) and _section_codes[-1] == "\\": _section_codes = _section_codes[:-1] _section_codes = _section_codes.split("\\") _section_dict = {} _section_index = 0 while _section_index < len(_section_codes)-1: # _____subfields_____ _section_code = _section_codes[_section_index] _section_title = _section_codes[_section_index+1] # control character are erased: end of line, tab, space # _____section_code_____ _section_code = self.delete_control_space(_section_code) # _____section_title_____ _section_title = self.delete_control_space(_section_title) if _section_code != "": _section_dict[_section_code] = _section_title _section_index = _section_index + 2 self.__budget.setSheetSections(_section_dict) self.num_valid_record = self.num_valid_record +1 else: # Any INFORMATION after last field separator is ignored # The record must have 5 fields if len(field_list) > 5: field_list = field_list[0:5] field_list = field_list[1:] # _____Fields_____ # _____Record Code_____ _record_code = self.delete_control_space(field_list[0]) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _record_code = self.validateCode(_record_code) _scodes_text = field_list[1] if _scodes_text == "": # TODO: rtf and html files print "Html and rtf files not implemented in ~L record" else: # _____Section-code_Section-text_____ # last \ is erased if len(_scodes_text) and _scodes_text[-1] == "\\": _scodes_text = _scodes_text[:-1] _scodes_text = _scodes_text.split("\\") _paragraph_dict = {} _section_dict = {} _section_index = 0 while _section_index < len(_scodes_text)-1: # _____subfields_____ _section_code = _scodes_text[_section_index] _section_text = _scodes_text[_section_index+1] # control character are erased: end of line, tab, space # _____section_code_____ _section_code = self.delete_control_space(_section_code) # _____section_text_____ if _section_code != "" and _section_text != "": #-# paragraph #-# _paragraph_code = _record_code + _section_code + "*" _paragraph_dict[ _paragraph_code ] = _section_text _section_dict[_section_code] = _paragraph_code _section_index = _section_index + 2 self.__budget.setSheetParagraphs(_paragraph_dict) self.__budget.setSheetRecord(_record_code, "*", _section_dict) self.num_valid_record = self.num_valid_record +1 def _parseQ(self, field_list): """_parseQ(self, field_list) field_list: field list of the record 0- Q: Sheet of Conditions 2 1- Record Code 2- {Section Code\Paragraph key\{Field key;}\}| """ # _____Number of fields_____ # The record must have at least 3 fields if len(field_list) < 3: return _code = field_list[1] # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] # _____Fields_____ # _____Record Code_____ _record_code = self.delete_control_space(field_list[0]) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _record_code = self.validateCode(_record_code) _scodes_pkey = field_list[1] # last \ is erased if len(_scodes_pkey) and _scodes_pkey[-1] == "\\": _scodes_pkey = _scodes_pkey[:-1] _scodes_pkey = _scodes_pkey.split("\\") _field_dict = {} _section_index = 0 while _section_index < len(_scodes_pkey) -1: # _____subfields_____ _section_code = _scodes_pkey[_section_index] _paragraph_key = _scodes_text[_section_index+1] _field_keys = _scodes_text[_section_index+2] # control character are erased: end of line, tab, space # _____section_code_____ _section_code = self.delete_control_space(_section_code) # _____section_text_____ _paragraph_key = self.delete_control_space(_paragraph_key) # _____Fields keys_____ _field_keys = self.delete_control_space(_field_keys) # last ; is erased if len(_field_keys) and _field_keys[-1] == ";": _field_keys = _field_keys[:-1] _field_keys_list = _scodes_pkey.split(";") for _field_key in _field_keys_list: if _field_key != "" and _section_code != "" and \ _paragraph_key != "": if _field_key in _field_dict: _section_dict = _field_dict[_field_key] else: _section_dict = {} _field_dict[_field_key] = _section_dict _section_dict[_section_code] = _paragraph_code _section_index = _section_index + 3 for _field, _section_dict in _field_dict.iteritems(): self.__budget.setSheetRecord(_record_code, _field, _section_dict) self.num_valid_record = self.num_valid_record +1 def _parseJ(self, field_list): """_parseJ(self, field_list) field_list: field list of the record 0- J: Sheet of Conditions 3 1- Paragraph code 2- [Paragraph text] 3- [RTF file] 4- [HTML file] """ # _____Number of fields_____ # The record must have at least 3 fields if len(field_list) < 3: return # Any INFORMATION after last field separator is ignored # The record must have 5 fields if len(field_list) > 5: field_list = field_list[0:5] field_list = field_list[1:] # _____Fields_____ # _____Paragraph code_____ _paragraph_code = self.delete_control_space(field_list[0]) # _____Paragraph text_____ _paragraph_text = field_list[1] if _paragraph_text == "": # TODO: rtf and html files print "Html and rtf files not implemented in ~J record" else: self.__budget.setSheetParagraph(paragraph_code, paragraph_text) self.num_valid_record = self.num_valid_record +1 def _parseG(self, field_list): """_parseG(self, field_list) field_list: field list of the record 0- G: Grafic info 1- record code 2- <grafic_file.ext\> """ # _____Number of fields_____ # The record must have at least 3 fields if len(field_list) < 3: return # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] # _____Fields_____ # _____Record Code_____ _record_code = self.delete_control_space(field_list[0]) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _record_code = self.validateCode(_record_code) # _____Grafic files_____ _grafic_files = self.delete_control(field_list[1]) # _____subfields_____ # last \ is erased if len(_grafic_files) and _grafic_files[-1] == "\\": _grafic_files = _grafic_files[:-1] _grafic_file_list = _grafic_files.split("\\") _tested_grafic_file_list = [] for _grafic_file in _grafic_file_list: _path = os.path.dirname(self.__filename) _grafic_file_path = os.path.join(_path, _grafic_file) if os.path.exists(_grafic_file_path): _tested_grafic_file_list.append(_grafic_file_path) else: _name_ext = os.path.splitext(_grafic_file) _grafic_file_name = _name_ext[0] _grafic_file_ext = _name_ext[1] _grafic_file_name_u = _grafic_file_name.upper() _grafic_file_name_l = _grafic_file_name.lower() _grafic_file_ext_u = _grafic_file_ext.upper() _grafic_file_ext_l = _grafic_file_ext.lower() _uu = _grafic_file_name_u + _grafic_file_ext_u _ul = _grafic_file_name_u + _grafic_file_ext_l _lu = _grafic_file_name_l + _grafic_file_ext_u _ll = _grafic_file_name_l + _grafic_file_ext_l _grafic_file_path_uu = os.path.join(_path, _uu) _grafic_file_path_ul = os.path.join(_path, _ul) _grafic_file_path_lu = os.path.join(_path, _lu) _grafic_file_path_ll = os.path.join(_path, _ll) if os.path.exists(_grafic_file_path_uu): _tested_grafic_file_list.append(_grafic_file_path_uu) elif os.path.exists(_grafic_file_path_ul): _tested_grafic_file_list.append(_grafic_file_path_ul) elif os.path.exists(_grafic_file_path_lu): _tested_grafic_file_list.append(_grafic_file_path_lu) elif os.path.exists(_grafic_file_path_ll): _tested_grafic_file_list.append(_grafic_file_path_ll) else: print utils.mapping(_("The file $1 do not exist"), (_grafic_file_path,)) if len(_grafic_file_list) > 0: for _grafic_file in _tested_grafic_file_list: self.__budget.addFile(_record_code, _grafic_file, "img", "") self.num_valid_record = self.num_valid_record +1 def _parseE(self, field_list): """_parseE(self, field_list) field_list: field list of the record 0- E: Company 1- company Code 2 [ summary ] 3- [ name ] 4- { [ type ] \ [ subname ] \ [ address ] \ [ postal_code ] \ [ town ] \ [ province ] \ [ country ] \ { phone; } \ { fax; } \ {contact_person; } \ } 5- [ cif ] \ [ web ] \ [ email ] \ """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 6 fields if len(field_list) > 6: field_list = field_list[1:6] # If there are no sufficient fields, the fields are added # with empty value:"" else: field_list = field_list[1:] + [""]*(6-len(field_list)) # _____Fields_____ # _____company Code_____ _company_code = self.delete_control_space(field_list[0]) if _company_code == "": return # _____Summary_____ _sumamary = self.delete_control(field_list[1]) # _____Name_____ _name = self.delete_control(field_list[2]) # _____local_offices_____ _local_offices = self.delete_control(field_list[3]) # _____subfields of local_offices_____ # last \ is erased if len(_local_offices) and _local_offices[-1] == "\\": _local_offices = _local_offices[:-1] _local_offices_list = _local_offices.split("\\") # If there are no sufficent subfields, the subfields are added # whith empty value _nsub = len(_local_offices_list) % 10 if _nsub != 0: _local_offices_list = _local_offices_list + \ [""]*(10-len(field_list)) _local_offices = [] _local_offices_index = 0 while _local_offices_index < len(_local_offices_list)-9: # _____subfields_____ _type = _local_offices_list[_local_offices_index] _subname = _local_offices_list[_local_offices_index+1] _address = _local_offices_list[_local_offices_index+2] _postal_code = _local_offices_list[_local_offices_index+3] _town = _local_offices_list[_local_offices_index+4] _province = _local_offices_list[_local_offices_index+5] _country = _local_offices_list[_local_offices_index+6] _phone = _local_offices_list[_local_offices_index+7] # last ; is erased if len(_phone) and _phone[-1] == ";": _phone = _phone[:-1] _phone_list = _phone.split(";") _fax = _local_offices_list[_local_offices_index+8] # last ; is erased if len(_fax) and _fax[-1] == ";": _fax = _fax[:-1] _fax_list = _fax.split(";") _contact_person = _local_offices_list[_local_offices_index+9] if _type != "" or _subname != "" or _address != "" or \ _postal_code != "" or _town != "" or _province != "" or \ _country != "" or _phone != "" or _fax != "" or \ _contact_person != "": _local_offices.append([_type, _subname, _address, _postal_code, _town, _province, _country, _phone_list, _fax_list, _contact_person]) _local_offices_index = _local_offices_index + 10 # _____cif web email_____ _c_w_e = self.delete_control_space(field_list[4]) # last \ is erased if len(_c_w_e) and _c_w_e[-1] == "\\": _c_w_e = _c_w_e[:-1] _c_w_e_list = _c_w_e.split("\\") # _____subfields_____ # If there are no sufficient fields, the fields are added # with empty value:"" _c_w_e_list = _c_w_e_list + [""]*(3-len(_c_w_e_list)) _cif = _c_w_e_list[0] _web = _c_w_e_list[1] _email = _c_w_e_list[2] self.__budget.setCompany(_company_code, _sumamary, _name, _local_offices, _cif, _web, _email) self.num_valid_record = self.num_valid_record +1 def _parseX(self, field_list): """_parseX(self, field_list) field_list: field list of the record A) 0- X: Tecnical information 1- Empty 2- < TI_Code \ TI_Descitption \ TI_Unit > B) 0- X: Tecnical information 1- Record_code 2- < TI_Code \ TI_value > """ # Tecnical information # The record must have at least 3 fields if len(field_list) < 3: return # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] # _____Fields_____ # "control": "[\t \n\r]" _field_1 = self.delete_control_space(field_list[0]) _field_2 = self.delete_control_space(field_list[1]) if _field_1 == "": # A) _field_2_list = _field_2.split("\\") _ti_index = 0 while _ti_index < len(_field_2_list)-3: _ti_code = _field_2_list[_ti_index] _ti_description = _field_2_list[_ti_index+1] _ti_unit = _field_2_list[_ti_index+2] if _ti_code != "": self.__budget.addTecInfo(_ti_code, _ti_description, _ti_unit) _ti_index = _ti_index + 3 else: # B) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _record_code = self.validateCode(_field_1) _field_2_list = _field_2.split("\\") _ti_index = 0 _ti_dict = {} while _ti_index < len(_field_2_list)-2: _ti_code = _field_2_list[_ti_index] _ti_value = _field_2_list[_ti_index+1] if _ti_code != "" and _ty_value != "": _ti_dict[_ti_code] = _ty_value _ti_index = _ti_index + 2 self.__budget.setTecnicalInformation(_record_code, _ti_dict) self.num_valid_record = self.num_valid_record +1 def _parseF(self, field_list): """_parseF(self, field_list) field_list: field list of the record 0- F: Files 1- Record code 2- { Type \ { Filenames; } \ [Description] } """ # _____Number of fields_____ # The record must have at least 3 fields if len(field_list) < 3: return # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] # _____Fields_____ # _____Record Code_____ _record_code = self.delete_control_space(field_list[0]) # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _record_code = self.validateCode(_record_code) # _____Grafic files_____ _files = self.delete_control(field_list[1]) # _____subfields_____ # last \ is erased if len(_files) and _files[-1] == "\\": _files = _files[:-1] _files_list = _files.split("\\") # adding empty subfiels if necesary if len(_files_list)%3 > 0: _files_list.extend[""]*(3 - len(_files_list)%3) _file_index = 0 _tested_files_list = [] while _file_index < len(_files_list)-3: _type = _files_list[_file_index].replace(" ","") ## _types = { ## "0": _("others"), ## "1": _("características técnicas y de fabricación"), ## "2": _("manual de colocación, uso y mantenimiento"), ## "3": _("certificado/s de elementos y sistemas"), ## "4": _("normativa y bibliografía"), ## "5": _("tarifa de precios"), ## "6": _("condiciones de venta"), ## "7": _("carta de colores"), ## "8": _("ámbito de aplicación y criterios selección"), ## "9": _("cálculo de elementos y sistemas"), ## "10": _("presentación, datos generales, objetivos, etc. de "\ ## "empresa"), ## "11": _("certificado/s de empresa"), ## "12": _("obras realizadas")} _types = ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12"] if not _type in _types: _type = "0" _filenames = _files_list[_file_index + 1] _description = _files_list[_file_index + 2] _file_index += 3 if len(_filenames) and _filenames[-1] == ";": _files = _files[:-1] _filenames_list = _files.split(";") _path = os.path.dirname(self.__filename) for _filename in filenames_list: _file_path = os.path.join(_path, _filename) if os.path.exists(_file_path): _tested_files_list.append([_file_path, _type, _description]) else: _name_ext = os.path.splitext(_filename) _file_name = _name_ext[0] _file_ext = _name_ext[1] _file_name_u = _file_name.upper() _file_name_l = _file_name.lower() _file_ext_u = _file_ext.upper() _file_ext_l = _file_ext.lower() _uu = _file_name_u + _file_ext_u _ul = _file_name_u + _file_ext_l _lu = _file_name_l + _file_ext_u _ll = _file_name_l + _file_ext_l _file_path_uu = os.path.join(_path, _uu) _file_path_ul = os.path.join(_path, _ul) _file_path_lu = os.path.join(_path, _lu) _file_path_ll = os.path.join(_path, _ll) if os.path.exists(_file_path_uu): _tested_files_list.append([_file_path_uu, _type, _description]) elif os.path.exists(_grafic_file_path_ul): _tested_files_list.append([_file_path_ul, _type, _description]) elif os.path.exists(_grafic_file_path_lu): _tested_files_list.append([_file_path_lu, _type, _description]) elif os.path.exists(_grafic_file_path_ll): _tested_files_list.append([_file_path_ll, _type, _description]) else: print utils.mapping(_("The file $1 do not exist"), (_file_path,)) if len(_tested_files_list) > 0: for _file in _tested_file_list: self.__budget.addFile(_record_code, _file[0], file[1], file[2]) self.num_valid_record = self.num_valid_record +1 def _parseB(self, field_list): """_parseB(self, field_list) field_list: field list of the record 0- B: Change code 1- Record Code 2- New code """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] if len(field_list) != 2: return # control character are erased: end of line, tab, space # _____Fields_____ _code = self.delete_control_space(field_list[0]) _new_code = self.delete_control_space(field_list[1]) # _____Codes_____ # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _code = self.validateCode(_code) _new_code = self.validateCode(_new_code) # change code self.__budget.changeCode(_code, _new_code) self.num_valid_record = self.num_valid_record + 1 def _parseA(self, field_list): """_parseA(self, field_list) field_list: field list of the record 0- A: Labels 1- Record Code 2- <Label\> """ # _____Number of fields_____ # Any INFORMATION after last field separator is ignored # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] if len(field_list) != 2: return # control character are erased: end of line, tab, space # _____Fields_____ # "control": "[\t \n\r]" _code = self.delete_control_space(field_list[0]) _labels = self.delete_control_space(field_list[1]) # _____Codes_____ # "#" and "##" characters at the end of the code are erased # invalid characters are also erased _code = self.validateCode(_code) # _____Labels_____ # last \ is erased # TODO: change the others parsers to this: while len(_labels) > 0 and _labels[-1] == "\\": _labels = _labels[:-1] # replace "_" to " " _labels = _labels.replace("_"," ") _label_list = _labels.split("\\") for _label in _label_list: self.__budget.addLabel(_code, _label) self.num_valid_record = self.num_valid_record + 1 def _parseP(self, field_list): """_parseP(self, field_list) field_list: Parametric record A) Global paremetric record 0- P: Parametric 1- Empty 2- [Parametric description] 3- [library.DLL] B) Family Parametric record 0- P: Parametric 1- Family Code 2- [Parametric description] """ # TODO: Use global parametric record if len(field_list) > 2: # delete control caracters and spaces _family_code = self.delete_control_space(field_list[1]) if _family_code == "": # A)Global paremetric record # The record must have 3 or 4 fields if len(field_list) > 4: field_list = field_list[0:4] field_list = field_list[1:] if len(field_list) == 2: field_list.append("") if len(field_list) != 3: return else: # B)Family Parametric record # The record must have 3 fields if len(field_list) > 3: field_list = field_list[0:3] field_list = field_list[1:] if len(field_list) != 2: print _("PyArq hates parametric DLLs") return else: return # _____Description_____ _description = field_list[1] if _description == "": print _("PyArq hates parametric DLLs") return # Adding last end of line _description = _description + "\r\n" # Delete comments # "comment" : "#.*\r\n" _description = self.__pattern["comment"].sub("\r\n",_description) # Tabs to spaces _description = _description.replace("\t"," ") # Delete empty lines # "empty_line": r"(\r\n) *\r\n" while self.__pattern["empty_line"].search(_description): _description = self.__pattern["empty_line"].sub( lambda x: x.groups()[0], _description) # Delete spaces before and after / # "space_before_backslash" : r"( )+\\" _description = self.__pattern["space_before_backslash"].sub( r"\\",_description) # "space_after_backslash" : r"\\( )+" _description = self.__pattern["space_after_backslash"].sub( r"\\",_description) # Join lines that start but not end with / _description = "\r\n" + _description # add leading end of line # "start_noend_backslash": "(\r\n\\\.*[^\\\])\r\n" while self.__pattern["start_noend_backslash"].search(_description): _description = self.__pattern["start_noend_backslash"].sub( lambda x: x.groups()[0], _description) # Join lines that end with a + - * / ^ and @ & < > <= >= = <> ! # "end_oper" : "(\+|-|\*|/|/^|@|&|<|>|<=|>=|=|!) *\r\n" _description = self.__pattern["end_oper"].sub( lambda x: x.groups()[0], _description) # Join lines for matricial vars # matricial_var : "(\r\n *[%|\$][A-ZÑ].*=.*,) *\r\n" while self.__pattern["matricial_var"].search(_description): _description = self.__pattern["matricial_var"].sub( lambda x: x.groups()[0], _description) _description = _description[2:] # remove leading end of line #_description = re.sub(r"\\( )+",r"\\",_description) _lines = _description.split("\r\n") _final_description = "" _pass_line = 0 for index in range(len(_lines)): _line = _lines[index] # Parse lines if len(_line) != 0: # Delete empty lines if _pass_line > 0: _pass_line = _pass_line -1 _line = "" elif _line.isspace(): _line = "" elif _line[0] != "\\": # Delete spaces out "" delimiter _list = _line.split('"') _final_line = "" for index1 in range(len(_list)): if index1 % 2 != 0: _parcial_line = '"' + _list[index1] else: _parcial_line = '"' + _list[index1].replace(" ","") _final_line = _final_line + _parcial_line _line = _final_line[1:] _lines[index] = _line # parse data if len(_line) > 2 and _line[:2] == "::": # Delete spaces out " delimiter #print "__PRECIO__" + _line[2:] pass elif len(_line) > 2 and _line[:2] == "%:": # Delete spaces out " delimiter #print "__%AUX__" + _line[2:] pass elif len(_line) > 3 and _line[:2] == "%%:": # Delete spaces out " delimiter #print "__%%AUX__" + _line[2:] pass elif self.__pattern["var"].search(_line): # Delete spaces out " delimiter #print "line =", _line while _line.count('"') % 2 == 1 and \ index + _pass_line + 1 < len(_lines) -1: _line = _line + _lines[index + _pass_line + 1] _pass_line = _pass_line + 1 _search = self.__pattern["var"].search(_line) if _search is not None: _var = _search.groups()[0] + " = " + _search.groups()[1] #print "__VAR__" + str(_var) pass else: #print "no __VAR__", _line pass elif self.__pattern["descomposition"].search(_line): # Delete spaces out " delimiter #_patern = "(^[^:]*):(.*)$" _search = self.__pattern["descomposition"].search(_line) if _search is not None: _var = _search.groups()[0] + ":" + _search.groups()[1] #print "__Descomposición__" + str(_var) pass else: #print "no __Descomposición__", _line pass else: print "Parametric: code: " + _family_code print "******* Desconocido *** : " + _line if index-10 > 0: print "-11 :", _lines[index-11] if index-10 > 0: print "-10 :", _lines[index-10] if index-9 > 0: print "-9 :", _lines[index-9] if index-8 > 0: print "-8 :", _lines[index-8] if index-7 > 0: print "-7 :", _lines[index-7] if index-6 > 0: print "-6 :", _lines[index-6] if index-5 > 0: print "-5 :", _lines[index-5] if index-4 > 0: print "-4 :", _lines[index-4] if index-3 > 0: print "-3 :", _lines[index-3] if index-2 > 0: print "-2 :", _lines[index-2] if index-1 > 0: print "-1 :", _lines[index-1] print "-0 :", _lines[index-0] pass else: _parameter_list = _line.split("\\")[1:-1] if len(_parameter_list) >= 2: if _parameter_list[0] == "C" or \ _parameter_list[0] == "COMENTARIO": #print "__COMENTARIO__" + _parameter_list[1] self.__budget.setParametricSelectComment( _family_code, _parameter_list[1]) elif _parameter_list[0] == "R" or \ _parameter_list[0] == "RESUMEN": #print "__RESUMEN__" + _parameter_list[1] self.__budget.setParametricSummary(_family_code, _parameter_list[1]) elif _parameter_list[0] == "T" or \ _parameter_list[0] == "TEXTO": #print "__TEXTO__" + _parameter_list[1] self.__budget.setParametricText(_family_code, _parameter_list[1]) elif _parameter_list[0] == "P" or \ _parameter_list[0] == "PLIEGO": #print "__PLIEGO__" + str(_parameter_list[1:]) pass elif _parameter_list[0] == "K" or \ _parameter_list[0] == "CLAVES": #print "__CLAVES__" + str(_parameter_list[1:]) pass elif _parameter_list[0] == "F" or \ _parameter_list[0] == "COMERCIAL": #print "__COMERCIAL__" + str(_parameter_list[1:]) pass else: #print "==PARAMETRO==" + str(_parameter_list[:]) pass _final_description = _final_description + _line + "\r\n" #print _line # Delete last empty line _description = _final_description[:-2] _lines = _description.split("\r\n") for _line in _lines: pass #print _line self.num_valid_record = self.num_valid_record + 1 def readFile(self, budget=None, filename=None, interface=None): """readFile(self, budget=None, filename=None) filename: the filename of the fiebdc file budget: base.obra object interface: a object to send messages must have printf(message) progress(percent) recordStatistics(...) Return the budget objetc or None if the file can be readed """ if not filename is None and not budget is None: self.__filename = filename self.__budget = budget self.__budget.filename = self.__filename if self.__filename is None or self.__budget is None or self.__cancel: return None if not os.path.exists(self.__filename): return None _time = time.time() try: _file = open(self.__filename, 'r') except IOError: print utils.mapping("IOError: $1", (self.__filename,)) return None self.__budget.filename = self.__filename self._record_number = 0 self.num_valid_record = 0 self._record_V_number = 0 self._record_C_number = 0 self._record_D_number = 0 self._record_Y_number = 0 self._record_M_number = 0 self._record_N_number = 0 self._record_T_number = 0 self._record_K_number = 0 self._record_W_number = 0 self._record_L_number = 0 self._record_Q_number = 0 self._record_J_number = 0 self._record_G_number = 0 self._record_E_number = 0 self._record_O_number = 0 self._record_P_number = 0 self._record_X_number = 0 self._record_B_number = 0 self._record_F_number = 0 self._record_A_number = 0 self._record_Unknow_number = 0 print utils.mapping(_("Loading file $1"), (self.__filename,)) _filesize = float(os.path.getsize(self.__filename)) interface.progress(_file.tell() / _filesize) _buffer = _file.read(1000) # set codepage from V record _record_list = _buffer.split("~") registro_V = _record_list[1] # ~V|[PROPIEDAD_ARCHIVO]|VERSION_FORMATO[\DDMMAAAA]|[PROGRAMA_EMISION]| # [CABECERA]\{ ROTULO_IDENTIFICACION \}|[JUEGO_CARACTERES]| # [COMENTARIO]|[TIPO INFORMACIÓN]|[NÚMERO CERTIFICACIÓN]| # [FECHA CERTIFICACIÓN ] | registro_V = registro_V.split("|") if registro_V[0] == "V": #_codepage = registro_V[5] if len(registro_V) > 5: _version = registro_V[5].strip() # remove leading spaces if _version in self.__character_sets_dict: self.__character_set = self.__character_sets_dict[_version] else: print utils.mapping(_("This codepage do not exist in "\ "FIEBDC3! Default codepage: $1"), (self.__character_set,)) else: print utils.mapping(_("This V record dot have a codepage! "\ "Default codepage: $1"), (self.__character_set,)) else: print utils.mapping(_("Not 'V' record in File! Default codepage: "\ "$1"), (self.__character_set,)) if self.__character_set != "utf8": _buffer = unicode(_buffer, self.__character_set) _buffer = _buffer.encode("utf8") # Any INFORMATION between the beginning of the file and the # beginning of the first registry “~” is ignored #"after_first_tilde" : "^[^~]*~" _buffer = self.__pattern["after_first_tilde"].sub("",_buffer) while _buffer != "" and not self.__cancel: #-# the blank characters (32), tabs (9) and end of line (13 and 10) # before the separators '~', '|' are erased. # Before separator \ not deleted because it affects the reading of # the record ~P _buffer = self.eraseControlCharacters(_buffer) _record_list = _buffer.split("~") # The last record can be incomplete unless it is the last one of # the file if len(_record_list) > 1: # not the end _last_record = _record_list.pop() else: # the end record # The blank characters (32), tabs (9) and end of line # (13 and 10) at the end of the file are ignored. #"end_control" : "((\r\n)| |\t)+$" _record_list[-1] = self.__pattern["end_control"].sub("", _record_list[-1]) _last_record = "" for record in _record_list: if self.__cancel: break self.parseRecord(record) interface.progress(_file.tell() / _filesize) _buffer2 = _file.read(100000) if self.__character_set != "utf8": _buffer2 = unicode(_buffer2, self.__character_set) _buffer2 = _buffer2.encode("utf8") _buffer = _last_record + _buffer2 _file.close() if self.__cancel: print _("Process terminated") return None else: print utils.mapping(_("Time to load: $1 seconds"), (("%.2f" %(time.time()-_time)),)) print utils.mapping(_("Records/Valid Records: $1/$2"), (self._record_number, self.num_valid_record)) if self._record_O_number > 0: print utils.mapping(_("$1 unsuported record type O: "\ "Comercial Relationship"), (self._record_O_number,)) if self.num_valid_record == 0: print _("This file is not a valid FIBDC3 file") return None _str = "" for type in \ [("V", self._record_V_number), ("C", self._record_C_number), ("D", self._record_D_number), ("Y", self._record_Y_number), ("M", self._record_M_number), ("N", self._record_N_number), ("T", self._record_T_number), ("K", self._record_K_number), ("W", self._record_W_number), ("L", self._record_L_number), ("Q", self._record_Q_number), ("J", self._record_J_number), ("G", self._record_G_number), ("E", self._record_E_number), ("O", self._record_O_number), ("P", self._record_P_number), ("X", self._record_X_number), ("B", self._record_B_number), ("F", self._record_F_number), ("A", self._record_A_number), ("?", self._record_Unknow_number)]: _str = _str + "%s: %s\n" %(type[0], type[1]) print _str self._testBudget(self.__budget) return self.__budget def _testBudget(self, budget): """testBudget(self,budget) budget: base.obra object Test and repair budget object after read it from bc3 file """ # TODO: more to do here print _("Testing budget ...") # Add price to records without price _iter = budget.iter() _titlelist = budget.getTitleList()[1] if len(_titlelist) == 0: _titlenum = 1 else: _titlenum = len(_titlelist) for _code in _iter: _record = budget.getRecord(_code) _prices = _record.getPrices() _len_prices = len(_prices) if _titlenum > _len_prices: _leftprices = _titlenum - _len_prices for _index in range(0,_leftprices): _root = budget.getRecord(budget.getRoot()) _price = [0.0, _root.getDate(_len_prices + _index)] budget.addPriceToRecord(_price,_record) print _("End Test") def delete_control_space(self, text): text = self.delete_control(text) text = text.replace(" ", "") return text def delete_control(self, text): text = text.replace("\t", "") text = text.replace("\r", "") text = text.replace("\n", "") return text