mirror of https://github.com/acidanthera/audk.git
1075 lines
49 KiB
Python
1075 lines
49 KiB
Python
## @file
|
|
# Collect all defined strings in multiple uni files.
|
|
#
|
|
# Copyright (c) 2014 - 2019, Intel Corporation. All rights reserved.<BR>
|
|
#
|
|
# SPDX-License-Identifier: BSD-2-Clause-Patent
|
|
#
|
|
"""
|
|
Collect all defined strings in multiple uni files
|
|
"""
|
|
from __future__ import print_function
|
|
|
|
##
|
|
# Import Modules
|
|
#
|
|
import os, codecs, re
|
|
import shlex
|
|
from Logger import ToolError
|
|
from Logger import Log as EdkLogger
|
|
from Logger import StringTable as ST
|
|
from Library.StringUtils import GetLineNo
|
|
from Library.Misc import PathClass
|
|
from Library.Misc import GetCharIndexOutStr
|
|
from Library import DataType as DT
|
|
from Library.ParserValidate import CheckUTF16FileHeader
|
|
|
|
##
|
|
# Static definitions
|
|
#
|
|
UNICODE_WIDE_CHAR = u'\\wide'
|
|
UNICODE_NARROW_CHAR = u'\\narrow'
|
|
UNICODE_NON_BREAKING_CHAR = u'\\nbr'
|
|
UNICODE_UNICODE_CR = '\r'
|
|
UNICODE_UNICODE_LF = '\n'
|
|
|
|
NARROW_CHAR = u'\uFFF0'
|
|
WIDE_CHAR = u'\uFFF1'
|
|
NON_BREAKING_CHAR = u'\uFFF2'
|
|
CR = u'\u000D'
|
|
LF = u'\u000A'
|
|
NULL = u'\u0000'
|
|
TAB = u'\t'
|
|
BACK_SPLASH = u'\\'
|
|
|
|
gLANG_CONV_TABLE = {'eng':'en', 'fra':'fr', \
|
|
'aar':'aa', 'abk':'ab', 'ave':'ae', 'afr':'af', 'aka':'ak', 'amh':'am', \
|
|
'arg':'an', 'ara':'ar', 'asm':'as', 'ava':'av', 'aym':'ay', 'aze':'az', \
|
|
'bak':'ba', 'bel':'be', 'bul':'bg', 'bih':'bh', 'bis':'bi', 'bam':'bm', \
|
|
'ben':'bn', 'bod':'bo', 'bre':'br', 'bos':'bs', 'cat':'ca', 'che':'ce', \
|
|
'cha':'ch', 'cos':'co', 'cre':'cr', 'ces':'cs', 'chu':'cu', 'chv':'cv', \
|
|
'cym':'cy', 'dan':'da', 'deu':'de', 'div':'dv', 'dzo':'dz', 'ewe':'ee', \
|
|
'ell':'el', 'epo':'eo', 'spa':'es', 'est':'et', 'eus':'eu', 'fas':'fa', \
|
|
'ful':'ff', 'fin':'fi', 'fij':'fj', 'fao':'fo', 'fry':'fy', 'gle':'ga', \
|
|
'gla':'gd', 'glg':'gl', 'grn':'gn', 'guj':'gu', 'glv':'gv', 'hau':'ha', \
|
|
'heb':'he', 'hin':'hi', 'hmo':'ho', 'hrv':'hr', 'hat':'ht', 'hun':'hu', \
|
|
'hye':'hy', 'her':'hz', 'ina':'ia', 'ind':'id', 'ile':'ie', 'ibo':'ig', \
|
|
'iii':'ii', 'ipk':'ik', 'ido':'io', 'isl':'is', 'ita':'it', 'iku':'iu', \
|
|
'jpn':'ja', 'jav':'jv', 'kat':'ka', 'kon':'kg', 'kik':'ki', 'kua':'kj', \
|
|
'kaz':'kk', 'kal':'kl', 'khm':'km', 'kan':'kn', 'kor':'ko', 'kau':'kr', \
|
|
'kas':'ks', 'kur':'ku', 'kom':'kv', 'cor':'kw', 'kir':'ky', 'lat':'la', \
|
|
'ltz':'lb', 'lug':'lg', 'lim':'li', 'lin':'ln', 'lao':'lo', 'lit':'lt', \
|
|
'lub':'lu', 'lav':'lv', 'mlg':'mg', 'mah':'mh', 'mri':'mi', 'mkd':'mk', \
|
|
'mal':'ml', 'mon':'mn', 'mar':'mr', 'msa':'ms', 'mlt':'mt', 'mya':'my', \
|
|
'nau':'na', 'nob':'nb', 'nde':'nd', 'nep':'ne', 'ndo':'ng', 'nld':'nl', \
|
|
'nno':'nn', 'nor':'no', 'nbl':'nr', 'nav':'nv', 'nya':'ny', 'oci':'oc', \
|
|
'oji':'oj', 'orm':'om', 'ori':'or', 'oss':'os', 'pan':'pa', 'pli':'pi', \
|
|
'pol':'pl', 'pus':'ps', 'por':'pt', 'que':'qu', 'roh':'rm', 'run':'rn', \
|
|
'ron':'ro', 'rus':'ru', 'kin':'rw', 'san':'sa', 'srd':'sc', 'snd':'sd', \
|
|
'sme':'se', 'sag':'sg', 'sin':'si', 'slk':'sk', 'slv':'sl', 'smo':'sm', \
|
|
'sna':'sn', 'som':'so', 'sqi':'sq', 'srp':'sr', 'ssw':'ss', 'sot':'st', \
|
|
'sun':'su', 'swe':'sv', 'swa':'sw', 'tam':'ta', 'tel':'te', 'tgk':'tg', \
|
|
'tha':'th', 'tir':'ti', 'tuk':'tk', 'tgl':'tl', 'tsn':'tn', 'ton':'to', \
|
|
'tur':'tr', 'tso':'ts', 'tat':'tt', 'twi':'tw', 'tah':'ty', 'uig':'ug', \
|
|
'ukr':'uk', 'urd':'ur', 'uzb':'uz', 'ven':'ve', 'vie':'vi', 'vol':'vo', \
|
|
'wln':'wa', 'wol':'wo', 'xho':'xh', 'yid':'yi', 'yor':'yo', 'zha':'za', \
|
|
'zho':'zh', 'zul':'zu'}
|
|
|
|
## Convert a python unicode string to a normal string
|
|
#
|
|
# Convert a python unicode string to a normal string
|
|
# UniToStr(u'I am a string') is 'I am a string'
|
|
#
|
|
# @param Uni: The python unicode string
|
|
#
|
|
# @retval: The formatted normal string
|
|
#
|
|
def UniToStr(Uni):
|
|
return repr(Uni)[2:-1]
|
|
|
|
## Convert a unicode string to a Hex list
|
|
#
|
|
# Convert a unicode string to a Hex list
|
|
# UniToHexList('ABC') is ['0x41', '0x00', '0x42', '0x00', '0x43', '0x00']
|
|
#
|
|
# @param Uni: The python unicode string
|
|
#
|
|
# @retval List: The formatted hex list
|
|
#
|
|
def UniToHexList(Uni):
|
|
List = []
|
|
for Item in Uni:
|
|
Temp = '%04X' % ord(Item)
|
|
List.append('0x' + Temp[2:4])
|
|
List.append('0x' + Temp[0:2])
|
|
return List
|
|
|
|
## Convert special unicode characters
|
|
#
|
|
# Convert special characters to (c), (r) and (tm).
|
|
#
|
|
# @param Uni: The python unicode string
|
|
#
|
|
# @retval NewUni: The converted unicode string
|
|
#
|
|
def ConvertSpecialUnicodes(Uni):
|
|
OldUni = NewUni = Uni
|
|
NewUni = NewUni.replace(u'\u00A9', '(c)')
|
|
NewUni = NewUni.replace(u'\u00AE', '(r)')
|
|
NewUni = NewUni.replace(u'\u2122', '(tm)')
|
|
if OldUni == NewUni:
|
|
NewUni = OldUni
|
|
return NewUni
|
|
|
|
## GetLanguageCode1766
|
|
#
|
|
# Check the language code read from .UNI file and convert RFC 4646 codes to RFC 1766 codes
|
|
# RFC 1766 language codes supported in compatibility mode
|
|
# RFC 4646 language codes supported in native mode
|
|
#
|
|
# @param LangName: Language codes read from .UNI file
|
|
#
|
|
# @retval LangName: Valid language code in RFC 1766 format or None
|
|
#
|
|
def GetLanguageCode1766(LangName, File=None):
|
|
return LangName
|
|
|
|
length = len(LangName)
|
|
if length == 2:
|
|
if LangName.isalpha():
|
|
for Key in gLANG_CONV_TABLE.keys():
|
|
if gLANG_CONV_TABLE.get(Key) == LangName.lower():
|
|
return Key
|
|
elif length == 3:
|
|
if LangName.isalpha() and gLANG_CONV_TABLE.get(LangName.lower()):
|
|
return LangName
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FORMAT_INVALID,
|
|
"Invalid RFC 1766 language code : %s" % LangName,
|
|
File)
|
|
elif length == 5:
|
|
if LangName[0:2].isalpha() and LangName[2] == '-':
|
|
for Key in gLANG_CONV_TABLE.keys():
|
|
if gLANG_CONV_TABLE.get(Key) == LangName[0:2].lower():
|
|
return Key
|
|
elif length >= 6:
|
|
if LangName[0:2].isalpha() and LangName[2] == '-':
|
|
for Key in gLANG_CONV_TABLE.keys():
|
|
if gLANG_CONV_TABLE.get(Key) == LangName[0:2].lower():
|
|
return Key
|
|
if LangName[0:3].isalpha() and gLANG_CONV_TABLE.get(LangName.lower()) is None and LangName[3] == '-':
|
|
for Key in gLANG_CONV_TABLE.keys():
|
|
if Key == LangName[0:3].lower():
|
|
return Key
|
|
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FORMAT_INVALID,
|
|
"Invalid RFC 4646 language code : %s" % LangName,
|
|
File)
|
|
|
|
## GetLanguageCode
|
|
#
|
|
# Check the language code read from .UNI file and convert RFC 1766 codes to RFC 4646 codes if appropriate
|
|
# RFC 1766 language codes supported in compatibility mode
|
|
# RFC 4646 language codes supported in native mode
|
|
#
|
|
# @param LangName: Language codes read from .UNI file
|
|
#
|
|
# @retval LangName: Valid lanugage code in RFC 4646 format or None
|
|
#
|
|
def GetLanguageCode(LangName, IsCompatibleMode, File):
|
|
length = len(LangName)
|
|
if IsCompatibleMode:
|
|
if length == 3 and LangName.isalpha():
|
|
TempLangName = gLANG_CONV_TABLE.get(LangName.lower())
|
|
if TempLangName is not None:
|
|
return TempLangName
|
|
return LangName
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FORMAT_INVALID,
|
|
"Invalid RFC 1766 language code : %s" % LangName,
|
|
File)
|
|
if (LangName[0] == 'X' or LangName[0] == 'x') and LangName[1] == '-':
|
|
return LangName
|
|
if length == 2:
|
|
if LangName.isalpha():
|
|
return LangName
|
|
elif length == 3:
|
|
if LangName.isalpha() and gLANG_CONV_TABLE.get(LangName.lower()) is None:
|
|
return LangName
|
|
elif length == 5:
|
|
if LangName[0:2].isalpha() and LangName[2] == '-':
|
|
return LangName
|
|
elif length >= 6:
|
|
if LangName[0:2].isalpha() and LangName[2] == '-':
|
|
return LangName
|
|
if LangName[0:3].isalpha() and gLANG_CONV_TABLE.get(LangName.lower()) is None and LangName[3] == '-':
|
|
return LangName
|
|
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FORMAT_INVALID,
|
|
"Invalid RFC 4646 language code : %s" % LangName,
|
|
File)
|
|
|
|
## FormatUniEntry
|
|
#
|
|
# Formatted the entry in Uni file.
|
|
#
|
|
# @param StrTokenName StrTokenName.
|
|
# @param TokenValueList A list need to be processed.
|
|
# @param ContainerFile ContainerFile.
|
|
#
|
|
# @return formatted entry
|
|
def FormatUniEntry(StrTokenName, TokenValueList, ContainerFile):
|
|
SubContent = ''
|
|
PreFormatLength = 40
|
|
if len(StrTokenName) > PreFormatLength:
|
|
PreFormatLength = len(StrTokenName) + 1
|
|
for (Lang, Value) in TokenValueList:
|
|
if not Value or Lang == DT.TAB_LANGUAGE_EN_X:
|
|
continue
|
|
if Lang == '':
|
|
Lang = DT.TAB_LANGUAGE_EN_US
|
|
if Lang == 'eng':
|
|
Lang = DT.TAB_LANGUAGE_EN_US
|
|
elif len(Lang.split('-')[0]) == 3:
|
|
Lang = GetLanguageCode(Lang.split('-')[0], True, ContainerFile)
|
|
else:
|
|
Lang = GetLanguageCode(Lang, False, ContainerFile)
|
|
ValueList = Value.split('\n')
|
|
SubValueContent = ''
|
|
for SubValue in ValueList:
|
|
if SubValue.strip():
|
|
SubValueContent += \
|
|
' ' * (PreFormatLength + len('#language en-US ')) + '\"%s\\n\"' % SubValue.strip() + '\r\n'
|
|
SubValueContent = SubValueContent[(PreFormatLength + len('#language en-US ')):SubValueContent.rfind('\\n')] \
|
|
+ '\"' + '\r\n'
|
|
SubContent += ' '*PreFormatLength + '#language %-5s ' % Lang + SubValueContent
|
|
if SubContent:
|
|
SubContent = StrTokenName + ' '*(PreFormatLength - len(StrTokenName)) + SubContent[PreFormatLength:]
|
|
return SubContent
|
|
|
|
|
|
## StringDefClassObject
|
|
#
|
|
# A structure for language definition
|
|
#
|
|
class StringDefClassObject(object):
|
|
def __init__(self, Name = None, Value = None, Referenced = False, Token = None, UseOtherLangDef = ''):
|
|
self.StringName = ''
|
|
self.StringNameByteList = []
|
|
self.StringValue = ''
|
|
self.StringValueByteList = ''
|
|
self.Token = 0
|
|
self.Referenced = Referenced
|
|
self.UseOtherLangDef = UseOtherLangDef
|
|
self.Length = 0
|
|
|
|
if Name is not None:
|
|
self.StringName = Name
|
|
self.StringNameByteList = UniToHexList(Name)
|
|
if Value is not None:
|
|
self.StringValue = Value
|
|
self.StringValueByteList = UniToHexList(self.StringValue)
|
|
self.Length = len(self.StringValueByteList)
|
|
if Token is not None:
|
|
self.Token = Token
|
|
|
|
def __str__(self):
|
|
return repr(self.StringName) + ' ' + \
|
|
repr(self.Token) + ' ' + \
|
|
repr(self.Referenced) + ' ' + \
|
|
repr(self.StringValue) + ' ' + \
|
|
repr(self.UseOtherLangDef)
|
|
|
|
def UpdateValue(self, Value = None):
|
|
if Value is not None:
|
|
if self.StringValue:
|
|
self.StringValue = self.StringValue + '\r\n' + Value
|
|
else:
|
|
self.StringValue = Value
|
|
self.StringValueByteList = UniToHexList(self.StringValue)
|
|
self.Length = len(self.StringValueByteList)
|
|
|
|
## UniFileClassObject
|
|
#
|
|
# A structure for .uni file definition
|
|
#
|
|
class UniFileClassObject(object):
|
|
def __init__(self, FileList = None, IsCompatibleMode = False, IncludePathList = None):
|
|
self.FileList = FileList
|
|
self.File = None
|
|
self.IncFileList = FileList
|
|
self.UniFileHeader = ''
|
|
self.Token = 2
|
|
self.LanguageDef = [] #[ [u'LanguageIdentifier', u'PrintableName'], ... ]
|
|
self.OrderedStringList = {} #{ u'LanguageIdentifier' : [StringDefClassObject] }
|
|
self.OrderedStringDict = {} #{ u'LanguageIdentifier' : {StringName:(IndexInList)} }
|
|
self.OrderedStringListByToken = {} #{ u'LanguageIdentifier' : {Token: StringDefClassObject} }
|
|
self.IsCompatibleMode = IsCompatibleMode
|
|
if not IncludePathList:
|
|
self.IncludePathList = []
|
|
else:
|
|
self.IncludePathList = IncludePathList
|
|
if len(self.FileList) > 0:
|
|
self.LoadUniFiles(FileList)
|
|
|
|
#
|
|
# Get Language definition
|
|
#
|
|
def GetLangDef(self, File, Line):
|
|
Lang = shlex.split(Line.split(u"//")[0])
|
|
if len(Lang) != 3:
|
|
try:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_8').readlines()
|
|
except UnicodeError as Xstr:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_16').readlines()
|
|
except UnicodeError as Xstr:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_16_le').readlines()
|
|
except:
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FILE_OPEN_FAILURE,
|
|
"File read failure: %s" % str(Xstr),
|
|
ExtraData=File)
|
|
LineNo = GetLineNo(FileIn, Line, False)
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.PARSER_ERROR,
|
|
"Wrong language definition",
|
|
ExtraData="""%s\n\t*Correct format is like '#langdef en-US "English"'""" % Line,
|
|
File = File, Line = LineNo)
|
|
else:
|
|
LangName = GetLanguageCode(Lang[1], self.IsCompatibleMode, self.File)
|
|
LangPrintName = Lang[2]
|
|
|
|
IsLangInDef = False
|
|
for Item in self.LanguageDef:
|
|
if Item[0] == LangName:
|
|
IsLangInDef = True
|
|
break
|
|
|
|
if not IsLangInDef:
|
|
self.LanguageDef.append([LangName, LangPrintName])
|
|
|
|
#
|
|
# Add language string
|
|
#
|
|
self.AddStringToList(u'$LANGUAGE_NAME', LangName, LangName, 0, True, Index=0)
|
|
self.AddStringToList(u'$PRINTABLE_LANGUAGE_NAME', LangName, LangPrintName, 1, True, Index=1)
|
|
|
|
if not IsLangInDef:
|
|
#
|
|
# The found STRING tokens will be added into new language string list
|
|
# so that the unique STRING identifier is reserved for all languages in the package list.
|
|
#
|
|
FirstLangName = self.LanguageDef[0][0]
|
|
if LangName != FirstLangName:
|
|
for Index in range (2, len (self.OrderedStringList[FirstLangName])):
|
|
Item = self.OrderedStringList[FirstLangName][Index]
|
|
if Item.UseOtherLangDef != '':
|
|
OtherLang = Item.UseOtherLangDef
|
|
else:
|
|
OtherLang = FirstLangName
|
|
self.OrderedStringList[LangName].append (StringDefClassObject(Item.StringName,
|
|
'',
|
|
Item.Referenced,
|
|
Item.Token,
|
|
OtherLang))
|
|
self.OrderedStringDict[LangName][Item.StringName] = len(self.OrderedStringList[LangName]) - 1
|
|
return True
|
|
|
|
#
|
|
# Get String name and value
|
|
#
|
|
def GetStringObject(self, Item):
|
|
Language = ''
|
|
Value = ''
|
|
|
|
Name = Item.split()[1]
|
|
# Check the string name is the upper character
|
|
if Name != '':
|
|
MatchString = re.match('[A-Z0-9_]+', Name, re.UNICODE)
|
|
if MatchString is None or MatchString.end(0) != len(Name):
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FORMAT_INVALID,
|
|
'The string token name %s in UNI file %s must be upper case character.' %(Name, self.File))
|
|
LanguageList = Item.split(u'#language ')
|
|
for IndexI in range(len(LanguageList)):
|
|
if IndexI == 0:
|
|
continue
|
|
else:
|
|
Language = LanguageList[IndexI].split()[0]
|
|
#.replace(u'\r\n', u'')
|
|
Value = \
|
|
LanguageList[IndexI][LanguageList[IndexI].find(u'\"') + len(u'\"') : LanguageList[IndexI].rfind(u'\"')]
|
|
Language = GetLanguageCode(Language, self.IsCompatibleMode, self.File)
|
|
self.AddStringToList(Name, Language, Value)
|
|
|
|
#
|
|
# Get include file list and load them
|
|
#
|
|
def GetIncludeFile(self, Item, Dir = None):
|
|
if Dir:
|
|
pass
|
|
FileName = Item[Item.find(u'!include ') + len(u'!include ') :Item.find(u' ', len(u'!include '))][1:-1]
|
|
self.LoadUniFile(FileName)
|
|
|
|
#
|
|
# Pre-process before parse .uni file
|
|
#
|
|
def PreProcess(self, File, IsIncludeFile=False):
|
|
if not os.path.exists(File.Path) or not os.path.isfile(File.Path):
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FILE_NOT_FOUND,
|
|
ExtraData=File.Path)
|
|
|
|
#
|
|
# Check file header of the Uni file
|
|
#
|
|
# if not CheckUTF16FileHeader(File.Path):
|
|
# EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
# ExtraData='The file %s is either invalid UTF-16LE or it is missing the BOM.' % File.Path)
|
|
|
|
try:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_8').readlines()
|
|
except UnicodeError as Xstr:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_16').readlines()
|
|
except UnicodeError:
|
|
FileIn = codecs.open(File.Path, mode='rb', encoding='utf_16_le').readlines()
|
|
except:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FILE_OPEN_FAILURE, ExtraData=File.Path)
|
|
|
|
|
|
#
|
|
# get the file header
|
|
#
|
|
Lines = []
|
|
HeaderStart = False
|
|
HeaderEnd = False
|
|
if not self.UniFileHeader:
|
|
FirstGenHeader = True
|
|
else:
|
|
FirstGenHeader = False
|
|
for Line in FileIn:
|
|
Line = Line.strip()
|
|
if Line == u'':
|
|
continue
|
|
if Line.startswith(DT.TAB_COMMENT_EDK1_SPLIT) and (Line.find(DT.TAB_HEADER_COMMENT) > -1) \
|
|
and not HeaderEnd and not HeaderStart:
|
|
HeaderStart = True
|
|
if not Line.startswith(DT.TAB_COMMENT_EDK1_SPLIT) and HeaderStart and not HeaderEnd:
|
|
HeaderEnd = True
|
|
if Line.startswith(DT.TAB_COMMENT_EDK1_SPLIT) and HeaderStart and not HeaderEnd and FirstGenHeader:
|
|
self.UniFileHeader += Line + '\r\n'
|
|
continue
|
|
|
|
#
|
|
# Use unique identifier
|
|
#
|
|
FindFlag = -1
|
|
LineCount = 0
|
|
MultiLineFeedExits = False
|
|
#
|
|
# 0: initial value
|
|
# 1: single String entry exist
|
|
# 2: line feed exist under the some single String entry
|
|
#
|
|
StringEntryExistsFlag = 0
|
|
for Line in FileIn:
|
|
Line = FileIn[LineCount]
|
|
LineCount += 1
|
|
Line = Line.strip()
|
|
#
|
|
# Ignore comment line and empty line
|
|
#
|
|
if Line == u'' or Line.startswith(u'//'):
|
|
#
|
|
# Change the single line String entry flag status
|
|
#
|
|
if StringEntryExistsFlag == 1:
|
|
StringEntryExistsFlag = 2
|
|
#
|
|
# If the '#string' line and the '#language' line are not in the same line,
|
|
# there should be only one line feed character between them
|
|
#
|
|
if MultiLineFeedExits:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
continue
|
|
|
|
MultiLineFeedExits = False
|
|
#
|
|
# Process comment embedded in string define lines
|
|
#
|
|
FindFlag = Line.find(u'//')
|
|
if FindFlag != -1 and Line.find(u'//') < Line.find(u'"'):
|
|
Line = Line.replace(Line[FindFlag:], u' ')
|
|
if FileIn[LineCount].strip().startswith('#language'):
|
|
Line = Line + FileIn[LineCount]
|
|
FileIn[LineCount-1] = Line
|
|
FileIn[LineCount] = '\r\n'
|
|
LineCount -= 1
|
|
for Index in range (LineCount + 1, len (FileIn) - 1):
|
|
if (Index == len(FileIn) -1):
|
|
FileIn[Index] = '\r\n'
|
|
else:
|
|
FileIn[Index] = FileIn[Index + 1]
|
|
continue
|
|
CommIndex = GetCharIndexOutStr(u'/', Line)
|
|
if CommIndex > -1:
|
|
if (len(Line) - 1) > CommIndex:
|
|
if Line[CommIndex+1] == u'/':
|
|
Line = Line[:CommIndex].strip()
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
Line = Line.replace(UNICODE_WIDE_CHAR, WIDE_CHAR)
|
|
Line = Line.replace(UNICODE_NARROW_CHAR, NARROW_CHAR)
|
|
Line = Line.replace(UNICODE_NON_BREAKING_CHAR, NON_BREAKING_CHAR)
|
|
|
|
Line = Line.replace(u'\\\\', u'\u0006')
|
|
Line = Line.replace(u'\\r\\n', CR + LF)
|
|
Line = Line.replace(u'\\n', CR + LF)
|
|
Line = Line.replace(u'\\r', CR)
|
|
Line = Line.replace(u'\\t', u'\t')
|
|
Line = Line.replace(u'''\"''', u'''"''')
|
|
Line = Line.replace(u'\t', u' ')
|
|
Line = Line.replace(u'\u0006', u'\\')
|
|
|
|
#
|
|
# Check if single line has correct '"'
|
|
#
|
|
if Line.startswith(u'#string') and Line.find(u'#language') > -1 and Line.find('"') > Line.find(u'#language'):
|
|
if not Line.endswith('"'):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
ExtraData='''The line %s misses '"' at the end of it in file %s'''
|
|
% (LineCount, File.Path))
|
|
|
|
#
|
|
# Between Name entry and Language entry can not contain line feed
|
|
#
|
|
if Line.startswith(u'#string') and Line.find(u'#language') == -1:
|
|
MultiLineFeedExits = True
|
|
|
|
if Line.startswith(u'#string') and Line.find(u'#language') > 0 and Line.find(u'"') < 0:
|
|
MultiLineFeedExits = True
|
|
|
|
#
|
|
# Between Language entry and String entry can not contain line feed
|
|
#
|
|
if Line.startswith(u'#language') and len(Line.split()) == 2:
|
|
MultiLineFeedExits = True
|
|
|
|
#
|
|
# Check the situation that there only has one '"' for the language entry
|
|
#
|
|
if Line.startswith(u'#string') and Line.find(u'#language') > 0 and Line.count(u'"') == 1:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
ExtraData='''The line %s misses '"' at the end of it in file %s'''
|
|
% (LineCount, File.Path))
|
|
|
|
#
|
|
# Check the situation that there has more than 2 '"' for the language entry
|
|
#
|
|
if Line.startswith(u'#string') and Line.find(u'#language') > 0 and Line.replace(u'\\"', '').count(u'"') > 2:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
ExtraData='''The line %s has more than 2 '"' for language entry in file %s'''
|
|
% (LineCount, File.Path))
|
|
|
|
#
|
|
# Between two String entry, can not contain line feed
|
|
#
|
|
if Line.startswith(u'"'):
|
|
if StringEntryExistsFlag == 2:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
Message=ST.ERR_UNIPARSE_LINEFEED_UP_EXIST % Line, ExtraData=File.Path)
|
|
|
|
StringEntryExistsFlag = 1
|
|
if not Line.endswith('"'):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
ExtraData='''The line %s misses '"' at the end of it in file %s'''
|
|
% (LineCount, File.Path))
|
|
|
|
#
|
|
# Check the situation that there has more than 2 '"' for the language entry
|
|
#
|
|
if Line.strip() and Line.replace(u'\\"', '').count(u'"') > 2:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
ExtraData='''The line %s has more than 2 '"' for language entry in file %s'''
|
|
% (LineCount, File.Path))
|
|
|
|
elif Line.startswith(u'#language'):
|
|
if StringEntryExistsFlag == 2:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
Message=ST.ERR_UNI_MISS_STRING_ENTRY % Line, ExtraData=File.Path)
|
|
StringEntryExistsFlag = 0
|
|
else:
|
|
StringEntryExistsFlag = 0
|
|
|
|
Lines.append(Line)
|
|
|
|
#
|
|
# Convert string def format as below
|
|
#
|
|
# #string MY_STRING_1
|
|
# #language eng
|
|
# "My first English string line 1"
|
|
# "My first English string line 2"
|
|
# #string MY_STRING_1
|
|
# #language spa
|
|
# "Mi segunda secuencia 1"
|
|
# "Mi segunda secuencia 2"
|
|
#
|
|
|
|
if not IsIncludeFile and not Lines:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_NO_SECTION_EXIST, \
|
|
ExtraData=File.Path)
|
|
|
|
NewLines = []
|
|
StrName = u''
|
|
ExistStrNameList = []
|
|
for Line in Lines:
|
|
if StrName and not StrName.split()[1].startswith(DT.TAB_STR_TOKENCNAME + DT.TAB_UNDERLINE_SPLIT):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_STRNAME_FORMAT_ERROR % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
|
|
if StrName and len(StrName.split()[1].split(DT.TAB_UNDERLINE_SPLIT)) == 4:
|
|
StringTokenList = StrName.split()[1].split(DT.TAB_UNDERLINE_SPLIT)
|
|
if (StringTokenList[3].upper() in [DT.TAB_STR_TOKENPROMPT, DT.TAB_STR_TOKENHELP] and \
|
|
StringTokenList[3] not in [DT.TAB_STR_TOKENPROMPT, DT.TAB_STR_TOKENHELP]) or \
|
|
(StringTokenList[2].upper() == DT.TAB_STR_TOKENERR and StringTokenList[2] != DT.TAB_STR_TOKENERR):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_STRTOKEN_FORMAT_ERROR % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
|
|
if Line.count(u'#language') > 1:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_SEP_LANGENTRY_LINE % Line, \
|
|
ExtraData=File.Path)
|
|
|
|
if Line.startswith(u'//'):
|
|
continue
|
|
elif Line.startswith(u'#langdef'):
|
|
if len(Line.split()) == 2:
|
|
NewLines.append(Line)
|
|
continue
|
|
elif len(Line.split()) > 2 and Line.find(u'"') > 0:
|
|
NewLines.append(Line[:Line.find(u'"')].strip())
|
|
NewLines.append(Line[Line.find(u'"'):])
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
elif Line.startswith(u'#string'):
|
|
if len(Line.split()) == 2:
|
|
StrName = Line
|
|
if StrName:
|
|
if StrName.split()[1] not in ExistStrNameList:
|
|
ExistStrNameList.append(StrName.split()[1].strip())
|
|
elif StrName.split()[1] in [DT.TAB_INF_ABSTRACT, DT.TAB_INF_DESCRIPTION, \
|
|
DT.TAB_INF_BINARY_ABSTRACT, DT.TAB_INF_BINARY_DESCRIPTION, \
|
|
DT.TAB_DEC_PACKAGE_ABSTRACT, DT.TAB_DEC_PACKAGE_DESCRIPTION, \
|
|
DT.TAB_DEC_BINARY_ABSTRACT, DT.TAB_DEC_BINARY_DESCRIPTION]:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_MULTI_ENTRY_EXIST % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
continue
|
|
elif len(Line.split()) == 4 and Line.find(u'#language') > 0:
|
|
if Line[Line.find(u'#language')-1] != ' ' or \
|
|
Line[Line.find(u'#language')+len(u'#language')] != u' ':
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
if Line.find(u'"') > 0:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
StrName = Line.split()[0] + u' ' + Line.split()[1]
|
|
if StrName:
|
|
if StrName.split()[1] not in ExistStrNameList:
|
|
ExistStrNameList.append(StrName.split()[1].strip())
|
|
elif StrName.split()[1] in [DT.TAB_INF_ABSTRACT, DT.TAB_INF_DESCRIPTION, \
|
|
DT.TAB_INF_BINARY_ABSTRACT, DT.TAB_INF_BINARY_DESCRIPTION, \
|
|
DT.TAB_DEC_PACKAGE_ABSTRACT, DT.TAB_DEC_PACKAGE_DESCRIPTION, \
|
|
DT.TAB_DEC_BINARY_ABSTRACT, DT.TAB_DEC_BINARY_DESCRIPTION]:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_MULTI_ENTRY_EXIST % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
if IsIncludeFile:
|
|
if StrName not in NewLines:
|
|
NewLines.append((Line[:Line.find(u'#language')]).strip())
|
|
else:
|
|
NewLines.append((Line[:Line.find(u'#language')]).strip())
|
|
NewLines.append((Line[Line.find(u'#language'):]).strip())
|
|
elif len(Line.split()) > 4 and Line.find(u'#language') > 0 and Line.find(u'"') > 0:
|
|
if Line[Line.find(u'#language')-1] != u' ' or \
|
|
Line[Line.find(u'#language')+len(u'#language')] != u' ':
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
if Line[Line.find(u'"')-1] != u' ':
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
StrName = Line.split()[0] + u' ' + Line.split()[1]
|
|
if StrName:
|
|
if StrName.split()[1] not in ExistStrNameList:
|
|
ExistStrNameList.append(StrName.split()[1].strip())
|
|
elif StrName.split()[1] in [DT.TAB_INF_ABSTRACT, DT.TAB_INF_DESCRIPTION, \
|
|
DT.TAB_INF_BINARY_ABSTRACT, DT.TAB_INF_BINARY_DESCRIPTION, \
|
|
DT.TAB_DEC_PACKAGE_ABSTRACT, DT.TAB_DEC_PACKAGE_DESCRIPTION, \
|
|
DT.TAB_DEC_BINARY_ABSTRACT, DT.TAB_DEC_BINARY_DESCRIPTION]:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_MULTI_ENTRY_EXIST % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
if IsIncludeFile:
|
|
if StrName not in NewLines:
|
|
NewLines.append((Line[:Line.find(u'#language')]).strip())
|
|
else:
|
|
NewLines.append((Line[:Line.find(u'#language')]).strip())
|
|
NewLines.append((Line[Line.find(u'#language'):Line.find(u'"')]).strip())
|
|
NewLines.append((Line[Line.find(u'"'):]).strip())
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
elif Line.startswith(u'#language'):
|
|
if len(Line.split()) == 2:
|
|
if IsIncludeFile:
|
|
if StrName not in NewLines:
|
|
NewLines.append(StrName)
|
|
else:
|
|
NewLines.append(StrName)
|
|
NewLines.append(Line)
|
|
elif len(Line.split()) > 2 and Line.find(u'"') > 0:
|
|
if IsIncludeFile:
|
|
if StrName not in NewLines:
|
|
NewLines.append(StrName)
|
|
else:
|
|
NewLines.append(StrName)
|
|
NewLines.append((Line[:Line.find(u'"')]).strip())
|
|
NewLines.append((Line[Line.find(u'"'):]).strip())
|
|
else:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
elif Line.startswith(u'"'):
|
|
if u'#string' in Line or u'#language' in Line:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
NewLines.append(Line)
|
|
else:
|
|
print(Line)
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, ExtraData=File.Path)
|
|
|
|
if StrName and not StrName.split()[1].startswith(u'STR_'):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_STRNAME_FORMAT_ERROR % StrName.split()[1], \
|
|
ExtraData=File.Path)
|
|
|
|
if StrName and not NewLines:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNI_MISS_LANGENTRY % StrName, \
|
|
ExtraData=File.Path)
|
|
|
|
#
|
|
# Check Abstract, Description, BinaryAbstract and BinaryDescription order,
|
|
# should be Abstract, Description, BinaryAbstract, BinaryDescription
|
|
AbstractPosition = -1
|
|
DescriptionPosition = -1
|
|
BinaryAbstractPosition = -1
|
|
BinaryDescriptionPosition = -1
|
|
for StrName in ExistStrNameList:
|
|
if DT.TAB_HEADER_ABSTRACT.upper() in StrName:
|
|
if 'BINARY' in StrName:
|
|
BinaryAbstractPosition = ExistStrNameList.index(StrName)
|
|
else:
|
|
AbstractPosition = ExistStrNameList.index(StrName)
|
|
if DT.TAB_HEADER_DESCRIPTION.upper() in StrName:
|
|
if 'BINARY' in StrName:
|
|
BinaryDescriptionPosition = ExistStrNameList.index(StrName)
|
|
else:
|
|
DescriptionPosition = ExistStrNameList.index(StrName)
|
|
|
|
OrderList = sorted([AbstractPosition, DescriptionPosition])
|
|
BinaryOrderList = sorted([BinaryAbstractPosition, BinaryDescriptionPosition])
|
|
Min = OrderList[0]
|
|
Max = OrderList[1]
|
|
BinaryMin = BinaryOrderList[0]
|
|
BinaryMax = BinaryOrderList[1]
|
|
if BinaryDescriptionPosition > -1:
|
|
if not(BinaryDescriptionPosition == BinaryMax and BinaryAbstractPosition == BinaryMin and \
|
|
BinaryMax > Max):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_ENTRY_ORDER_WRONG, \
|
|
ExtraData=File.Path)
|
|
elif BinaryAbstractPosition > -1:
|
|
if not(BinaryAbstractPosition > Max):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_ENTRY_ORDER_WRONG, \
|
|
ExtraData=File.Path)
|
|
|
|
if DescriptionPosition > -1:
|
|
if not(DescriptionPosition == Max and AbstractPosition == Min and \
|
|
DescriptionPosition > AbstractPosition):
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID, \
|
|
Message=ST.ERR_UNIPARSE_ENTRY_ORDER_WRONG, \
|
|
ExtraData=File.Path)
|
|
|
|
if not self.UniFileHeader:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
Message = ST.ERR_NO_SOURCE_HEADER,
|
|
ExtraData=File.Path)
|
|
|
|
return NewLines
|
|
|
|
#
|
|
# Load a .uni file
|
|
#
|
|
def LoadUniFile(self, File = None):
|
|
if File is None:
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.PARSER_ERROR,
|
|
Message='No unicode file is given',
|
|
ExtraData=File.Path)
|
|
|
|
self.File = File
|
|
|
|
#
|
|
# Process special char in file
|
|
#
|
|
Lines = self.PreProcess(File)
|
|
|
|
#
|
|
# Get Unicode Information
|
|
#
|
|
for IndexI in range(len(Lines)):
|
|
Line = Lines[IndexI]
|
|
if (IndexI + 1) < len(Lines):
|
|
SecondLine = Lines[IndexI + 1]
|
|
if (IndexI + 2) < len(Lines):
|
|
ThirdLine = Lines[IndexI + 2]
|
|
|
|
#
|
|
# Get Language def information
|
|
#
|
|
if Line.find(u'#langdef ') >= 0:
|
|
self.GetLangDef(File, Line + u' ' + SecondLine)
|
|
continue
|
|
|
|
Name = ''
|
|
Language = ''
|
|
Value = ''
|
|
CombineToken = False
|
|
#
|
|
# Get string def information format as below
|
|
#
|
|
# #string MY_STRING_1
|
|
# #language eng
|
|
# "My first English string line 1"
|
|
# "My first English string line 2"
|
|
# #string MY_STRING_1
|
|
# #language spa
|
|
# "Mi segunda secuencia 1"
|
|
# "Mi segunda secuencia 2"
|
|
#
|
|
if Line.find(u'#string ') >= 0 and Line.find(u'#language ') < 0 and \
|
|
SecondLine.find(u'#string ') < 0 and SecondLine.find(u'#language ') >= 0 and \
|
|
ThirdLine.find(u'#string ') < 0 and ThirdLine.find(u'#language ') < 0:
|
|
if Line.find('"') > 0 or SecondLine.find('"') > 0:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FORMAT_INVALID,
|
|
Message=ST.ERR_UNIPARSE_DBLQUOTE_UNMATCHED,
|
|
ExtraData=File.Path)
|
|
|
|
Name = Line[Line.find(u'#string ') + len(u'#string ') : ].strip(' ')
|
|
Language = SecondLine[SecondLine.find(u'#language ') + len(u'#language ') : ].strip(' ')
|
|
for IndexJ in range(IndexI + 2, len(Lines)):
|
|
if Lines[IndexJ].find(u'#string ') < 0 and Lines[IndexJ].find(u'#language ') < 0 and \
|
|
Lines[IndexJ].strip().startswith(u'"') and Lines[IndexJ].strip().endswith(u'"'):
|
|
if Lines[IndexJ][-2] == ' ':
|
|
CombineToken = True
|
|
if CombineToken:
|
|
if Lines[IndexJ].strip()[1:-1].strip():
|
|
Value = Value + Lines[IndexJ].strip()[1:-1].rstrip() + ' '
|
|
else:
|
|
Value = Value + Lines[IndexJ].strip()[1:-1]
|
|
CombineToken = False
|
|
else:
|
|
Value = Value + Lines[IndexJ].strip()[1:-1] + '\r\n'
|
|
else:
|
|
IndexI = IndexJ
|
|
break
|
|
if Value.endswith('\r\n'):
|
|
Value = Value[: Value.rfind('\r\n')]
|
|
Language = GetLanguageCode(Language, self.IsCompatibleMode, self.File)
|
|
self.AddStringToList(Name, Language, Value)
|
|
continue
|
|
|
|
#
|
|
# Load multiple .uni files
|
|
#
|
|
def LoadUniFiles(self, FileList):
|
|
if len(FileList) > 0:
|
|
for File in FileList:
|
|
FilePath = File.Path.strip()
|
|
if FilePath.endswith('.uni') or FilePath.endswith('.UNI') or FilePath.endswith('.Uni'):
|
|
self.LoadUniFile(File)
|
|
|
|
#
|
|
# Add a string to list
|
|
#
|
|
def AddStringToList(self, Name, Language, Value, Token = 0, Referenced = False, UseOtherLangDef = '', Index = -1):
|
|
for LangNameItem in self.LanguageDef:
|
|
if Language == LangNameItem[0]:
|
|
break
|
|
|
|
if Language not in self.OrderedStringList:
|
|
self.OrderedStringList[Language] = []
|
|
self.OrderedStringDict[Language] = {}
|
|
|
|
IsAdded = True
|
|
if Name in self.OrderedStringDict[Language]:
|
|
IsAdded = False
|
|
if Value is not None:
|
|
ItemIndexInList = self.OrderedStringDict[Language][Name]
|
|
Item = self.OrderedStringList[Language][ItemIndexInList]
|
|
Item.UpdateValue(Value)
|
|
Item.UseOtherLangDef = ''
|
|
|
|
if IsAdded:
|
|
Token = len(self.OrderedStringList[Language])
|
|
if Index == -1:
|
|
self.OrderedStringList[Language].append(StringDefClassObject(Name,
|
|
Value,
|
|
Referenced,
|
|
Token,
|
|
UseOtherLangDef))
|
|
self.OrderedStringDict[Language][Name] = Token
|
|
for LangName in self.LanguageDef:
|
|
#
|
|
# New STRING token will be added into all language string lists.
|
|
# so that the unique STRING identifier is reserved for all languages in the package list.
|
|
#
|
|
if LangName[0] != Language:
|
|
if UseOtherLangDef != '':
|
|
OtherLangDef = UseOtherLangDef
|
|
else:
|
|
OtherLangDef = Language
|
|
self.OrderedStringList[LangName[0]].append(StringDefClassObject(Name,
|
|
'',
|
|
Referenced,
|
|
Token,
|
|
OtherLangDef))
|
|
self.OrderedStringDict[LangName[0]][Name] = len(self.OrderedStringList[LangName[0]]) - 1
|
|
else:
|
|
self.OrderedStringList[Language].insert(Index, StringDefClassObject(Name,
|
|
Value,
|
|
Referenced,
|
|
Token,
|
|
UseOtherLangDef))
|
|
self.OrderedStringDict[Language][Name] = Index
|
|
|
|
#
|
|
# Set the string as referenced
|
|
#
|
|
def SetStringReferenced(self, Name):
|
|
#
|
|
# String stoken are added in the same order in all language string lists.
|
|
# So, only update the status of string stoken in first language string list.
|
|
#
|
|
Lang = self.LanguageDef[0][0]
|
|
if Name in self.OrderedStringDict[Lang]:
|
|
ItemIndexInList = self.OrderedStringDict[Lang][Name]
|
|
Item = self.OrderedStringList[Lang][ItemIndexInList]
|
|
Item.Referenced = True
|
|
|
|
#
|
|
# Search the string in language definition by Name
|
|
#
|
|
def FindStringValue(self, Name, Lang):
|
|
if Name in self.OrderedStringDict[Lang]:
|
|
ItemIndexInList = self.OrderedStringDict[Lang][Name]
|
|
return self.OrderedStringList[Lang][ItemIndexInList]
|
|
|
|
return None
|
|
|
|
#
|
|
# Search the string in language definition by Token
|
|
#
|
|
def FindByToken(self, Token, Lang):
|
|
for Item in self.OrderedStringList[Lang]:
|
|
if Item.Token == Token:
|
|
return Item
|
|
|
|
return None
|
|
|
|
#
|
|
# Re-order strings and re-generate tokens
|
|
#
|
|
def ReToken(self):
|
|
if len(self.LanguageDef) == 0:
|
|
return None
|
|
#
|
|
# Retoken all language strings according to the status of string stoken in the first language string.
|
|
#
|
|
FirstLangName = self.LanguageDef[0][0]
|
|
|
|
# Convert the OrderedStringList to be OrderedStringListByToken in order to faciliate future search by token
|
|
for LangNameItem in self.LanguageDef:
|
|
self.OrderedStringListByToken[LangNameItem[0]] = {}
|
|
|
|
#
|
|
# Use small token for all referred string stoken.
|
|
#
|
|
RefToken = 0
|
|
for Index in range (0, len (self.OrderedStringList[FirstLangName])):
|
|
FirstLangItem = self.OrderedStringList[FirstLangName][Index]
|
|
if FirstLangItem.Referenced == True:
|
|
for LangNameItem in self.LanguageDef:
|
|
LangName = LangNameItem[0]
|
|
OtherLangItem = self.OrderedStringList[LangName][Index]
|
|
OtherLangItem.Referenced = True
|
|
OtherLangItem.Token = RefToken
|
|
self.OrderedStringListByToken[LangName][OtherLangItem.Token] = OtherLangItem
|
|
RefToken = RefToken + 1
|
|
|
|
#
|
|
# Use big token for all unreferred string stoken.
|
|
#
|
|
UnRefToken = 0
|
|
for Index in range (0, len (self.OrderedStringList[FirstLangName])):
|
|
FirstLangItem = self.OrderedStringList[FirstLangName][Index]
|
|
if FirstLangItem.Referenced == False:
|
|
for LangNameItem in self.LanguageDef:
|
|
LangName = LangNameItem[0]
|
|
OtherLangItem = self.OrderedStringList[LangName][Index]
|
|
OtherLangItem.Token = RefToken + UnRefToken
|
|
self.OrderedStringListByToken[LangName][OtherLangItem.Token] = OtherLangItem
|
|
UnRefToken = UnRefToken + 1
|
|
|
|
#
|
|
# Show the instance itself
|
|
#
|
|
def ShowMe(self):
|
|
print(self.LanguageDef)
|
|
#print self.OrderedStringList
|
|
for Item in self.OrderedStringList:
|
|
print(Item)
|
|
for Member in self.OrderedStringList[Item]:
|
|
print(str(Member))
|
|
|
|
#
|
|
# Read content from '!include' UNI file
|
|
#
|
|
def ReadIncludeUNIfile(self, FilaPath):
|
|
if self.File:
|
|
pass
|
|
|
|
if not os.path.exists(FilaPath) or not os.path.isfile(FilaPath):
|
|
EdkLogger.Error("Unicode File Parser",
|
|
ToolError.FILE_NOT_FOUND,
|
|
ExtraData=FilaPath)
|
|
try:
|
|
FileIn = codecs.open(FilaPath, mode='rb', encoding='utf_8').readlines()
|
|
except UnicodeError as Xstr:
|
|
FileIn = codecs.open(FilaPath, mode='rb', encoding='utf_16').readlines()
|
|
except UnicodeError:
|
|
FileIn = codecs.open(FilaPath, mode='rb', encoding='utf_16_le').readlines()
|
|
except:
|
|
EdkLogger.Error("Unicode File Parser", ToolError.FILE_OPEN_FAILURE, ExtraData=FilaPath)
|
|
return FileIn
|
|
|