## @file # This file is used to parse meta files # # Copyright (c) 2008 - 2018, Intel Corporation. All rights reserved.
# (C) Copyright 2015-2018 Hewlett Packard Enterprise Development LP
# SPDX-License-Identifier: BSD-2-Clause-Patent # ## # Import Modules # from __future__ import print_function from __future__ import absolute_import import Common.LongFilePathOs as os import re import time import copy from hashlib import md5 import Common.EdkLogger as EdkLogger import Common.GlobalData as GlobalData from CommonDataClass.DataClass import * from Common.DataType import * from Common.StringUtils import * from Common.Misc import GuidStructureStringToGuidString, CheckPcdDatum, PathClass, AnalyzePcdData, AnalyzeDscPcd, AnalyzePcdExpression, ParseFieldValue, StructPattern from Common.Expression import * from CommonDataClass.Exceptions import * from Common.LongFilePathSupport import OpenLongFilePath as open from collections import defaultdict from .MetaFileTable import MetaFileStorage from .MetaFileCommentParser import CheckInfComment from Common.DataType import TAB_COMMENT_EDK_START, TAB_COMMENT_EDK_END ## RegEx for finding file versions hexVersionPattern = re.compile(r'0[xX][\da-f-A-F]{5,8}') decVersionPattern = re.compile(r'\d+\.\d+') CODEPattern = re.compile(r"{CODE\([a-fA-F0-9Xx\{\},\s]*\)}") ## A decorator used to parse macro definition def ParseMacro(Parser): def MacroParser(self): Match = GlobalData.gMacroDefPattern.match(self._CurrentLine) if not Match: # Not 'DEFINE/EDK_GLOBAL' statement, call decorated method Parser(self) return TokenList = GetSplitValueList(self._CurrentLine[Match.end(1):], TAB_EQUAL_SPLIT, 1) # Syntax check if not TokenList[0]: EdkLogger.error('Parser', FORMAT_INVALID, "No macro name given", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if len(TokenList) < 2: TokenList.append('') Type = Match.group(1) Name, Value = TokenList # Global macros can be only defined via environment variable if Name in GlobalData.gGlobalDefines: EdkLogger.error('Parser', FORMAT_INVALID, "%s can only be defined via environment variable" % Name, ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) # Only upper case letters, digit and '_' are allowed if not GlobalData.gMacroNamePattern.match(Name): EdkLogger.error('Parser', FORMAT_INVALID, "The macro name must be in the pattern [A-Z][A-Z0-9_]*", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) Value = ReplaceMacro(Value, self._Macros) if Type in self.DataType: self._ItemType = self.DataType[Type] else: self._ItemType = MODEL_META_DATA_DEFINE # DEFINE defined macros if Type == TAB_DSC_DEFINES_DEFINE: # # First judge whether this DEFINE is in conditional directive statements or not. # if isinstance(self, DscParser) and self._InDirective > -1: pass else: if isinstance(self, DecParser): if MODEL_META_DATA_HEADER in self._SectionType: self._FileLocalMacros[Name] = Value else: self._ConstructSectionMacroDict(Name, Value) elif self._SectionType == MODEL_META_DATA_HEADER: self._FileLocalMacros[Name] = Value else: self._ConstructSectionMacroDict(Name, Value) # EDK_GLOBAL defined macros elif not isinstance(self, DscParser): EdkLogger.error('Parser', FORMAT_INVALID, "EDK_GLOBAL can only be used in .dsc file", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) elif self._SectionType != MODEL_META_DATA_HEADER: EdkLogger.error('Parser', FORMAT_INVALID, "EDK_GLOBAL can only be used under [Defines] section", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) elif (Name in self._FileLocalMacros) and (self._FileLocalMacros[Name] != Value): EdkLogger.error('Parser', FORMAT_INVALID, "EDK_GLOBAL defined a macro with the same name and different value as one defined by 'DEFINE'", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList = [Type, Name, Value] return MacroParser ## Base class of parser # # This class is used for derivation purpose. The specific parser for one kind # type file must derive this class and implement some public interfaces. # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Table Database used to retrieve module/package information # @param Macros Macros used for replacement in file # @param Owner Owner ID (for sub-section parsing) # @param From ID from which the data comes (for !INCLUDE directive) # class MetaFileParser(object): # data type (file content) for specific file type DataType = {} # Parser objects used to implement singleton MetaFiles = {} ## Factory method # # One file, one parser object. This factory method makes sure that there's # only one object constructed for one meta file. # # @param Class class object of real AutoGen class # (InfParser, DecParser or DscParser) # @param FilePath The path of meta file # @param *args The specific class related parameters # @param **kwargs The specific class related dict parameters # def __new__(Class, FilePath, *args, **kwargs): if FilePath in Class.MetaFiles: return Class.MetaFiles[FilePath] else: ParserObject = super(MetaFileParser, Class).__new__(Class) Class.MetaFiles[FilePath] = ParserObject return ParserObject ## Constructor of MetaFileParser # # Initialize object of MetaFileParser # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Arch Default Arch value for filtering sections # @param Table Database used to retrieve module/package information # @param Owner Owner ID (for sub-section parsing) # @param From ID from which the data comes (for !INCLUDE directive) # def __init__(self, FilePath, FileType, Arch, Table, Owner= -1, From= -1): self._Table = Table self._RawTable = Table self._Arch = Arch self._FileType = FileType self.MetaFile = FilePath self._FileDir = self.MetaFile.Dir self._Defines = {} self._Packages = [] self._FileLocalMacros = {} self._SectionsMacroDict = defaultdict(dict) # for recursive parsing self._Owner = [Owner] self._From = From # parsr status for parsing self._ValueList = ['', '', '', '', ''] self._Scope = [] self._LineIndex = 0 self._CurrentLine = '' self._SectionType = MODEL_UNKNOWN self._SectionName = '' self._InSubsection = False self._SubsectionType = MODEL_UNKNOWN self._SubsectionName = '' self._ItemType = MODEL_UNKNOWN self._LastItem = -1 self._Enabled = 0 self._Finished = False self._PostProcessed = False # Different version of meta-file has different way to parse. self._Version = 0 self._GuidDict = {} # for Parser PCD value {GUID(gTokeSpaceGuidName)} self._PcdCodeValue = "" self._PcdDataTypeCODE = False self._CurrentPcdName = "" ## Store the parsed data in table def _Store(self, *Args): return self._Table.Insert(*Args) ## Virtual method for starting parse def Start(self): raise NotImplementedError ## Notify a post-process is needed def DoPostProcess(self): self._PostProcessed = False ## Set parsing complete flag in both class and table def _Done(self): self._Finished = True self._Table.SetEndFlag() def _PostProcess(self): self._PostProcessed = True ## Get the parse complete flag @property def Finished(self): return self._Finished ## Set the complete flag @Finished.setter def Finished(self, Value): self._Finished = Value ## Remove records that do not match given Filter Arch def _FilterRecordList(self, RecordList, FilterArch): NewRecordList = [] for Record in RecordList: Arch = Record[3] if Arch == TAB_ARCH_COMMON or Arch == FilterArch: NewRecordList.append(Record) return NewRecordList ## Use [] style to query data in table, just for readability # # DataInfo = [data_type, scope1(arch), scope2(platform/moduletype)] # def __getitem__(self, DataInfo): if not isinstance(DataInfo, type(())): DataInfo = (DataInfo,) # Parse the file first, if necessary self.StartParse() # No specific ARCH or Platform given, use raw data if self._RawTable and (len(DataInfo) == 1 or DataInfo[1] is None): return self._FilterRecordList(self._RawTable.Query(*DataInfo), self._Arch) # Do post-process if necessary if not self._PostProcessed: self._PostProcess() return self._FilterRecordList(self._Table.Query(*DataInfo), DataInfo[1]) def StartParse(self): if not self._Finished: if self._RawTable.IsIntegrity(): self._Finished = True else: self._Table = self._RawTable self._PostProcessed = False self.Start() ## Data parser for the common format in different type of file # # The common format in the meatfile is like # # xxx1 | xxx2 | xxx3 # @ParseMacro def _CommonParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) self._ValueList[0:len(TokenList)] = TokenList ## Data parser for the format in which there's path # # Only path can have macro used. So we need to replace them before use. # @ParseMacro def _PathParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) self._ValueList[0:len(TokenList)] = TokenList # Don't do macro replacement for dsc file at this point if not isinstance(self, DscParser): Macros = self._Macros self._ValueList = [ReplaceMacro(Value, Macros) for Value in self._ValueList] ## Skip unsupported data def _Skip(self): EdkLogger.warn("Parser", "Unrecognized content", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine); self._ValueList[0:1] = [self._CurrentLine] ## Skip unsupported data for UserExtension Section def _SkipUserExtension(self): self._ValueList[0:1] = [self._CurrentLine] ## Section header parser # # The section header is always in following format: # # [section_name.arch<.platform|module_type>] # def _SectionHeaderParser(self): self._Scope = [] self._SectionName = '' ArchList = set() for Item in GetSplitValueList(self._CurrentLine[1:-1], TAB_COMMA_SPLIT): if Item == '': continue ItemList = GetSplitValueList(Item, TAB_SPLIT, 3) # different section should not mix in one section if self._SectionName != '' and self._SectionName != ItemList[0].upper(): EdkLogger.error('Parser', FORMAT_INVALID, "Different section names in the same section", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) self._SectionName = ItemList[0].upper() if self._SectionName in self.DataType: self._SectionType = self.DataType[self._SectionName] # Check if the section name is valid if self._SectionName not in SECTIONS_HAVE_ITEM_AFTER_ARCH_SET and len(ItemList) > 3: EdkLogger.error("Parser", FORMAT_UNKNOWN_ERROR, "%s is not a valid section name" % Item, self.MetaFile, self._LineIndex + 1, self._CurrentLine) elif self._Version >= 0x00010005: EdkLogger.error("Parser", FORMAT_UNKNOWN_ERROR, "%s is not a valid section name" % Item, self.MetaFile, self._LineIndex + 1, self._CurrentLine) else: self._SectionType = MODEL_UNKNOWN # S1 is always Arch if len(ItemList) > 1: S1 = ItemList[1].upper() else: S1 = TAB_ARCH_COMMON S1 = ReplaceMacro(S1, self._Macros) ArchList.add(S1) # S2 may be Platform or ModuleType if len(ItemList) > 2: if self._SectionName.upper() in SECTIONS_HAVE_ITEM_PCD_SET: S2 = ItemList[2] else: S2 = ItemList[2].upper() else: S2 = TAB_COMMON if len(ItemList) > 3: S3 = ItemList[3] else: S3 = TAB_COMMON self._Scope.append([S1, S2, S3]) # 'COMMON' must not be used with specific ARCHs at the same section if TAB_ARCH_COMMON in ArchList and len(ArchList) > 1: EdkLogger.error('Parser', FORMAT_INVALID, "'common' ARCH must not be used with specific ARCHs", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) # If the section information is needed later, it should be stored in database self._ValueList[0] = self._SectionName ## [packages] section parser @ParseMacro def _PackageParser(self): self._CurrentLine = CleanString(self._CurrentLine) self._Packages.append(self._CurrentLine) self._ValueList[0] = self._CurrentLine ## [defines] section parser @ParseMacro def _DefineParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) self._ValueList[1:len(TokenList)] = TokenList if not self._ValueList[1]: EdkLogger.error('Parser', FORMAT_INVALID, "No name specified", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if not self._ValueList[2]: EdkLogger.error('Parser', FORMAT_INVALID, "No value specified", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList = [ReplaceMacro(Value, self._Macros) for Value in self._ValueList] Name, Value = self._ValueList[1], self._ValueList[2] MacroUsed = GlobalData.gMacroRefPattern.findall(Value) if len(MacroUsed) != 0: for Macro in MacroUsed: if Macro in GlobalData.gGlobalDefines: EdkLogger.error("Parser", FORMAT_INVALID, "Global macro %s is not permitted." % (Macro), ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) else: EdkLogger.error("Parser", FORMAT_INVALID, "%s not defined" % (Macro), ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) # Sometimes, we need to make differences between EDK and EDK2 modules if Name == 'INF_VERSION': if hexVersionPattern.match(Value): self._Version = int(Value, 0) elif decVersionPattern.match(Value): ValueList = Value.split('.') Major = int(ValueList[0], 0) Minor = int(ValueList[1], 0) if Major > 0xffff or Minor > 0xffff: EdkLogger.error('Parser', FORMAT_INVALID, "Invalid version number", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._Version = int('0x{0:04x}{1:04x}'.format(Major, Minor), 0) else: EdkLogger.error('Parser', FORMAT_INVALID, "Invalid version number", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if isinstance(self, InfParser) and self._Version < 0x00010005: # EDK module allows using defines as macros self._FileLocalMacros[Name] = Value self._Defines[Name] = Value ## [BuildOptions] section parser @ParseMacro def _BuildOptionParser(self): self._CurrentLine = CleanString(self._CurrentLine, BuildOption=True) TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) TokenList2 = GetSplitValueList(TokenList[0], ':', 1) if len(TokenList2) == 2: self._ValueList[0] = TokenList2[0] # toolchain family self._ValueList[1] = TokenList2[1] # keys else: self._ValueList[1] = TokenList[0] if len(TokenList) == 2 and not isinstance(self, DscParser): # value self._ValueList[2] = ReplaceMacro(TokenList[1], self._Macros) if self._ValueList[1].count('_') != 4: EdkLogger.error( 'Parser', FORMAT_INVALID, "'%s' must be in format of ____FLAGS" % self._ValueList[1], ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1 ) def GetValidExpression(self, TokenSpaceGuid, PcdCName): return self._Table.GetValidExpression(TokenSpaceGuid, PcdCName) @property def _Macros(self): Macros = {} Macros.update(self._FileLocalMacros) Macros.update(self._GetApplicableSectionMacro()) return Macros ## Construct section Macro dict def _ConstructSectionMacroDict(self, Name, Value): ScopeKey = [(Scope[0], Scope[1], Scope[2]) for Scope in self._Scope] ScopeKey = tuple(ScopeKey) # # DecParser SectionType is a list, will contain more than one item only in Pcd Section # As Pcd section macro usage is not allowed, so here it is safe # if isinstance(self, DecParser): SectionDictKey = self._SectionType[0], ScopeKey else: SectionDictKey = self._SectionType, ScopeKey self._SectionsMacroDict[SectionDictKey][Name] = Value ## Get section Macros that are applicable to current line, which may come from other sections ## that share the same name while scope is wider def _GetApplicableSectionMacro(self): Macros = {} ComComMacroDict = {} ComSpeMacroDict = {} SpeSpeMacroDict = {} ActiveSectionType = self._SectionType if isinstance(self, DecParser): ActiveSectionType = self._SectionType[0] for (SectionType, Scope) in self._SectionsMacroDict: if SectionType != ActiveSectionType: continue for ActiveScope in self._Scope: Scope0, Scope1, Scope2= ActiveScope[0], ActiveScope[1], ActiveScope[2] if(Scope0, Scope1, Scope2) not in Scope: break else: SpeSpeMacroDict.update(self._SectionsMacroDict[(SectionType, Scope)]) for ActiveScope in self._Scope: Scope0, Scope1, Scope2 = ActiveScope[0], ActiveScope[1], ActiveScope[2] if(Scope0, Scope1, Scope2) not in Scope and (Scope0, TAB_COMMON, TAB_COMMON) not in Scope and (TAB_COMMON, Scope1, TAB_COMMON) not in Scope: break else: ComSpeMacroDict.update(self._SectionsMacroDict[(SectionType, Scope)]) if (TAB_COMMON, TAB_COMMON, TAB_COMMON) in Scope: ComComMacroDict.update(self._SectionsMacroDict[(SectionType, Scope)]) Macros.update(ComComMacroDict) Macros.update(ComSpeMacroDict) Macros.update(SpeSpeMacroDict) return Macros def ProcessMultipleLineCODEValue(self,Content): CODEBegin = False CODELine = "" continuelinecount = 0 newContent = [] for Index in range(0, len(Content)): Line = Content[Index] if CODEBegin: CODELine = CODELine + Line continuelinecount +=1 if ")}" in Line: newContent.append(CODELine) for _ in range(continuelinecount): newContent.append("") CODEBegin = False CODELine = "" continuelinecount = 0 else: if not Line: newContent.append(Line) continue if "{CODE(" not in Line: newContent.append(Line) continue elif CODEPattern.findall(Line): newContent.append(Line) continue else: CODEBegin = True CODELine = Line return newContent _SectionParser = {} ## INF file parser class # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Table Database used to retrieve module/package information # @param Macros Macros used for replacement in file # class InfParser(MetaFileParser): # INF file supported data types (one type per section) DataType = { TAB_UNKNOWN.upper() : MODEL_UNKNOWN, TAB_INF_DEFINES.upper() : MODEL_META_DATA_HEADER, TAB_DSC_DEFINES_DEFINE : MODEL_META_DATA_DEFINE, TAB_BUILD_OPTIONS.upper() : MODEL_META_DATA_BUILD_OPTION, TAB_INCLUDES.upper() : MODEL_EFI_INCLUDE, TAB_LIBRARIES.upper() : MODEL_EFI_LIBRARY_INSTANCE, TAB_LIBRARY_CLASSES.upper() : MODEL_EFI_LIBRARY_CLASS, TAB_PACKAGES.upper() : MODEL_META_DATA_PACKAGE, TAB_NMAKE.upper() : MODEL_META_DATA_NMAKE, TAB_INF_FIXED_PCD.upper() : MODEL_PCD_FIXED_AT_BUILD, TAB_INF_PATCH_PCD.upper() : MODEL_PCD_PATCHABLE_IN_MODULE, TAB_INF_FEATURE_PCD.upper() : MODEL_PCD_FEATURE_FLAG, TAB_INF_PCD_EX.upper() : MODEL_PCD_DYNAMIC_EX, TAB_INF_PCD.upper() : MODEL_PCD_DYNAMIC, TAB_SOURCES.upper() : MODEL_EFI_SOURCE_FILE, TAB_GUIDS.upper() : MODEL_EFI_GUID, TAB_PROTOCOLS.upper() : MODEL_EFI_PROTOCOL, TAB_PPIS.upper() : MODEL_EFI_PPI, TAB_DEPEX.upper() : MODEL_EFI_DEPEX, TAB_BINARIES.upper() : MODEL_EFI_BINARY_FILE, TAB_USER_EXTENSIONS.upper() : MODEL_META_DATA_USER_EXTENSION } ## Constructor of InfParser # # Initialize object of InfParser # # @param FilePath The path of module description file # @param FileType The raw data of DSC file # @param Arch Default Arch value for filtering sections # @param Table Database used to retrieve module/package information # def __init__(self, FilePath, FileType, Arch, Table): # prevent re-initialization if hasattr(self, "_Table"): return MetaFileParser.__init__(self, FilePath, FileType, Arch, Table) self.PcdsDict = {} ## Parser starter def Start(self): NmakeLine = '' Content = '' try: with open(str(self.MetaFile), 'r') as File: Content = File.readlines() except: EdkLogger.error("Parser", FILE_READ_FAILURE, ExtraData=self.MetaFile) # parse the file line by line IsFindBlockComment = False GetHeaderComment = False TailComments = [] SectionComments = [] Comments = [] for Index in range(0, len(Content)): # skip empty, commented, block commented lines Line, Comment = CleanString2(Content[Index], AllowCppStyleComment=True) NextLine = '' if Index + 1 < len(Content): NextLine, NextComment = CleanString2(Content[Index + 1]) if Line == '': if Comment: Comments.append((Comment, Index + 1)) elif GetHeaderComment: SectionComments.extend(Comments) Comments = [] continue if Line.find(TAB_COMMENT_EDK_START) > -1: IsFindBlockComment = True continue if Line.find(TAB_COMMENT_EDK_END) > -1: IsFindBlockComment = False continue if IsFindBlockComment: continue self._LineIndex = Index self._CurrentLine = Line # section header if Line[0] == TAB_SECTION_START and Line[-1] == TAB_SECTION_END: if not GetHeaderComment: for Cmt, LNo in Comments: self._Store(MODEL_META_DATA_HEADER_COMMENT, Cmt, '', '', TAB_COMMON, TAB_COMMON, self._Owner[-1], LNo, -1, LNo, -1, 0) GetHeaderComment = True else: TailComments.extend(SectionComments + Comments) Comments = [] self._SectionHeaderParser() # Check invalid sections if self._Version < 0x00010005: if self._SectionType in [MODEL_META_DATA_BUILD_OPTION, MODEL_EFI_LIBRARY_CLASS, MODEL_META_DATA_PACKAGE, MODEL_PCD_FIXED_AT_BUILD, MODEL_PCD_PATCHABLE_IN_MODULE, MODEL_PCD_FEATURE_FLAG, MODEL_PCD_DYNAMIC_EX, MODEL_PCD_DYNAMIC, MODEL_EFI_GUID, MODEL_EFI_PROTOCOL, MODEL_EFI_PPI, MODEL_META_DATA_USER_EXTENSION]: EdkLogger.error('Parser', FORMAT_INVALID, "Section [%s] is not allowed in inf file without version" % (self._SectionName), ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) elif self._SectionType in [MODEL_EFI_INCLUDE, MODEL_EFI_LIBRARY_INSTANCE, MODEL_META_DATA_NMAKE]: EdkLogger.error('Parser', FORMAT_INVALID, "Section [%s] is not allowed in inf file with version 0x%08x" % (self._SectionName, self._Version), ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) continue # merge two lines specified by '\' in section NMAKE elif self._SectionType == MODEL_META_DATA_NMAKE: if Line[-1] == '\\': if NextLine == '': self._CurrentLine = NmakeLine + Line[0:-1] NmakeLine = '' else: if NextLine[0] == TAB_SECTION_START and NextLine[-1] == TAB_SECTION_END: self._CurrentLine = NmakeLine + Line[0:-1] NmakeLine = '' else: NmakeLine = NmakeLine + ' ' + Line[0:-1] continue else: self._CurrentLine = NmakeLine + Line NmakeLine = '' # section content self._ValueList = ['', '', ''] # parse current line, result will be put in self._ValueList self._SectionParser[self._SectionType](self) if self._ValueList is None or self._ItemType == MODEL_META_DATA_DEFINE: self._ItemType = -1 Comments = [] continue if Comment: Comments.append((Comment, Index + 1)) if GlobalData.gOptions and GlobalData.gOptions.CheckUsage: CheckInfComment(self._SectionType, Comments, str(self.MetaFile), Index + 1, self._ValueList) # # Model, Value1, Value2, Value3, Arch, Platform, BelongsToItem=-1, # LineBegin=-1, ColumnBegin=-1, LineEnd=-1, ColumnEnd=-1, Enabled=-1 # for Arch, Platform, _ in self._Scope: LastItem = self._Store(self._SectionType, self._ValueList[0], self._ValueList[1], self._ValueList[2], Arch, Platform, self._Owner[-1], self._LineIndex + 1, - 1, self._LineIndex + 1, - 1, 0 ) for Comment, LineNo in Comments: self._Store(MODEL_META_DATA_COMMENT, Comment, '', '', Arch, Platform, LastItem, LineNo, -1, LineNo, -1, 0) Comments = [] SectionComments = [] TailComments.extend(SectionComments + Comments) if IsFindBlockComment: EdkLogger.error("Parser", FORMAT_INVALID, "Open block comments (starting with /*) are expected to end with */", File=self.MetaFile) # If there are tail comments in INF file, save to database whatever the comments are for Comment in TailComments: self._Store(MODEL_META_DATA_TAIL_COMMENT, Comment[0], '', '', TAB_COMMON, TAB_COMMON, self._Owner[-1], -1, -1, -1, -1, 0) self._Done() ## Data parser for the format in which there's path # # Only path can have macro used. So we need to replace them before use. # def _IncludeParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) self._ValueList[0:len(TokenList)] = TokenList Macros = self._Macros if Macros: for Index in range(0, len(self._ValueList)): Value = self._ValueList[Index] if not Value: continue self._ValueList[Index] = ReplaceMacro(Value, Macros) ## Parse [Sources] section # # Only path can have macro used. So we need to replace them before use. # @ParseMacro def _SourceFileParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) self._ValueList[0:len(TokenList)] = TokenList Macros = self._Macros # For Acpi tables, remove macro like ' TABLE_NAME=Sata1' if 'COMPONENT_TYPE' in Macros: if self._Defines['COMPONENT_TYPE'].upper() == 'ACPITABLE': self._ValueList[0] = GetSplitValueList(self._ValueList[0], ' ', 1)[0] if self._Defines['BASE_NAME'] == 'Microcode': pass self._ValueList = [ReplaceMacro(Value, Macros) for Value in self._ValueList] ## Parse [Binaries] section # # Only path can have macro used. So we need to replace them before use. # @ParseMacro def _BinaryFileParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT, 2) if len(TokenList) < 2: EdkLogger.error('Parser', FORMAT_INVALID, "No file type or path specified", ExtraData=self._CurrentLine + " ( | [| ])", File=self.MetaFile, Line=self._LineIndex + 1) if not TokenList[0]: EdkLogger.error('Parser', FORMAT_INVALID, "No file type specified", ExtraData=self._CurrentLine + " ( | [| ])", File=self.MetaFile, Line=self._LineIndex + 1) if not TokenList[1]: EdkLogger.error('Parser', FORMAT_INVALID, "No file path specified", ExtraData=self._CurrentLine + " ( | [| ])", File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0:len(TokenList)] = TokenList self._ValueList[1] = ReplaceMacro(self._ValueList[1], self._Macros) ## [nmake] section parser (Edk.x style only) def _NmakeParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) self._ValueList[0:len(TokenList)] = TokenList # remove macros self._ValueList[1] = ReplaceMacro(self._ValueList[1], self._Macros) # remove self-reference in macro setting #self._ValueList[1] = ReplaceMacro(self._ValueList[1], {self._ValueList[0]:''}) ## [FixedPcd], [FeaturePcd], [PatchPcd], [Pcd] and [PcdEx] sections parser @ParseMacro def _PcdParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT, 1) ValueList = GetSplitValueList(TokenList[0], TAB_SPLIT) if len(ValueList) != 2: EdkLogger.error('Parser', FORMAT_INVALID, "Illegal token space GUID and PCD name format", ExtraData=self._CurrentLine + " (.)", File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0:1] = ValueList if len(TokenList) > 1: self._ValueList[2] = TokenList[1] if self._ValueList[0] == '' or self._ValueList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No token space GUID or PCD name specified", ExtraData=self._CurrentLine + " (.)", File=self.MetaFile, Line=self._LineIndex + 1) # if value are 'True', 'true', 'TRUE' or 'False', 'false', 'FALSE', replace with integer 1 or 0. if self._ValueList[2] != '': InfPcdValueList = GetSplitValueList(TokenList[1], TAB_VALUE_SPLIT, 1) if InfPcdValueList[0] in ['True', 'true', 'TRUE']: self._ValueList[2] = TokenList[1].replace(InfPcdValueList[0], '1', 1) elif InfPcdValueList[0] in ['False', 'false', 'FALSE']: self._ValueList[2] = TokenList[1].replace(InfPcdValueList[0], '0', 1) elif isinstance(InfPcdValueList[0], str) and InfPcdValueList[0].find('$(') >= 0: Value = ReplaceExprMacro(InfPcdValueList[0],self._Macros) if Value != '0': self._ValueList[2] = Value if (self._ValueList[0], self._ValueList[1]) not in self.PcdsDict: self.PcdsDict[self._ValueList[0], self._ValueList[1]] = self._SectionType elif self.PcdsDict[self._ValueList[0], self._ValueList[1]] != self._SectionType: EdkLogger.error('Parser', FORMAT_INVALID, "It is not permissible to list a specified PCD in different PCD type sections.", ExtraData=self._CurrentLine + " (.)", File=self.MetaFile, Line=self._LineIndex + 1) ## [depex] section parser @ParseMacro def _DepexParser(self): self._ValueList[0:1] = [self._CurrentLine] _SectionParser = { MODEL_UNKNOWN : MetaFileParser._Skip, MODEL_META_DATA_HEADER : MetaFileParser._DefineParser, MODEL_META_DATA_BUILD_OPTION : MetaFileParser._BuildOptionParser, MODEL_EFI_INCLUDE : _IncludeParser, # for Edk.x modules MODEL_EFI_LIBRARY_INSTANCE : MetaFileParser._CommonParser, # for Edk.x modules MODEL_EFI_LIBRARY_CLASS : MetaFileParser._PathParser, MODEL_META_DATA_PACKAGE : MetaFileParser._PathParser, MODEL_META_DATA_NMAKE : _NmakeParser, # for Edk.x modules MODEL_PCD_FIXED_AT_BUILD : _PcdParser, MODEL_PCD_PATCHABLE_IN_MODULE : _PcdParser, MODEL_PCD_FEATURE_FLAG : _PcdParser, MODEL_PCD_DYNAMIC_EX : _PcdParser, MODEL_PCD_DYNAMIC : _PcdParser, MODEL_EFI_SOURCE_FILE : _SourceFileParser, MODEL_EFI_GUID : MetaFileParser._CommonParser, MODEL_EFI_PROTOCOL : MetaFileParser._CommonParser, MODEL_EFI_PPI : MetaFileParser._CommonParser, MODEL_EFI_DEPEX : _DepexParser, MODEL_EFI_BINARY_FILE : _BinaryFileParser, MODEL_META_DATA_USER_EXTENSION : MetaFileParser._SkipUserExtension, } ## DSC file parser class # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Table Database used to retrieve module/package information # @param Macros Macros used for replacement in file # @param Owner Owner ID (for sub-section parsing) # @param From ID from which the data comes (for !INCLUDE directive) # class DscParser(MetaFileParser): # DSC file supported data types (one type per section) DataType = { TAB_SKUIDS.upper() : MODEL_EFI_SKU_ID, TAB_DEFAULT_STORES.upper() : MODEL_EFI_DEFAULT_STORES, TAB_LIBRARIES.upper() : MODEL_EFI_LIBRARY_INSTANCE, TAB_LIBRARY_CLASSES.upper() : MODEL_EFI_LIBRARY_CLASS, TAB_BUILD_OPTIONS.upper() : MODEL_META_DATA_BUILD_OPTION, TAB_PACKAGES.upper() : MODEL_META_DATA_PACKAGE, TAB_PCDS_FIXED_AT_BUILD_NULL.upper() : MODEL_PCD_FIXED_AT_BUILD, TAB_PCDS_PATCHABLE_IN_MODULE_NULL.upper() : MODEL_PCD_PATCHABLE_IN_MODULE, TAB_PCDS_FEATURE_FLAG_NULL.upper() : MODEL_PCD_FEATURE_FLAG, TAB_PCDS_DYNAMIC_DEFAULT_NULL.upper() : MODEL_PCD_DYNAMIC_DEFAULT, TAB_PCDS_DYNAMIC_HII_NULL.upper() : MODEL_PCD_DYNAMIC_HII, TAB_PCDS_DYNAMIC_VPD_NULL.upper() : MODEL_PCD_DYNAMIC_VPD, TAB_PCDS_DYNAMIC_EX_DEFAULT_NULL.upper() : MODEL_PCD_DYNAMIC_EX_DEFAULT, TAB_PCDS_DYNAMIC_EX_HII_NULL.upper() : MODEL_PCD_DYNAMIC_EX_HII, TAB_PCDS_DYNAMIC_EX_VPD_NULL.upper() : MODEL_PCD_DYNAMIC_EX_VPD, TAB_COMPONENTS.upper() : MODEL_META_DATA_COMPONENT, TAB_DSC_DEFINES.upper() : MODEL_META_DATA_HEADER, TAB_DSC_DEFINES_DEFINE : MODEL_META_DATA_DEFINE, TAB_DSC_DEFINES_EDKGLOBAL : MODEL_META_DATA_GLOBAL_DEFINE, TAB_INCLUDE.upper() : MODEL_META_DATA_INCLUDE, TAB_IF.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_IF, TAB_IF_DEF.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_IFDEF, TAB_IF_N_DEF.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF, TAB_ELSE_IF.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSEIF, TAB_ELSE.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSE, TAB_END_IF.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_ENDIF, TAB_USER_EXTENSIONS.upper() : MODEL_META_DATA_USER_EXTENSION, TAB_ERROR.upper() : MODEL_META_DATA_CONDITIONAL_STATEMENT_ERROR, } # Valid names in define section DefineKeywords = [ "DSC_SPECIFICATION", "PLATFORM_NAME", "PLATFORM_GUID", "PLATFORM_VERSION", "SKUID_IDENTIFIER", "PCD_INFO_GENERATION", "PCD_VAR_CHECK_GENERATION", "SUPPORTED_ARCHITECTURES", "BUILD_TARGETS", "OUTPUT_DIRECTORY", "FLASH_DEFINITION", "BUILD_NUMBER", "RFC_LANGUAGES", "ISO_LANGUAGES", "TIME_STAMP_FILE", "VPD_TOOL_GUID", "FIX_LOAD_TOP_MEMORY_ADDRESS", "PREBUILD", "POSTBUILD" ] SubSectionDefineKeywords = [ "FILE_GUID" ] SymbolPattern = ValueExpression.SymbolPattern IncludedFiles = set() ## Constructor of DscParser # # Initialize object of DscParser # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Arch Default Arch value for filtering sections # @param Table Database used to retrieve module/package information # @param Owner Owner ID (for sub-section parsing) # @param From ID from which the data comes (for !INCLUDE directive) # def __init__(self, FilePath, FileType, Arch, Table, Owner= -1, From= -1): # prevent re-initialization if hasattr(self, "_Table") and self._Table is Table: return MetaFileParser.__init__(self, FilePath, FileType, Arch, Table, Owner, From) self._Version = 0x00010005 # Only EDK2 dsc file is supported # to store conditional directive evaluation result self._DirectiveStack = [] self._DirectiveEvalStack = [] self._Enabled = 1 # # Specify whether current line is in uncertain condition # self._InDirective = -1 # Final valid replacable symbols self._Symbols = {} # # Map the ID between the original table and new table to track # the owner item # self._IdMapping = {-1:-1} self._Content = None ## Parser starter def Start(self): Content = '' try: with open(str(self.MetaFile), 'r') as File: Content = File.readlines() except: EdkLogger.error("Parser", FILE_READ_FAILURE, ExtraData=self.MetaFile) OwnerId = {} Content = self.ProcessMultipleLineCODEValue(Content) for Index in range(0, len(Content)): Line = CleanString(Content[Index]) # skip empty line if Line == '': continue self._CurrentLine = Line self._LineIndex = Index if self._InSubsection and self._Owner[-1] == -1: self._Owner.append(self._LastItem) # section header if Line[0] == TAB_SECTION_START and Line[-1] == TAB_SECTION_END: self._SectionType = MODEL_META_DATA_SECTION_HEADER # subsection ending elif Line[0] == '}' and self._InSubsection: self._InSubsection = False self._SubsectionType = MODEL_UNKNOWN self._SubsectionName = '' self._Owner[-1] = -1 OwnerId.clear() continue # subsection header elif Line[0] == TAB_OPTION_START and Line[-1] == TAB_OPTION_END: self._SubsectionType = MODEL_META_DATA_SUBSECTION_HEADER # directive line elif Line[0] == '!': TokenList = GetSplitValueList(Line, ' ', 1) if TokenList[0] == TAB_INCLUDE: for Arch, ModuleType, DefaultStore in self._Scope: if self._SubsectionType != MODEL_UNKNOWN and Arch in OwnerId: self._Owner[-1] = OwnerId[Arch] self._DirectiveParser() else: self._DirectiveParser() continue if Line[0] == TAB_OPTION_START and not self._InSubsection: EdkLogger.error("Parser", FILE_READ_FAILURE, "Missing the '{' before %s in Line %s" % (Line, Index+1), ExtraData=self.MetaFile) if self._InSubsection: SectionType = self._SubsectionType else: SectionType = self._SectionType self._ItemType = SectionType self._ValueList = ['', '', ''] # "SET pcd = pcd_expression" syntax is not supported in Dsc file. if self._CurrentLine.upper().strip().startswith("SET "): EdkLogger.error('Parser', FORMAT_INVALID, '''"SET pcd = pcd_expression" syntax is not support in Dsc file''', ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._SectionParser[SectionType](self) if self._ValueList is None: continue # # Model, Value1, Value2, Value3, Arch, ModuleType, BelongsToItem=-1, BelongsToFile=-1, # LineBegin=-1, ColumnBegin=-1, LineEnd=-1, ColumnEnd=-1, Enabled=-1 # for Arch, ModuleType, DefaultStore in self._Scope: Owner = self._Owner[-1] if self._SubsectionType != MODEL_UNKNOWN and Arch in OwnerId: Owner = OwnerId[Arch] self._LastItem = self._Store( self._ItemType, self._ValueList[0], self._ValueList[1], self._ValueList[2], Arch, ModuleType, DefaultStore, Owner, self._From, self._LineIndex + 1, - 1, self._LineIndex + 1, - 1, self._Enabled ) if self._SubsectionType == MODEL_UNKNOWN and self._InSubsection: OwnerId[Arch] = self._LastItem if self._DirectiveStack: Type, Line, Text = self._DirectiveStack[-1] EdkLogger.error('Parser', FORMAT_INVALID, "No matching '!endif' found", ExtraData=Text, File=self.MetaFile, Line=Line) self._Done() ## parser def _SubsectionHeaderParser(self): self._SubsectionName = self._CurrentLine[1:-1].upper() if self._SubsectionName in self.DataType: self._SubsectionType = self.DataType[self._SubsectionName] else: self._SubsectionType = MODEL_UNKNOWN EdkLogger.warn("Parser", "Unrecognized sub-section", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) self._ValueList[0] = self._SubsectionName ## Directive statement parser def _DirectiveParser(self): self._ValueList = ['', '', ''] TokenList = GetSplitValueList(self._CurrentLine, ' ', 1) self._ValueList[0:len(TokenList)] = TokenList # Syntax check DirectiveName = self._ValueList[0].upper() if DirectiveName not in self.DataType: EdkLogger.error("Parser", FORMAT_INVALID, "Unknown directive [%s]" % DirectiveName, File=self.MetaFile, Line=self._LineIndex + 1) if DirectiveName in ['!IF', '!IFDEF', '!IFNDEF']: self._InDirective += 1 if DirectiveName in ['!ENDIF']: self._InDirective -= 1 if DirectiveName in ['!IF', '!IFDEF', '!INCLUDE', '!IFNDEF', '!ELSEIF'] and self._ValueList[1] == '': EdkLogger.error("Parser", FORMAT_INVALID, "Missing expression", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) ItemType = self.DataType[DirectiveName] Scope = [[TAB_COMMON, TAB_COMMON, TAB_COMMON]] if ItemType == MODEL_META_DATA_INCLUDE: Scope = self._Scope elif ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ERROR: Scope = self._Scope if ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ENDIF: # Remove all directives between !if and !endif, including themselves while self._DirectiveStack: # Remove any !else or !elseif DirectiveInfo = self._DirectiveStack.pop() if DirectiveInfo[0] in [MODEL_META_DATA_CONDITIONAL_STATEMENT_IF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFDEF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF]: break else: EdkLogger.error("Parser", FORMAT_INVALID, "Redundant '!endif'", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) elif ItemType not in {MODEL_META_DATA_INCLUDE, MODEL_META_DATA_CONDITIONAL_STATEMENT_ERROR}: # Break if there's a !else is followed by a !elseif if ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSEIF and \ self._DirectiveStack and \ self._DirectiveStack[-1][0] == MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSE: EdkLogger.error("Parser", FORMAT_INVALID, "'!elseif' after '!else'", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) self._DirectiveStack.append((ItemType, self._LineIndex + 1, self._CurrentLine)) # # Model, Value1, Value2, Value3, Arch, ModuleType, BelongsToItem=-1, BelongsToFile=-1, # LineBegin=-1, ColumnBegin=-1, LineEnd=-1, ColumnEnd=-1, Enabled=-1 # for Arch, ModuleType, DefaultStore in Scope: self._LastItem = self._Store( ItemType, self._ValueList[0], self._ValueList[1], self._ValueList[2], Arch, ModuleType, DefaultStore, self._Owner[-1], self._From, self._LineIndex + 1, - 1, self._LineIndex + 1, - 1, 0 ) ## [defines] section parser @ParseMacro def _DefineParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) self._ValueList[1:len(TokenList)] = TokenList # Syntax check if not self._ValueList[1]: EdkLogger.error('Parser', FORMAT_INVALID, "No name specified", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if not self._ValueList[2]: EdkLogger.error('Parser', FORMAT_INVALID, "No value specified", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if (not self._ValueList[1] in self.DefineKeywords and (self._InSubsection and self._ValueList[1] not in self.SubSectionDefineKeywords)): EdkLogger.error('Parser', FORMAT_INVALID, "Unknown keyword found: %s. " "If this is a macro you must " "add it as a DEFINE in the DSC" % self._ValueList[1], ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if not self._InSubsection: self._Defines[self._ValueList[1]] = self._ValueList[2] self._ItemType = self.DataType[TAB_DSC_DEFINES.upper()] @ParseMacro def _SkuIdParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) if len(TokenList) not in (2, 3): EdkLogger.error('Parser', FORMAT_INVALID, "Correct format is '|[|]'", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0:len(TokenList)] = TokenList @ParseMacro def _DefaultStoresParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) if len(TokenList) != 2: EdkLogger.error('Parser', FORMAT_INVALID, "Correct format is '|'", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0:len(TokenList)] = TokenList ## Parse Edk style of library modules @ParseMacro def _LibraryInstanceParser(self): self._ValueList[0] = self._CurrentLine def _DecodeCODEData(self): pass ## PCD sections parser # # [PcdsFixedAtBuild] # [PcdsPatchableInModule] # [PcdsFeatureFlag] # [PcdsDynamicEx # [PcdsDynamicExDefault] # [PcdsDynamicExVpd] # [PcdsDynamicExHii] # [PcdsDynamic] # [PcdsDynamicDefault] # [PcdsDynamicVpd] # [PcdsDynamicHii] # @ParseMacro def _PcdParser(self): if self._PcdDataTypeCODE: self._PcdCodeValue = self._PcdCodeValue + "\n " + self._CurrentLine if self._CurrentLine.endswith(")}"): self._CurrentLine = "|".join((self._CurrentPcdName, self._PcdCodeValue)) self._PcdDataTypeCODE = False self._PcdCodeValue = "" else: self._ValueList = None return TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT, 1) self._CurrentPcdName = TokenList[0] if len(TokenList) == 2 and TokenList[1].strip().startswith("{CODE"): self._PcdDataTypeCODE = True self._PcdCodeValue = TokenList[1].strip() if self._PcdDataTypeCODE: if self._CurrentLine.endswith(")}"): self._PcdDataTypeCODE = False self._PcdCodeValue = "" else: self._ValueList = None return self._ValueList[0:1] = GetSplitValueList(TokenList[0], TAB_SPLIT) PcdNameTockens = GetSplitValueList(TokenList[0], TAB_SPLIT) if len(PcdNameTockens) == 2: self._ValueList[0], self._ValueList[1] = PcdNameTockens[0], PcdNameTockens[1] elif len(PcdNameTockens) == 3: self._ValueList[0], self._ValueList[1] = ".".join((PcdNameTockens[0], PcdNameTockens[1])), PcdNameTockens[2] elif len(PcdNameTockens) > 3: self._ValueList[0], self._ValueList[1] = ".".join((PcdNameTockens[0], PcdNameTockens[1])), ".".join(PcdNameTockens[2:]) if len(TokenList) == 2: self._ValueList[2] = TokenList[1] if self._ValueList[0] == '' or self._ValueList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No token space GUID or PCD name specified", ExtraData=self._CurrentLine + " (.|)", File=self.MetaFile, Line=self._LineIndex + 1) if self._ValueList[2] == '': # # The PCD values are optional for FIXEDATBUILD, PATCHABLEINMODULE, Dynamic/DynamicEx default # if self._SectionType in (MODEL_PCD_FIXED_AT_BUILD, MODEL_PCD_PATCHABLE_IN_MODULE, MODEL_PCD_DYNAMIC_DEFAULT, MODEL_PCD_DYNAMIC_EX_DEFAULT): return EdkLogger.error('Parser', FORMAT_INVALID, "No PCD value given", ExtraData=self._CurrentLine + " (.|)", File=self.MetaFile, Line=self._LineIndex + 1) # Validate the datum type of Dynamic Defaul PCD and DynamicEx Default PCD ValueList = GetSplitValueList(self._ValueList[2]) if len(ValueList) > 1 and ValueList[1] in [TAB_UINT8, TAB_UINT16, TAB_UINT32, TAB_UINT64] \ and self._ItemType in [MODEL_PCD_DYNAMIC_DEFAULT, MODEL_PCD_DYNAMIC_EX_DEFAULT]: EdkLogger.error('Parser', FORMAT_INVALID, "The datum type '%s' of PCD is wrong" % ValueList[1], ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) # Validate the VariableName of DynamicHii and DynamicExHii for PCD Entry must not be an empty string if self._ItemType in [MODEL_PCD_DYNAMIC_HII, MODEL_PCD_DYNAMIC_EX_HII]: DscPcdValueList = GetSplitValueList(TokenList[1], TAB_VALUE_SPLIT, 1) if len(DscPcdValueList[0].replace('L', '').replace('"', '').strip()) == 0: EdkLogger.error('Parser', FORMAT_INVALID, "The VariableName field in the HII format PCD entry must not be an empty string", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) # if value are 'True', 'true', 'TRUE' or 'False', 'false', 'FALSE', replace with integer 1 or 0. DscPcdValueList = GetSplitValueList(TokenList[1], TAB_VALUE_SPLIT, 1) if DscPcdValueList[0] in ['True', 'true', 'TRUE']: self._ValueList[2] = TokenList[1].replace(DscPcdValueList[0], '1', 1); elif DscPcdValueList[0] in ['False', 'false', 'FALSE']: self._ValueList[2] = TokenList[1].replace(DscPcdValueList[0], '0', 1); ## [components] section parser @ParseMacro def _ComponentParser(self): if self._CurrentLine[-1] == '{': self._ValueList[0] = self._CurrentLine[0:-1].strip() self._InSubsection = True self._SubsectionType = MODEL_UNKNOWN else: self._ValueList[0] = self._CurrentLine ## [LibraryClasses] section @ParseMacro def _LibraryClassParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT) if len(TokenList) < 2: EdkLogger.error('Parser', FORMAT_INVALID, "No library class or instance specified", ExtraData=self._CurrentLine + " (|)", File=self.MetaFile, Line=self._LineIndex + 1) if TokenList[0] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No library class specified", ExtraData=self._CurrentLine + " (|)", File=self.MetaFile, Line=self._LineIndex + 1) if TokenList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No library instance specified", ExtraData=self._CurrentLine + " (|)", File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0:len(TokenList)] = TokenList ## [BuildOptions] section parser @ParseMacro def _BuildOptionParser(self): self._CurrentLine = CleanString(self._CurrentLine, BuildOption=True) TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) TokenList2 = GetSplitValueList(TokenList[0], ':', 1) if len(TokenList2) == 2: self._ValueList[0] = TokenList2[0] # toolchain family self._ValueList[1] = TokenList2[1] # keys else: self._ValueList[1] = TokenList[0] if len(TokenList) == 2: # value self._ValueList[2] = TokenList[1] if self._ValueList[1].count('_') != 4: EdkLogger.error( 'Parser', FORMAT_INVALID, "'%s' must be in format of ____FLAGS" % self._ValueList[1], ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1 ) ## Override parent's method since we'll do all macro replacements in parser @property def _Macros(self): Macros = {} Macros.update(self._FileLocalMacros) Macros.update(self._GetApplicableSectionMacro()) Macros.update(GlobalData.gEdkGlobal) Macros.update(GlobalData.gPlatformDefines) Macros.update(GlobalData.gCommandLineDefines) # PCD cannot be referenced in macro definition if self._ItemType not in [MODEL_META_DATA_DEFINE, MODEL_META_DATA_GLOBAL_DEFINE]: Macros.update(self._Symbols) if GlobalData.BuildOptionPcd: for Item in GlobalData.BuildOptionPcd: if isinstance(Item, tuple): continue PcdName, TmpValue = Item.split("=") TmpValue = BuildOptionValue(TmpValue, self._GuidDict) Macros[PcdName.strip()] = TmpValue return Macros def _PostProcess(self): Processer = { MODEL_META_DATA_SECTION_HEADER : self.__ProcessSectionHeader, MODEL_META_DATA_SUBSECTION_HEADER : self.__ProcessSubsectionHeader, MODEL_META_DATA_HEADER : self.__ProcessDefine, MODEL_META_DATA_DEFINE : self.__ProcessDefine, MODEL_META_DATA_GLOBAL_DEFINE : self.__ProcessDefine, MODEL_META_DATA_INCLUDE : self.__ProcessDirective, MODEL_META_DATA_PACKAGE : self.__ProcessPackages, MODEL_META_DATA_CONDITIONAL_STATEMENT_IF : self.__ProcessDirective, MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSE : self.__ProcessDirective, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFDEF : self.__ProcessDirective, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF : self.__ProcessDirective, MODEL_META_DATA_CONDITIONAL_STATEMENT_ENDIF : self.__ProcessDirective, MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSEIF : self.__ProcessDirective, MODEL_EFI_SKU_ID : self.__ProcessSkuId, MODEL_EFI_DEFAULT_STORES : self.__ProcessDefaultStores, MODEL_EFI_LIBRARY_INSTANCE : self.__ProcessLibraryInstance, MODEL_EFI_LIBRARY_CLASS : self.__ProcessLibraryClass, MODEL_PCD_FIXED_AT_BUILD : self.__ProcessPcd, MODEL_PCD_PATCHABLE_IN_MODULE : self.__ProcessPcd, MODEL_PCD_FEATURE_FLAG : self.__ProcessPcd, MODEL_PCD_DYNAMIC_DEFAULT : self.__ProcessPcd, MODEL_PCD_DYNAMIC_HII : self.__ProcessPcd, MODEL_PCD_DYNAMIC_VPD : self.__ProcessPcd, MODEL_PCD_DYNAMIC_EX_DEFAULT : self.__ProcessPcd, MODEL_PCD_DYNAMIC_EX_HII : self.__ProcessPcd, MODEL_PCD_DYNAMIC_EX_VPD : self.__ProcessPcd, MODEL_META_DATA_COMPONENT : self.__ProcessComponent, MODEL_META_DATA_BUILD_OPTION : self.__ProcessBuildOption, MODEL_UNKNOWN : self._Skip, MODEL_META_DATA_USER_EXTENSION : self._SkipUserExtension, MODEL_META_DATA_CONDITIONAL_STATEMENT_ERROR : self._ProcessError, } self._Table = MetaFileStorage(self._RawTable.DB, self.MetaFile, MODEL_FILE_DSC, True) self._DirectiveStack = [] self._DirectiveEvalStack = [] self._FileWithError = self.MetaFile self._FileLocalMacros = {} self._SectionsMacroDict.clear() GlobalData.gPlatformDefines = {} # Get all macro and PCD which has straitforward value self.__RetrievePcdValue() self._Content = self._RawTable.GetAll() self._ContentIndex = 0 self._InSubsection = False while self._ContentIndex < len(self._Content) : Id, self._ItemType, V1, V2, V3, S1, S2, S3, Owner, self._From, \ LineStart, ColStart, LineEnd, ColEnd, Enabled = self._Content[self._ContentIndex] if self._From < 0: self._FileWithError = self.MetaFile self._ContentIndex += 1 self._Scope = [[S1, S2, S3]] # # For !include directive, handle it specially, # merge arch and module type in case of duplicate items # while self._ItemType == MODEL_META_DATA_INCLUDE: if self._ContentIndex >= len(self._Content): break Record = self._Content[self._ContentIndex] if LineStart == Record[10] and LineEnd == Record[12]: if [Record[5], Record[6], Record[7]] not in self._Scope: self._Scope.append([Record[5], Record[6], Record[7]]) self._ContentIndex += 1 else: break self._LineIndex = LineStart - 1 self._ValueList = [V1, V2, V3] if Owner > 0 and Owner in self._IdMapping: self._InSubsection = True else: self._InSubsection = False try: Processer[self._ItemType]() except EvaluationException as Excpt: # # Only catch expression evaluation error here. We need to report # the precise number of line on which the error occurred # if hasattr(Excpt, 'Pcd'): if Excpt.Pcd in GlobalData.gPlatformOtherPcds: Info = GlobalData.gPlatformOtherPcds[Excpt.Pcd] EdkLogger.error('Parser', FORMAT_INVALID, "Cannot use this PCD (%s) in an expression as" " it must be defined in a [PcdsFixedAtBuild] or [PcdsFeatureFlag] section" " of the DSC file, and it is currently defined in this section:" " %s, line #: %d." % (Excpt.Pcd, Info[0], Info[1]), File=self._FileWithError, ExtraData=' '.join(self._ValueList), Line=self._LineIndex + 1) else: EdkLogger.error('Parser', FORMAT_INVALID, "PCD (%s) is not defined in DSC file" % Excpt.Pcd, File=self._FileWithError, ExtraData=' '.join(self._ValueList), Line=self._LineIndex + 1) else: EdkLogger.error('Parser', FORMAT_INVALID, "Invalid expression: %s" % str(Excpt), File=self._FileWithError, ExtraData=' '.join(self._ValueList), Line=self._LineIndex + 1) except MacroException as Excpt: EdkLogger.error('Parser', FORMAT_INVALID, str(Excpt), File=self._FileWithError, ExtraData=' '.join(self._ValueList), Line=self._LineIndex + 1) if self._ValueList is None: continue NewOwner = self._IdMapping.get(Owner, -1) self._Enabled = int((not self._DirectiveEvalStack) or (False not in self._DirectiveEvalStack)) self._LastItem = self._Store( self._ItemType, self._ValueList[0], self._ValueList[1], self._ValueList[2], S1, S2, S3, NewOwner, self._From, self._LineIndex + 1, - 1, self._LineIndex + 1, - 1, self._Enabled ) self._IdMapping[Id] = self._LastItem GlobalData.gPlatformDefines.update(self._FileLocalMacros) self._PostProcessed = True self._Content = None def _ProcessError(self): if not self._Enabled: return EdkLogger.error('Parser', ERROR_STATEMENT, self._ValueList[1], File=self.MetaFile, Line=self._LineIndex + 1) def __ProcessSectionHeader(self): self._SectionName = self._ValueList[0] if self._SectionName in self.DataType: self._SectionType = self.DataType[self._SectionName] else: self._SectionType = MODEL_UNKNOWN def __ProcessSubsectionHeader(self): self._SubsectionName = self._ValueList[0] if self._SubsectionName in self.DataType: self._SubsectionType = self.DataType[self._SubsectionName] else: self._SubsectionType = MODEL_UNKNOWN def __RetrievePcdValue(self): try: with open(str(self.MetaFile), 'r') as File: Content = File.readlines() except: EdkLogger.error("Parser", FILE_READ_FAILURE, ExtraData=self.MetaFile) GlobalData.gPlatformOtherPcds['DSCFILE'] = str(self.MetaFile) for PcdType in (MODEL_PCD_PATCHABLE_IN_MODULE, MODEL_PCD_DYNAMIC_DEFAULT, MODEL_PCD_DYNAMIC_HII, MODEL_PCD_DYNAMIC_VPD, MODEL_PCD_DYNAMIC_EX_DEFAULT, MODEL_PCD_DYNAMIC_EX_HII, MODEL_PCD_DYNAMIC_EX_VPD): Records = self._RawTable.Query(PcdType, BelongsToItem= -1.0) for TokenSpaceGuid, PcdName, Value, Dummy2, Dummy3, Dummy4, ID, Line in Records: Name = TokenSpaceGuid + '.' + PcdName if Name not in GlobalData.gPlatformOtherPcds: PcdLine = Line while not Content[Line - 1].lstrip().startswith(TAB_SECTION_START): Line -= 1 GlobalData.gPlatformOtherPcds[Name] = (CleanString(Content[Line - 1]), PcdLine, PcdType) def __ProcessDefine(self): if not self._Enabled: return Type, Name, Value = self._ValueList Value = ReplaceMacro(Value, self._Macros, False) # # If it is , return # if self._InSubsection: self._ValueList = [Type, Name, Value] return if self._ItemType == MODEL_META_DATA_DEFINE: if self._SectionType == MODEL_META_DATA_HEADER: self._FileLocalMacros[Name] = Value else: self._ConstructSectionMacroDict(Name, Value) elif self._ItemType == MODEL_META_DATA_GLOBAL_DEFINE: GlobalData.gEdkGlobal[Name] = Value # # Keyword in [Defines] section can be used as Macros # if (self._ItemType == MODEL_META_DATA_HEADER) and (self._SectionType == MODEL_META_DATA_HEADER): self._FileLocalMacros[Name] = Value self._ValueList = [Type, Name, Value] def __ProcessDirective(self): Result = None if self._ItemType in [MODEL_META_DATA_CONDITIONAL_STATEMENT_IF, MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSEIF]: Macros = self._Macros Macros.update(GlobalData.gGlobalDefines) try: Result = ValueExpression(self._ValueList[1], Macros)() except SymbolNotFound as Exc: EdkLogger.debug(EdkLogger.DEBUG_5, str(Exc), self._ValueList[1]) Result = False except WrnExpression as Excpt: # # Catch expression evaluation warning here. We need to report # the precise number of line and return the evaluation result # EdkLogger.warn('Parser', "Suspicious expression: %s" % str(Excpt), File=self._FileWithError, ExtraData=' '.join(self._ValueList), Line=self._LineIndex + 1) Result = Excpt.result if self._ItemType in [MODEL_META_DATA_CONDITIONAL_STATEMENT_IF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFDEF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF]: self._DirectiveStack.append(self._ItemType) if self._ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_IF: Result = bool(Result) else: Macro = self._ValueList[1] Macro = Macro[2:-1] if (Macro.startswith("$(") and Macro.endswith(")")) else Macro Result = Macro in self._Macros if self._ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF: Result = not Result self._DirectiveEvalStack.append(Result) elif self._ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSEIF: self._DirectiveStack.append(self._ItemType) self._DirectiveEvalStack[-1] = not self._DirectiveEvalStack[-1] self._DirectiveEvalStack.append(bool(Result)) elif self._ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ELSE: self._DirectiveStack.append(self._ItemType) self._DirectiveEvalStack[-1] = not self._DirectiveEvalStack[-1] self._DirectiveEvalStack.append(True) elif self._ItemType == MODEL_META_DATA_CONDITIONAL_STATEMENT_ENDIF: # Back to the nearest !if/!ifdef/!ifndef while self._DirectiveStack: self._DirectiveEvalStack.pop() Directive = self._DirectiveStack.pop() if Directive in [MODEL_META_DATA_CONDITIONAL_STATEMENT_IF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFDEF, MODEL_META_DATA_CONDITIONAL_STATEMENT_IFNDEF]: break elif self._ItemType == MODEL_META_DATA_INCLUDE: # The included file must be relative to workspace or same directory as DSC file __IncludeMacros = {} # # Allow using system environment variables in path after !include # __IncludeMacros['WORKSPACE'] = GlobalData.gGlobalDefines['WORKSPACE'] # # Allow using MACROs comes from [Defines] section to keep compatible. # __IncludeMacros.update(self._Macros) IncludedFile = NormPath(ReplaceMacro(self._ValueList[1], __IncludeMacros, RaiseError=True)) # # First search the include file under the same directory as DSC file # IncludedFile1 = PathClass(IncludedFile, self.MetaFile.Dir) if self._Enabled: ErrorCode, ErrorInfo1 = IncludedFile1.Validate() if ErrorCode != 0: # # Also search file under the WORKSPACE directory # IncludedFile1 = PathClass(IncludedFile, GlobalData.gWorkspace) ErrorCode, ErrorInfo2 = IncludedFile1.Validate() if ErrorCode != 0: EdkLogger.error('parser', ErrorCode, File=self._FileWithError, Line=self._LineIndex + 1, ExtraData=ErrorInfo1 + "\n" + ErrorInfo2) self._FileWithError = IncludedFile1 FromItem = self._Content[self._ContentIndex - 1][0] if self._InSubsection: Owner = self._Content[self._ContentIndex - 1][8] else: Owner = self._Content[self._ContentIndex - 1][0] IncludedFileTable = MetaFileStorage(self._RawTable.DB, IncludedFile1, MODEL_FILE_DSC, False, FromItem=FromItem) Parser = DscParser(IncludedFile1, self._FileType, self._Arch, IncludedFileTable, Owner=Owner, From=FromItem) self.IncludedFiles.add (IncludedFile1) # set the parser status with current status Parser._SectionName = self._SectionName Parser._SubsectionType = self._SubsectionType Parser._InSubsection = self._InSubsection Parser._SectionType = self._SectionType Parser._Scope = self._Scope Parser._Enabled = self._Enabled # Parse the included file Parser.StartParse() # Insert all records in the table for the included file into dsc file table Records = IncludedFileTable.GetAll() if Records: self._Content[self._ContentIndex:self._ContentIndex] = Records self._Content.pop(self._ContentIndex - 1) self._ValueList = None self._ContentIndex -= 1 def __ProcessPackages(self): self._ValueList[0] = ReplaceMacro(self._ValueList[0], self._Macros) def __ProcessSkuId(self): self._ValueList = [ReplaceMacro(Value, self._Macros, RaiseError=True) for Value in self._ValueList] def __ProcessDefaultStores(self): self._ValueList = [ReplaceMacro(Value, self._Macros, RaiseError=True) for Value in self._ValueList] def __ProcessLibraryInstance(self): self._ValueList = [ReplaceMacro(Value, self._Macros) for Value in self._ValueList] def __ProcessLibraryClass(self): self._ValueList[1] = ReplaceMacro(self._ValueList[1], self._Macros, RaiseError=True) def __ProcessPcd(self): if self._ItemType not in [MODEL_PCD_FEATURE_FLAG, MODEL_PCD_FIXED_AT_BUILD]: self._ValueList[2] = ReplaceMacro(self._ValueList[2], self._Macros, RaiseError=True) return ValList, Valid, Index = AnalyzeDscPcd(self._ValueList[2], self._ItemType) if not Valid: if self._ItemType in (MODEL_PCD_DYNAMIC_DEFAULT, MODEL_PCD_DYNAMIC_EX_DEFAULT, MODEL_PCD_FIXED_AT_BUILD, MODEL_PCD_PATCHABLE_IN_MODULE): if ValList[1] != TAB_VOID and StructPattern.match(ValList[1]) is None and ValList[2]: EdkLogger.error('build', FORMAT_INVALID, "Pcd format incorrect. The datum type info should be VOID* or a valid struct name.", File=self._FileWithError, Line=self._LineIndex + 1, ExtraData="%s.%s|%s" % (self._ValueList[0], self._ValueList[1], self._ValueList[2])) EdkLogger.error('build', FORMAT_INVALID, "Pcd format incorrect.", File=self._FileWithError, Line=self._LineIndex + 1, ExtraData="%s.%s|%s" % (self._ValueList[0], self._ValueList[1], self._ValueList[2])) PcdValue = ValList[Index] if PcdValue and "." not in self._ValueList[0]: try: ValList[Index] = ValueExpression(PcdValue, self._Macros)(True) except WrnExpression as Value: ValList[Index] = Value.result except: pass if ValList[Index] == 'True': ValList[Index] = '1' if ValList[Index] == 'False': ValList[Index] = '0' if (not self._DirectiveEvalStack) or (False not in self._DirectiveEvalStack): GlobalData.gPlatformPcds[TAB_SPLIT.join(self._ValueList[0:2])] = PcdValue self._Symbols[TAB_SPLIT.join(self._ValueList[0:2])] = PcdValue try: self._ValueList[2] = '|'.join(ValList) except Exception: print(ValList) def __ProcessComponent(self): self._ValueList[0] = ReplaceMacro(self._ValueList[0], self._Macros) def __ProcessBuildOption(self): self._ValueList = [ReplaceMacro(Value, self._Macros, RaiseError=False) for Value in self._ValueList] def DisableOverrideComponent(self,module_id): for ori_id in self._IdMapping: if self._IdMapping[ori_id] == module_id: self._RawTable.DisableComponent(ori_id) _SectionParser = { MODEL_META_DATA_HEADER : _DefineParser, MODEL_EFI_SKU_ID : _SkuIdParser, MODEL_EFI_DEFAULT_STORES : _DefaultStoresParser, MODEL_EFI_LIBRARY_INSTANCE : _LibraryInstanceParser, MODEL_EFI_LIBRARY_CLASS : _LibraryClassParser, MODEL_PCD_FIXED_AT_BUILD : _PcdParser, MODEL_PCD_PATCHABLE_IN_MODULE : _PcdParser, MODEL_PCD_FEATURE_FLAG : _PcdParser, MODEL_PCD_DYNAMIC_DEFAULT : _PcdParser, MODEL_PCD_DYNAMIC_HII : _PcdParser, MODEL_PCD_DYNAMIC_VPD : _PcdParser, MODEL_PCD_DYNAMIC_EX_DEFAULT : _PcdParser, MODEL_PCD_DYNAMIC_EX_HII : _PcdParser, MODEL_PCD_DYNAMIC_EX_VPD : _PcdParser, MODEL_META_DATA_COMPONENT : _ComponentParser, MODEL_META_DATA_BUILD_OPTION : _BuildOptionParser, MODEL_UNKNOWN : MetaFileParser._Skip, MODEL_META_DATA_PACKAGE : MetaFileParser._PackageParser, MODEL_META_DATA_USER_EXTENSION : MetaFileParser._SkipUserExtension, MODEL_META_DATA_SECTION_HEADER : MetaFileParser._SectionHeaderParser, MODEL_META_DATA_SUBSECTION_HEADER : _SubsectionHeaderParser, } ## DEC file parser class # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Table Database used to retrieve module/package information # @param Macros Macros used for replacement in file # class DecParser(MetaFileParser): # DEC file supported data types (one type per section) DataType = { TAB_DEC_DEFINES.upper() : MODEL_META_DATA_HEADER, TAB_DSC_DEFINES_DEFINE : MODEL_META_DATA_DEFINE, TAB_INCLUDES.upper() : MODEL_EFI_INCLUDE, TAB_LIBRARY_CLASSES.upper() : MODEL_EFI_LIBRARY_CLASS, TAB_GUIDS.upper() : MODEL_EFI_GUID, TAB_PPIS.upper() : MODEL_EFI_PPI, TAB_PROTOCOLS.upper() : MODEL_EFI_PROTOCOL, TAB_PCDS_FIXED_AT_BUILD_NULL.upper() : MODEL_PCD_FIXED_AT_BUILD, TAB_PCDS_PATCHABLE_IN_MODULE_NULL.upper() : MODEL_PCD_PATCHABLE_IN_MODULE, TAB_PCDS_FEATURE_FLAG_NULL.upper() : MODEL_PCD_FEATURE_FLAG, TAB_PCDS_DYNAMIC_NULL.upper() : MODEL_PCD_DYNAMIC, TAB_PCDS_DYNAMIC_EX_NULL.upper() : MODEL_PCD_DYNAMIC_EX, TAB_USER_EXTENSIONS.upper() : MODEL_META_DATA_USER_EXTENSION, } ## Constructor of DecParser # # Initialize object of DecParser # # @param FilePath The path of platform description file # @param FileType The raw data of DSC file # @param Arch Default Arch value for filtering sections # @param Table Database used to retrieve module/package information # def __init__(self, FilePath, FileType, Arch, Table): # prevent re-initialization if hasattr(self, "_Table"): return MetaFileParser.__init__(self, FilePath, FileType, Arch, Table, -1) self._Comments = [] self._Version = 0x00010005 # Only EDK2 dec file is supported self._AllPCDs = [] # Only for check duplicate PCD self._AllPcdDict = {} self._CurrentStructurePcdName = "" self._include_flag = False self._package_flag = False self._RestofValue = "" ## Parser starter def Start(self): Content = '' try: with open(str(self.MetaFile), 'r') as File: Content = File.readlines() except: EdkLogger.error("Parser", FILE_READ_FAILURE, ExtraData=self.MetaFile) Content = self.ProcessMultipleLineCODEValue(Content) self._DefinesCount = 0 for Index in range(0, len(Content)): Line, Comment = CleanString2(Content[Index]) self._CurrentLine = Line self._LineIndex = Index # save comment for later use if Comment: self._Comments.append((Comment, self._LineIndex + 1)) # skip empty line if Line == '': continue # section header if Line[0] == TAB_SECTION_START and Line[-1] == TAB_SECTION_END: self._SectionHeaderParser() if self._SectionName == TAB_DEC_DEFINES.upper(): self._DefinesCount += 1 self._Comments = [] continue if self._SectionType == MODEL_UNKNOWN: EdkLogger.error("Parser", FORMAT_INVALID, "" "Not able to determine \"%s\" in which section."%self._CurrentLine, self.MetaFile, self._LineIndex + 1) elif len(self._SectionType) == 0: self._Comments = [] continue # section content self._ValueList = ['', '', ''] self._SectionParser[self._SectionType[0]](self) if self._ValueList is None or self._ItemType == MODEL_META_DATA_DEFINE: self._ItemType = -1 self._Comments = [] continue # # Model, Value1, Value2, Value3, Arch, BelongsToItem=-1, LineBegin=-1, # ColumnBegin=-1, LineEnd=-1, ColumnEnd=-1, FeatureFlag='', Enabled=-1 # for Arch, ModuleType, Type in self._Scope: self._LastItem = self._Store( Type, self._ValueList[0], self._ValueList[1], self._ValueList[2], Arch, ModuleType, self._Owner[-1], self._LineIndex + 1, - 1, self._LineIndex + 1, - 1, 0 ) for Comment, LineNo in self._Comments: self._Store( MODEL_META_DATA_COMMENT, Comment, self._ValueList[0], self._ValueList[1], Arch, ModuleType, self._LastItem, LineNo, - 1, LineNo, - 1, 0 ) self._Comments = [] if self._DefinesCount > 1: EdkLogger.error('Parser', FORMAT_INVALID, 'Multiple [Defines] section is exist.', self.MetaFile ) if self._DefinesCount == 0: EdkLogger.error('Parser', FORMAT_INVALID, 'No [Defines] section exist.', self.MetaFile) self._Done() ## Section header parser # # The section header is always in following format: # # [section_name.arch<.platform|module_type>] # def _SectionHeaderParser(self): self._Scope = [] self._SectionName = '' self._SectionType = [] ArchList = set() PrivateList = set() Line = re.sub(r',[\s]*', TAB_COMMA_SPLIT, self._CurrentLine) for Item in Line[1:-1].split(TAB_COMMA_SPLIT): if Item == '': EdkLogger.error("Parser", FORMAT_UNKNOWN_ERROR, "section name can NOT be empty or incorrectly use separator comma", self.MetaFile, self._LineIndex + 1, self._CurrentLine) ItemList = Item.split(TAB_SPLIT) # different types of PCD are permissible in one section self._SectionName = ItemList[0].upper() if self._SectionName == TAB_DEC_DEFINES.upper() and (len(ItemList) > 1 or len(Line.split(TAB_COMMA_SPLIT)) > 1): EdkLogger.error("Parser", FORMAT_INVALID, "Defines section format is invalid", self.MetaFile, self._LineIndex + 1, self._CurrentLine) if self._SectionName in self.DataType: if self.DataType[self._SectionName] not in self._SectionType: self._SectionType.append(self.DataType[self._SectionName]) else: EdkLogger.error("Parser", FORMAT_UNKNOWN_ERROR, "%s is not a valid section name" % Item, self.MetaFile, self._LineIndex + 1, self._CurrentLine) if MODEL_PCD_FEATURE_FLAG in self._SectionType and len(self._SectionType) > 1: EdkLogger.error( 'Parser', FORMAT_INVALID, "%s must not be in the same section of other types of PCD" % TAB_PCDS_FEATURE_FLAG_NULL, File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine ) # S1 is always Arch if len(ItemList) > 1: S1 = ItemList[1].upper() else: S1 = TAB_ARCH_COMMON ArchList.add(S1) # S2 may be Platform or ModuleType if len(ItemList) > 2: S2 = ItemList[2].upper() # only Includes, GUIDs, PPIs, Protocols section have Private tag if self._SectionName in [TAB_INCLUDES.upper(), TAB_GUIDS.upper(), TAB_PROTOCOLS.upper(), TAB_PPIS.upper()]: if S2 != 'PRIVATE': EdkLogger.error("Parser", FORMAT_INVALID, 'Please use keyword "Private" as section tag modifier.', File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) else: S2 = TAB_COMMON PrivateList.add(S2) if [S1, S2, self.DataType[self._SectionName]] not in self._Scope: self._Scope.append([S1, S2, self.DataType[self._SectionName]]) # 'COMMON' must not be used with specific ARCHs at the same section if TAB_ARCH_COMMON in ArchList and len(ArchList) > 1: EdkLogger.error('Parser', FORMAT_INVALID, "'common' ARCH must not be used with specific ARCHs", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) # It is not permissible to mix section tags without the Private attribute with section tags with the Private attribute if TAB_COMMON in PrivateList and len(PrivateList) > 1: EdkLogger.error('Parser', FORMAT_INVALID, "Can't mix section tags without the Private attribute with section tags with the Private attribute", File=self.MetaFile, Line=self._LineIndex + 1, ExtraData=self._CurrentLine) ## [guids], [ppis] and [protocols] section parser @ParseMacro def _GuidParser(self): TokenList = GetSplitValueList(self._CurrentLine, TAB_EQUAL_SPLIT, 1) if len(TokenList) < 2: EdkLogger.error('Parser', FORMAT_INVALID, "No GUID name or value specified", ExtraData=self._CurrentLine + " ( = )", File=self.MetaFile, Line=self._LineIndex + 1) if TokenList[0] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No GUID name specified", ExtraData=self._CurrentLine + " ( = )", File=self.MetaFile, Line=self._LineIndex + 1) if TokenList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No GUID value specified", ExtraData=self._CurrentLine + " ( = )", File=self.MetaFile, Line=self._LineIndex + 1) if TokenList[1][0] != '{' or TokenList[1][-1] != '}' or GuidStructureStringToGuidString(TokenList[1]) == '': EdkLogger.error('Parser', FORMAT_INVALID, "Invalid GUID value format", ExtraData=self._CurrentLine + \ " ( = )", File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[0] = TokenList[0] self._ValueList[1] = TokenList[1] if self._ValueList[0] not in self._GuidDict: self._GuidDict[self._ValueList[0]] = self._ValueList[1] def ParsePcdName(self,namelist): if "[" in namelist[1]: pcdname = namelist[1][:namelist[1].index("[")] arrayindex = namelist[1][namelist[1].index("["):] namelist[1] = pcdname if len(namelist) == 2: namelist.append(arrayindex) else: namelist[2] = ".".join((arrayindex,namelist[2])) return namelist ## PCD sections parser # # [PcdsFixedAtBuild] # [PcdsPatchableInModule] # [PcdsFeatureFlag] # [PcdsDynamicEx # [PcdsDynamic] # @ParseMacro def _PcdParser(self): if self._CurrentStructurePcdName: self._ValueList[0] = self._CurrentStructurePcdName if "|" not in self._CurrentLine: if "" == self._CurrentLine: self._include_flag = True self._package_flag = False self._ValueList = None return if "" == self._CurrentLine: self._package_flag = True self._ValueList = None self._include_flag = False return if self._include_flag: self._ValueList[1] = "_" + md5(self._CurrentLine.encode('utf-8')).hexdigest() self._ValueList[2] = self._CurrentLine if self._package_flag and "}" != self._CurrentLine: self._ValueList[1] = "_" + md5(self._CurrentLine.encode('utf-8')).hexdigest() self._ValueList[2] = self._CurrentLine if self._CurrentLine == "}": self._package_flag = False self._include_flag = False self._ValueList = None return else: PcdTockens = self._CurrentLine.split(TAB_VALUE_SPLIT) PcdNames = self.ParsePcdName(PcdTockens[0].split(TAB_SPLIT)) if len(PcdNames) == 2: if PcdNames[1].strip().endswith("]"): PcdName = PcdNames[1][:PcdNames[1].index('[')] Index = PcdNames[1][PcdNames[1].index('['):] self._ValueList[0] = TAB_SPLIT.join((PcdNames[0],PcdName)) self._ValueList[1] = Index self._ValueList[2] = PcdTockens[1] else: self._CurrentStructurePcdName = "" else: if self._CurrentStructurePcdName != TAB_SPLIT.join(PcdNames[:2]): EdkLogger.error('Parser', FORMAT_INVALID, "Pcd Name does not match: %s and %s " % (self._CurrentStructurePcdName, TAB_SPLIT.join(PcdNames[:2])), File=self.MetaFile, Line=self._LineIndex + 1) self._ValueList[1] = TAB_SPLIT.join(PcdNames[2:]) self._ValueList[2] = PcdTockens[1] if not self._CurrentStructurePcdName: if self._PcdDataTypeCODE: if ")}" in self._CurrentLine: ValuePart,RestofValue = self._CurrentLine.split(")}") self._PcdCodeValue = self._PcdCodeValue + "\n " + ValuePart self._CurrentLine = "|".join((self._CurrentPcdName, self._PcdCodeValue,RestofValue)) self._PcdDataTypeCODE = False self._PcdCodeValue = "" else: self._PcdCodeValue = self._PcdCodeValue + "\n " + self._CurrentLine self._ValueList = None return TokenList = GetSplitValueList(self._CurrentLine, TAB_VALUE_SPLIT, 1) self._CurrentPcdName = TokenList[0] if len(TokenList) == 2 and TokenList[1].strip().startswith("{CODE"): if ")}" in self._CurrentLine: self._PcdDataTypeCODE = False self._PcdCodeValue = "" else: self._PcdDataTypeCODE = True self._PcdCodeValue = TokenList[1].strip() self._ValueList = None return self._ValueList[0:1] = GetSplitValueList(TokenList[0], TAB_SPLIT) ValueRe = re.compile(r'^[a-zA-Z_][a-zA-Z0-9_]*') # check PCD information if self._ValueList[0] == '' or self._ValueList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No token space GUID or PCD name specified", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check format of token space GUID CName if not ValueRe.match(self._ValueList[0]): EdkLogger.error('Parser', FORMAT_INVALID, "The format of the token space GUID CName is invalid. The correct format is '(a-zA-Z_)[a-zA-Z0-9_]*'", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check format of PCD CName if not ValueRe.match(self._ValueList[1]): EdkLogger.error('Parser', FORMAT_INVALID, "The format of the PCD CName is invalid. The correct format is '(a-zA-Z_)[a-zA-Z0-9_]*'", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check PCD datum information if len(TokenList) < 2 or TokenList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "No PCD Datum information given", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) ValueRe = re.compile(r'^\s*L?\".*\|.*\"') PtrValue = ValueRe.findall(TokenList[1]) # Has VOID* type string, may contain "|" character in the string. if len(PtrValue) != 0: ptrValueList = re.sub(ValueRe, '', TokenList[1]) ValueList = AnalyzePcdExpression(ptrValueList) ValueList[0] = PtrValue[0] else: ValueList = AnalyzePcdExpression(TokenList[1]) # check if there's enough datum information given if len(ValueList) != 3: EdkLogger.error('Parser', FORMAT_INVALID, "Invalid PCD Datum information given", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check default value if ValueList[0] == '': EdkLogger.error('Parser', FORMAT_INVALID, "Missing DefaultValue in PCD Datum information", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check datum type if ValueList[1] == '': EdkLogger.error('Parser', FORMAT_INVALID, "Missing DatumType in PCD Datum information", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) # check token of the PCD if ValueList[2] == '': EdkLogger.error('Parser', FORMAT_INVALID, "Missing Token in PCD Datum information", ExtraData=self._CurrentLine + \ " (.|||)", File=self.MetaFile, Line=self._LineIndex + 1) PcdValue = ValueList[0] if PcdValue: try: self._GuidDict.update(self._AllPcdDict) ValueList[0] = ValueExpressionEx(ValueList[0], ValueList[1], self._GuidDict)(True) except BadExpression as Value: EdkLogger.error('Parser', FORMAT_INVALID, Value, ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) # check format of default value against the datum type IsValid, Cause = CheckPcdDatum(ValueList[1], ValueList[0]) if not IsValid: EdkLogger.error('Parser', FORMAT_INVALID, Cause, ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) if Cause == "StructurePcd": self._CurrentStructurePcdName = TAB_SPLIT.join(self._ValueList[0:2]) self._ValueList[0] = self._CurrentStructurePcdName self._ValueList[1] = ValueList[1].strip() if ValueList[0] in ['True', 'true', 'TRUE']: ValueList[0] = '1' elif ValueList[0] in ['False', 'false', 'FALSE']: ValueList[0] = '0' # check for duplicate PCD definition if (self._Scope[0], self._ValueList[0], self._ValueList[1]) in self._AllPCDs: EdkLogger.error('Parser', FORMAT_INVALID, "The same PCD name and GUID have been already defined", ExtraData=self._CurrentLine, File=self.MetaFile, Line=self._LineIndex + 1) else: self._AllPCDs.append((self._Scope[0], self._ValueList[0], self._ValueList[1])) self._AllPcdDict[TAB_SPLIT.join(self._ValueList[0:2])] = ValueList[0] self._ValueList[2] = ValueList[0].strip() + '|' + ValueList[1].strip() + '|' + ValueList[2].strip() _SectionParser = { MODEL_META_DATA_HEADER : MetaFileParser._DefineParser, MODEL_EFI_INCLUDE : MetaFileParser._PathParser, MODEL_EFI_LIBRARY_CLASS : MetaFileParser._PathParser, MODEL_EFI_GUID : _GuidParser, MODEL_EFI_PPI : _GuidParser, MODEL_EFI_PROTOCOL : _GuidParser, MODEL_PCD_FIXED_AT_BUILD : _PcdParser, MODEL_PCD_PATCHABLE_IN_MODULE : _PcdParser, MODEL_PCD_FEATURE_FLAG : _PcdParser, MODEL_PCD_DYNAMIC : _PcdParser, MODEL_PCD_DYNAMIC_EX : _PcdParser, MODEL_UNKNOWN : MetaFileParser._Skip, MODEL_META_DATA_USER_EXTENSION : MetaFileParser._SkipUserExtension, } ## # # This acts like the main() function for the script, unless it is 'import'ed into another # script. # if __name__ == '__main__': pass