CloverBootloader/BaseTools/Source/Python/Trim/Trim.py
Sergey Isakov b1264ef1e3 changes of BaseTools
Signed-off-by: Sergey Isakov <isakov-sl@bk.ru>
2019-09-04 23:02:41 +03:00

552 lines
22 KiB
Python
Executable File

## @file
# Trim files preprocessed by compiler
#
# Copyright (c) 2007 - 2018, Intel Corporation. All rights reserved.<BR>
# SPDX-License-Identifier: BSD-2-Clause-Patent
#
##
# Import Modules
#
import Common.LongFilePathOs as os
import sys
import re
from io import BytesIO
import codecs
from optparse import OptionParser
from optparse import make_option
from Common.BuildToolError import *
from Common.Misc import *
from Common.DataType import *
from Common.BuildVersion import gBUILD_VERSION
import Common.EdkLogger as EdkLogger
from Common.LongFilePathSupport import OpenLongFilePath as open
# Version and Copyright
__version_number__ = ("0.10" + " " + gBUILD_VERSION)
__version__ = "%prog Version " + __version_number__
__copyright__ = "Copyright (c) 2007-2018, Intel Corporation. All rights reserved."
## Regular expression for matching Line Control directive like "#line xxx"
gLineControlDirective = re.compile('^\s*#(?:line)?\s+([0-9]+)\s+"*([^"]*)"')
## Regular expression for matching "typedef struct"
gTypedefPattern = re.compile("^\s*typedef\s+struct(\s+\w+)?\s*[{]*$", re.MULTILINE)
## Regular expression for matching "#pragma pack"
gPragmaPattern = re.compile("^\s*#pragma\s+pack", re.MULTILINE)
## Regular expression for matching "typedef"
gTypedef_SinglePattern = re.compile("^\s*typedef", re.MULTILINE)
## Regular expression for matching "typedef struct, typedef union, struct, union"
gTypedef_MulPattern = re.compile("^\s*(typedef)?\s+(struct|union)(\s+\w+)?\s*[{]*$", re.MULTILINE)
#
# The following number pattern match will only match if following criteria is met:
# There is leading non-(alphanumeric or _) character, and no following alphanumeric or _
# as the pattern is greedily match, so it is ok for the gDecNumberPattern or gHexNumberPattern to grab the maximum match
#
## Regular expression for matching HEX number
gHexNumberPattern = re.compile("(?<=[^a-zA-Z0-9_])(0[xX])([0-9a-fA-F]+)(U(?=$|[^a-zA-Z0-9_]))?")
## Regular expression for matching decimal number with 'U' postfix
gDecNumberPattern = re.compile("(?<=[^a-zA-Z0-9_])([0-9]+)U(?=$|[^a-zA-Z0-9_])")
## Regular expression for matching constant with 'ULL' 'LL' postfix
gLongNumberPattern = re.compile("(?<=[^a-zA-Z0-9_])(0[xX][0-9a-fA-F]+|[0-9]+)U?LL(?=$|[^a-zA-Z0-9_])")
## Regular expression for matching "Include ()" in asl file
gAslIncludePattern = re.compile("^(\s*)[iI]nclude\s*\(\"?([^\"\(\)]+)\"\)", re.MULTILINE)
## Regular expression for matching C style #include "XXX.asl" in asl file
gAslCIncludePattern = re.compile(r'^(\s*)#include\s*[<"]\s*([-\\/\w.]+)\s*([>"])', re.MULTILINE)
## Patterns used to convert EDK conventions to EDK2 ECP conventions
## file cache to avoid circular include in ASL file
gIncludedAslFile = []
## Trim preprocessed source code
#
# Remove extra content made by preprocessor. The preprocessor must enable the
# line number generation option when preprocessing.
#
# @param Source File to be trimmed
# @param Target File to store the trimmed content
# @param Convert If True, convert standard HEX format to MASM format
#
def TrimPreprocessedFile(Source, Target, ConvertHex, TrimLong):
CreateDirectory(os.path.dirname(Target))
try:
with open(Source, "r") as File:
Lines = File.readlines()
except IOError:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Source)
except:
EdkLogger.error("Trim", AUTOGEN_ERROR, "TrimPreprocessedFile: Error while processing file", File=Source)
PreprocessedFile = ""
InjectedFile = ""
LineIndexOfOriginalFile = None
NewLines = []
LineControlDirectiveFound = False
for Index in range(len(Lines)):
Line = Lines[Index]
#
# Find out the name of files injected by preprocessor from the lines
# with Line Control directive
#
MatchList = gLineControlDirective.findall(Line)
if MatchList != []:
MatchList = MatchList[0]
if len(MatchList) == 2:
LineNumber = int(MatchList[0], 0)
InjectedFile = MatchList[1]
InjectedFile = os.path.normpath(InjectedFile)
InjectedFile = os.path.normcase(InjectedFile)
# The first injected file must be the preprocessed file itself
if PreprocessedFile == "":
PreprocessedFile = InjectedFile
LineControlDirectiveFound = True
continue
elif PreprocessedFile == "" or InjectedFile != PreprocessedFile:
continue
if LineIndexOfOriginalFile is None:
#
# Any non-empty lines must be from original preprocessed file.
# And this must be the first one.
#
LineIndexOfOriginalFile = Index
EdkLogger.verbose("Found original file content starting from line %d"
% (LineIndexOfOriginalFile + 1))
if TrimLong:
Line = gLongNumberPattern.sub(r"\1", Line)
# convert HEX number format if indicated
if ConvertHex:
Line = gHexNumberPattern.sub(r"0\2h", Line)
else:
Line = gHexNumberPattern.sub(r"\1\2", Line)
# convert Decimal number format
Line = gDecNumberPattern.sub(r"\1", Line)
if LineNumber is not None:
EdkLogger.verbose("Got line directive: line=%d" % LineNumber)
# in case preprocessor removed some lines, like blank or comment lines
if LineNumber <= len(NewLines):
# possible?
NewLines[LineNumber - 1] = Line
else:
if LineNumber > (len(NewLines) + 1):
for LineIndex in range(len(NewLines), LineNumber-1):
NewLines.append(TAB_LINE_BREAK)
NewLines.append(Line)
LineNumber = None
EdkLogger.verbose("Now we have lines: %d" % len(NewLines))
else:
NewLines.append(Line)
# in case there's no line directive or linemarker found
if (not LineControlDirectiveFound) and NewLines == []:
MulPatternFlag = False
SinglePatternFlag = False
Brace = 0
for Index in range(len(Lines)):
Line = Lines[Index]
if MulPatternFlag == False and gTypedef_MulPattern.search(Line) is None:
if SinglePatternFlag == False and gTypedef_SinglePattern.search(Line) is None:
# remove "#pragram pack" directive
if gPragmaPattern.search(Line) is None:
NewLines.append(Line)
continue
elif SinglePatternFlag == False:
SinglePatternFlag = True
if Line.find(";") >= 0:
SinglePatternFlag = False
elif MulPatternFlag == False:
# found "typedef struct, typedef union, union, struct", keep its position and set a flag
MulPatternFlag = True
# match { and } to find the end of typedef definition
if Line.find("{") >= 0:
Brace += 1
elif Line.find("}") >= 0:
Brace -= 1
# "typedef struct, typedef union, union, struct" must end with a ";"
if Brace == 0 and Line.find(";") >= 0:
MulPatternFlag = False
# save to file
try:
with open(Target, 'w') as File:
File.writelines(NewLines)
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Target)
## Trim preprocessed VFR file
#
# Remove extra content made by preprocessor. The preprocessor doesn't need to
# enable line number generation option when preprocessing.
#
# @param Source File to be trimmed
# @param Target File to store the trimmed content
#
def TrimPreprocessedVfr(Source, Target):
CreateDirectory(os.path.dirname(Target))
try:
with open(Source, "r") as File:
Lines = File.readlines()
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Source)
# read whole file
FoundTypedef = False
Brace = 0
TypedefStart = 0
TypedefEnd = 0
for Index in range(len(Lines)):
Line = Lines[Index]
# don't trim the lines from "formset" definition to the end of file
if Line.strip() == 'formset':
break
if FoundTypedef == False and (Line.find('#line') == 0 or Line.find('# ') == 0):
# empty the line number directive if it's not aomong "typedef struct"
Lines[Index] = "\n"
continue
if FoundTypedef == False and gTypedefPattern.search(Line) is None:
# keep "#pragram pack" directive
if gPragmaPattern.search(Line) is None:
Lines[Index] = "\n"
continue
elif FoundTypedef == False:
# found "typedef struct", keept its position and set a flag
FoundTypedef = True
TypedefStart = Index
# match { and } to find the end of typedef definition
if Line.find("{") >= 0:
Brace += 1
elif Line.find("}") >= 0:
Brace -= 1
# "typedef struct" must end with a ";"
if Brace == 0 and Line.find(";") >= 0:
FoundTypedef = False
TypedefEnd = Index
# keep all "typedef struct" except to GUID, EFI_PLABEL and PAL_CALL_RETURN
if Line.strip("} ;\r\n") in [TAB_GUID, "EFI_PLABEL", "PAL_CALL_RETURN"]:
for i in range(TypedefStart, TypedefEnd+1):
Lines[i] = "\n"
# save all lines trimmed
try:
with open(Target, 'w') as File:
File.writelines(Lines)
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Target)
## Read the content ASL file, including ASL included, recursively
#
# @param Source File to be read
# @param Indent Spaces before the Include() statement
# @param IncludePathList The list of external include file
# @param LocalSearchPath If LocalSearchPath is specified, this path will be searched
# first for the included file; otherwise, only the path specified
# in the IncludePathList will be searched.
#
def DoInclude(Source, Indent='', IncludePathList=[], LocalSearchPath=None):
NewFileContent = []
try:
#
# Search LocalSearchPath first if it is specified.
#
if LocalSearchPath:
SearchPathList = [LocalSearchPath] + IncludePathList
else:
SearchPathList = IncludePathList
for IncludePath in SearchPathList:
IncludeFile = os.path.join(IncludePath, Source)
if os.path.isfile(IncludeFile):
try:
with open(IncludeFile, "r") as File:
F = File.readlines()
except:
with codecs.open(IncludeFile, "r", encoding='utf-8') as File:
F = File.readlines()
break
else:
EdkLogger.error("Trim", "Failed to find include file %s" % Source)
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Source)
# avoid A "include" B and B "include" A
IncludeFile = os.path.abspath(os.path.normpath(IncludeFile))
if IncludeFile in gIncludedAslFile:
EdkLogger.warn("Trim", "Circular include",
ExtraData= "%s -> %s" % (" -> ".join(gIncludedAslFile), IncludeFile))
return []
gIncludedAslFile.append(IncludeFile)
for Line in F:
LocalSearchPath = None
Result = gAslIncludePattern.findall(Line)
if len(Result) == 0:
Result = gAslCIncludePattern.findall(Line)
if len(Result) == 0 or os.path.splitext(Result[0][1])[1].lower() not in [".asl", ".asi"]:
NewFileContent.append("%s%s" % (Indent, Line))
continue
#
# We should first search the local directory if current file are using pattern #include "XXX"
#
if Result[0][2] == '"':
LocalSearchPath = os.path.dirname(IncludeFile)
CurrentIndent = Indent + Result[0][0]
IncludedFile = Result[0][1]
NewFileContent.extend(DoInclude(IncludedFile, CurrentIndent, IncludePathList, LocalSearchPath))
NewFileContent.append("\n")
gIncludedAslFile.pop()
return NewFileContent
## Trim ASL file
#
# Replace ASL include statement with the content the included file
#
# @param Source File to be trimmed
# @param Target File to store the trimmed content
# @param IncludePathFile The file to log the external include path
#
def TrimAslFile(Source, Target, IncludePathFile):
CreateDirectory(os.path.dirname(Target))
SourceDir = os.path.dirname(Source)
if SourceDir == '':
SourceDir = '.'
#
# Add source directory as the first search directory
#
IncludePathList = [SourceDir]
#
# If additional include path file is specified, append them all
# to the search directory list.
#
if IncludePathFile:
try:
LineNum = 0
with open(IncludePathFile, 'r') as File:
FileLines = File.readlines()
for Line in FileLines:
LineNum += 1
if Line.startswith("/I") or Line.startswith ("-I"):
IncludePathList.append(Line[2:].strip())
else:
EdkLogger.warn("Trim", "Invalid include line in include list file.", IncludePathFile, LineNum)
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=IncludePathFile)
Lines = DoInclude(Source, '', IncludePathList)
#
# Undef MIN and MAX to avoid collision in ASL source code
#
Lines.insert(0, "#undef MIN\n#undef MAX\n")
# save all lines trimmed
try:
with open(Target, 'w') as File:
File.writelines(Lines)
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, ExtraData=Target)
def GenerateVfrBinSec(ModuleName, DebugDir, OutputFile):
VfrNameList = []
if os.path.isdir(DebugDir):
for CurrentDir, Dirs, Files in os.walk(DebugDir):
for FileName in Files:
Name, Ext = os.path.splitext(FileName)
if Ext == '.c' and Name != 'AutoGen':
VfrNameList.append (Name + 'Bin')
VfrNameList.append (ModuleName + 'Strings')
EfiFileName = os.path.join(DebugDir, ModuleName + '.efi')
MapFileName = os.path.join(DebugDir, ModuleName + '.map')
VfrUniOffsetList = GetVariableOffset(MapFileName, EfiFileName, VfrNameList)
if not VfrUniOffsetList:
return
try:
fInputfile = open(OutputFile, "wb+")
except:
EdkLogger.error("Trim", FILE_OPEN_FAILURE, "File open failed for %s" %OutputFile, None)
# Use a instance of BytesIO to cache data
fStringIO = BytesIO()
for Item in VfrUniOffsetList:
if (Item[0].find("Strings") != -1):
#
# UNI offset in image.
# GUID + Offset
# { 0x8913c5e0, 0x33f6, 0x4d86, { 0x9b, 0xf1, 0x43, 0xef, 0x89, 0xfc, 0x6, 0x66 } }
#
UniGuid = b'\xe0\xc5\x13\x89\xf63\x86M\x9b\xf1C\xef\x89\xfc\x06f'
fStringIO.write(UniGuid)
UniValue = pack ('Q', int (Item[1], 16))
fStringIO.write (UniValue)
else:
#
# VFR binary offset in image.
# GUID + Offset
# { 0xd0bc7cb4, 0x6a47, 0x495f, { 0xaa, 0x11, 0x71, 0x7, 0x46, 0xda, 0x6, 0xa2 } };
#
VfrGuid = b'\xb4|\xbc\xd0Gj_I\xaa\x11q\x07F\xda\x06\xa2'
fStringIO.write(VfrGuid)
type (Item[1])
VfrValue = pack ('Q', int (Item[1], 16))
fStringIO.write (VfrValue)
#
# write data into file.
#
try :
fInputfile.write (fStringIO.getvalue())
except:
EdkLogger.error("Trim", FILE_WRITE_FAILURE, "Write data to file %s failed, please check whether the file been locked or using by other applications." %OutputFile, None)
fStringIO.close ()
fInputfile.close ()
## Parse command line options
#
# Using standard Python module optparse to parse command line option of this tool.
#
# @retval Options A optparse.Values object containing the parsed options
# @retval InputFile Path of file to be trimmed
#
def Options():
OptionList = [
make_option("-s", "--source-code", dest="FileType", const="SourceCode", action="store_const",
help="The input file is preprocessed source code, including C or assembly code"),
make_option("-r", "--vfr-file", dest="FileType", const="Vfr", action="store_const",
help="The input file is preprocessed VFR file"),
make_option("--Vfr-Uni-Offset", dest="FileType", const="VfrOffsetBin", action="store_const",
help="The input file is EFI image"),
make_option("-a", "--asl-file", dest="FileType", const="Asl", action="store_const",
help="The input file is ASL file"),
make_option("-c", "--convert-hex", dest="ConvertHex", action="store_true",
help="Convert standard hex format (0xabcd) to MASM format (abcdh)"),
make_option("-l", "--trim-long", dest="TrimLong", action="store_true",
help="Remove postfix of long number"),
make_option("-i", "--include-path-file", dest="IncludePathFile",
help="The input file is include path list to search for ASL include file"),
make_option("-o", "--output", dest="OutputFile",
help="File to store the trimmed content"),
make_option("--ModuleName", dest="ModuleName", help="The module's BASE_NAME"),
make_option("--DebugDir", dest="DebugDir",
help="Debug Output directory to store the output files"),
make_option("-v", "--verbose", dest="LogLevel", action="store_const", const=EdkLogger.VERBOSE,
help="Run verbosely"),
make_option("-d", "--debug", dest="LogLevel", type="int",
help="Run with debug information"),
make_option("-q", "--quiet", dest="LogLevel", action="store_const", const=EdkLogger.QUIET,
help="Run quietly"),
make_option("-?", action="help", help="show this help message and exit"),
]
# use clearer usage to override default usage message
UsageString = "%prog [-s|-r|-a|--Vfr-Uni-Offset] [-c] [-v|-d <debug_level>|-q] [-i <include_path_file>] [-o <output_file>] [--ModuleName <ModuleName>] [--DebugDir <DebugDir>] [<input_file>]"
Parser = OptionParser(description=__copyright__, version=__version__, option_list=OptionList, usage=UsageString)
Parser.set_defaults(FileType="Vfr")
Parser.set_defaults(ConvertHex=False)
Parser.set_defaults(LogLevel=EdkLogger.INFO)
Options, Args = Parser.parse_args()
# error check
if Options.FileType == 'VfrOffsetBin':
if len(Args) == 0:
return Options, ''
elif len(Args) > 1:
EdkLogger.error("Trim", OPTION_NOT_SUPPORTED, ExtraData=Parser.get_usage())
if len(Args) == 0:
EdkLogger.error("Trim", OPTION_MISSING, ExtraData=Parser.get_usage())
if len(Args) > 1:
EdkLogger.error("Trim", OPTION_NOT_SUPPORTED, ExtraData=Parser.get_usage())
InputFile = Args[0]
return Options, InputFile
## Entrance method
#
# This method mainly dispatch specific methods per the command line options.
# If no error found, return zero value so the caller of this tool can know
# if it's executed successfully or not.
#
# @retval 0 Tool was successful
# @retval 1 Tool failed
#
def Main():
try:
EdkLogger.Initialize()
CommandOptions, InputFile = Options()
if CommandOptions.LogLevel < EdkLogger.DEBUG_9:
EdkLogger.SetLevel(CommandOptions.LogLevel + 1)
else:
EdkLogger.SetLevel(CommandOptions.LogLevel)
except FatalError as X:
return 1
try:
if CommandOptions.FileType == "Vfr":
if CommandOptions.OutputFile is None:
CommandOptions.OutputFile = os.path.splitext(InputFile)[0] + '.iii'
TrimPreprocessedVfr(InputFile, CommandOptions.OutputFile)
elif CommandOptions.FileType == "Asl":
if CommandOptions.OutputFile is None:
CommandOptions.OutputFile = os.path.splitext(InputFile)[0] + '.iii'
TrimAslFile(InputFile, CommandOptions.OutputFile, CommandOptions.IncludePathFile)
elif CommandOptions.FileType == "VfrOffsetBin":
GenerateVfrBinSec(CommandOptions.ModuleName, CommandOptions.DebugDir, CommandOptions.OutputFile)
else :
if CommandOptions.OutputFile is None:
CommandOptions.OutputFile = os.path.splitext(InputFile)[0] + '.iii'
TrimPreprocessedFile(InputFile, CommandOptions.OutputFile, CommandOptions.ConvertHex, CommandOptions.TrimLong)
except FatalError as X:
import platform
import traceback
if CommandOptions is not None and CommandOptions.LogLevel <= EdkLogger.DEBUG_9:
EdkLogger.quiet("(Python %s on %s) " % (platform.python_version(), sys.platform) + traceback.format_exc())
return 1
except:
import traceback
import platform
EdkLogger.error(
"\nTrim",
CODE_ERROR,
"Unknown fatal error when trimming [%s]" % InputFile,
ExtraData="\n(Please send email to %s for help, attaching following call stack trace!)\n" % MSG_EDKII_MAIL_ADDR,
RaiseError=False
)
EdkLogger.quiet("(Python %s on %s) " % (platform.python_version(), sys.platform) + traceback.format_exc())
return 1
return 0
if __name__ == '__main__':
r = Main()
## 0-127 is a safe return range, and 1 is a standard default error
if r < 0 or r > 127: r = 1
sys.exit(r)