| # Copyright Deniz Bahadir 2015 |
| # |
| # Distributed under the Boost Software License, Version 1.0. |
| # (See accompanying file LICENSE_1_0.txt or copy at |
| # http://www.boost.org/LICENSE_1_0.txt) |
| # |
| # See http://www.boost.org/libs/mpl for documentation. |
| # See http://stackoverflow.com/a/20660264/3115457 for further information. |
| # See http://stackoverflow.com/a/29627158/3115457 for further information. |
| |
| import fix_boost_mpl_preprocess as fixmpl |
| import argparse |
| import sys |
| import os |
| import os.path |
| import re |
| import fileinput |
| import shutil |
| |
| |
| def create_more_container_files(sourceDir, suffix, maxElements, containers, containers2): |
| """Creates additional files for the individual MPL-containers.""" |
| |
| # Create files for each MPL-container with 20 to 'maxElements' elements |
| # which will be used during generation. |
| for container in containers: |
| for i in range(20, maxElements, 10): |
| # Create copy of "template"-file. |
| newFile = os.path.join( sourceDir, container, container + str(i+10) + suffix ) |
| shutil.copyfile( os.path.join( sourceDir, container, container + "20" + suffix ), newFile ) |
| # Adjust copy of "template"-file accordingly. |
| for line in fileinput.input( newFile, inplace=1, mode="rU" ): |
| line = re.sub(r'20', '%TWENTY%', line.rstrip()) |
| line = re.sub(r'11', '%ELEVEN%', line.rstrip()) |
| line = re.sub(r'10(?![0-9])', '%TEN%', line.rstrip()) |
| line = re.sub(r'%TWENTY%', re.escape(str(i+10)), line.rstrip()) |
| line = re.sub(r'%ELEVEN%', re.escape(str(i + 1)), line.rstrip()) |
| line = re.sub(r'%TEN%', re.escape(str(i)), line.rstrip()) |
| print(line) |
| for container in containers2: |
| for i in range(20, maxElements, 10): |
| # Create copy of "template"-file. |
| newFile = os.path.join( sourceDir, container, container + str(i+10) + "_c" + suffix ) |
| shutil.copyfile( os.path.join( sourceDir, container, container + "20_c" + suffix ), newFile ) |
| # Adjust copy of "template"-file accordingly. |
| for line in fileinput.input( newFile, inplace=1, mode="rU" ): |
| line = re.sub(r'20', '%TWENTY%', line.rstrip()) |
| line = re.sub(r'11', '%ELEVEN%', line.rstrip()) |
| line = re.sub(r'10(?![0-9])', '%TEN%', line.rstrip()) |
| line = re.sub(r'%TWENTY%', re.escape(str(i+10)), line.rstrip()) |
| line = re.sub(r'%ELEVEN%', re.escape(str(i + 1)), line.rstrip()) |
| line = re.sub(r'%TEN%', re.escape(str(i)), line.rstrip()) |
| print(line) |
| |
| |
| def create_input_for_numbered_sequences(headerDir, sourceDir, containers, maxElements): |
| """Creates additional source- and header-files for the numbered sequence MPL-containers.""" |
| # Create additional container-list without "map". |
| containersWithoutMap = containers[:] |
| try: |
| containersWithoutMap.remove('map') |
| except ValueError: |
| # We can safely ignore if "map" is not contained in 'containers'! |
| pass |
| # Create header/source-files. |
| create_more_container_files(headerDir, ".hpp", maxElements, containers, containersWithoutMap) |
| create_more_container_files(sourceDir, ".cpp", maxElements, containers, containersWithoutMap) |
| |
| |
| def adjust_container_limits_for_variadic_sequences(headerDir, containers, maxElements): |
| """Adjusts the limits of variadic sequence MPL-containers.""" |
| for container in containers: |
| headerFile = os.path.join( headerDir, "limits", container + ".hpp" ) |
| regexMatch = r'(define\s+BOOST_MPL_LIMIT_' + container.upper() + r'_SIZE\s+)[0-9]+' |
| regexReplace = r'\g<1>' + re.escape( str(maxElements) ) |
| for line in fileinput.input( headerFile, inplace=1, mode="rU" ): |
| line = re.sub(regexMatch, regexReplace, line.rstrip()) |
| print(line) |
| |
| |
| def current_boost_dir(): |
| """Returns the (relative) path to the Boost source-directory this file is located in (if any).""" |
| # Path to directory containing this script. |
| path = os.path.dirname( os.path.realpath(__file__) ) |
| # Making sure it is located in "${boost-dir}/libs/mpl/preprocessed". |
| for directory in reversed( ["libs", "mpl", "preprocessed"] ): |
| (head, tail) = os.path.split(path) |
| if tail == directory: |
| path = head |
| else: |
| return None |
| return os.path.relpath( path ) |
| |
| |
| |
| def to_positive_multiple_of_10(string): |
| """Converts a string into its encoded positive integer (greater zero) or throws an exception.""" |
| try: |
| value = int(string) |
| except ValueError: |
| msg = '"%r" is not a positive multiple of 10 (greater zero).' % string |
| raise argparse.ArgumentTypeError(msg) |
| if value <= 0 or value % 10 != 0: |
| msg = '"%r" is not a positive multiple of 10 (greater zero).' % string |
| raise argparse.ArgumentTypeError(msg) |
| return value |
| |
| |
| def to_existing_absolute_path(string): |
| """Converts a path into its absolute path and verifies that it exists or throws an exception.""" |
| value = os.path.abspath(string) |
| if not os.path.exists( value ) or not os.path.isdir( value ): |
| msg = '"%r" is not a valid path to a directory.' % string |
| raise argparse.ArgumentTypeError(msg) |
| return value |
| |
| |
| def main(): |
| """The main function.""" |
| |
| # Find the current Boost source-directory in which this script is located. |
| sourceDir = current_boost_dir() |
| if sourceDir == None: |
| sourceDir = "" |
| |
| # Prepare and run cmdline-parser. |
| cmdlineParser = argparse.ArgumentParser(description="A generator-script for pre-processed Boost.MPL headers.") |
| cmdlineParser.add_argument("-v", "--verbose", dest='verbose', action='store_true', |
| help="Be a little bit more verbose.") |
| cmdlineParser.add_argument("-s", "--sequence-type", dest='seqType', choices=['variadic', 'numbered', 'both'], |
| default='both', |
| help="Only update pre-processed headers for the selected sequence types, " |
| "either 'numbered' sequences, 'variadic' sequences or 'both' sequence " |
| "types. (Default=both)") |
| cmdlineParser.add_argument("--no-vector", dest='want_vector', action='store_false', |
| help="Do not update pre-processed headers for Boost.MPL Vector.") |
| cmdlineParser.add_argument("--no-list", dest='want_list', action='store_false', |
| help="Do not update pre-processed headers for Boost.MPL List.") |
| cmdlineParser.add_argument("--no-set", dest='want_set', action='store_false', |
| help="Do not update pre-processed headers for Boost.MPL Set.") |
| cmdlineParser.add_argument("--no-map", dest='want_map', action='store_false', |
| help="Do not update pre-processed headers for Boost.MPL Map.") |
| cmdlineParser.add_argument("--num-elements", dest='numElements', metavar="<num-elements>", |
| type=to_positive_multiple_of_10, default=100, |
| help="The maximal number of elements per container sequence. (Default=100)") |
| cmdlineParser.add_argument(dest='sourceDir', metavar="<source-dir>", default=current_boost_dir(), nargs='?', |
| type=to_existing_absolute_path, |
| help="The source-directory of Boost. (Default=\"" + sourceDir + "\")") |
| args = cmdlineParser.parse_args() |
| |
| # Some verbose debug output. |
| if args.verbose: |
| print "Arguments extracted from command-line:" |
| print " verbose = ", args.verbose |
| print " source directory = ", args.sourceDir |
| print " num elements = ", args.numElements |
| print " sequence type = ", args.seqType |
| print " want: vector = ", args.want_vector |
| print " want: list = ", args.want_list |
| print " want: set = ", args.want_set |
| print " want: map = ", args.want_map |
| |
| # Verify that we received any source-directory. |
| if args.sourceDir == None: |
| print "You should specify a valid path to the Boost source-directory." |
| sys.exit(0) |
| |
| # The directories for header- and source files of Boost.MPL. |
| # NOTE: Assuming 'args.sourceDir' is the source-directory of the entire boost project. |
| headerDir = os.path.join( args.sourceDir, "boost", "mpl" ) |
| sourceDir = os.path.join( args.sourceDir, "libs", "mpl", "preprocessed" ) |
| # Check that the header/source-directories exist. |
| if not os.path.exists( headerDir ) or not os.path.exists( sourceDir ): |
| # Maybe 'args.sourceDir' is not the source-directory of the entire boost project |
| # but instead of the Boost.MPL git-directory, only? |
| headerDir = os.path.join( args.sourceDir, "include", "boost", "mpl" ) |
| sourceDir = os.path.join( args.sourceDir, "preprocessed" ) |
| if not os.path.exists( headerDir ) or not os.path.exists( sourceDir ): |
| cmdlineParser.print_usage() |
| print "error: Cannot find Boost.MPL header/source files in given Boost source-directory!" |
| sys.exit(0) |
| |
| # Some verbose debug output. |
| if args.verbose: |
| print "Chosen header-directory: ", headerDir |
| print "Chosen source-directory: ", sourceDir |
| |
| # Create list of containers for which files shall be pre-processed. |
| containers = [] |
| if args.want_vector: |
| containers.append('vector') |
| if args.want_list: |
| containers.append('list') |
| if args.want_set: |
| containers.append('set') |
| if args.want_map: |
| containers.append('map') |
| if containers == []: |
| print "Nothing to do." |
| print "(Why did you prevent generating pre-processed headers for all Boost.MPL container types?)" |
| sys.exit(0) |
| |
| # Possibly fix the header-comments of input-files needed for pre-processing. |
| if args.verbose: |
| print "Checking if prior to pre-processing some input-files need fixing." |
| needFixing = fixmpl.check_input_files(headerDir, sourceDir, containers, args.seqType, args.verbose) |
| if needFixing: |
| if args.verbose: |
| print "Fixing of some input-files prior to pre-processing is needed." |
| print "Will fix them now!" |
| fixmpl.fix_input_files(headerDir, sourceDir, containers, args.seqType, args.verbose) |
| |
| # Some verbose debug output. |
| if args.verbose: |
| print "Containers for which to pre-process headers: ", containers |
| |
| # Create (additional) input files for generating pre-processed headers of numbered sequence MPL containers. |
| if args.seqType == "both" or args.seqType == "numbered": |
| create_input_for_numbered_sequences(headerDir, sourceDir, containers, args.numElements) |
| # Modify settings for generating pre-processed headers of variadic sequence MPL containers. |
| if args.seqType == "both" or args.seqType == "variadic": |
| adjust_container_limits_for_variadic_sequences(headerDir, containers, args.numElements) |
| |
| # Generate MPL-preprocessed files. |
| os.chdir( sourceDir ) |
| if args.seqType == "both" or args.seqType == "numbered": |
| if args.want_vector: |
| if args.verbose: |
| print "Pre-process headers for Boost.MPL numbered vectors." |
| os.system( "python " + os.path.join( sourceDir, "preprocess_vector.py" ) + " all " + args.sourceDir ) |
| if args.want_list: |
| if args.verbose: |
| print "Pre-process headers for Boost.MPL numbered lists." |
| os.system( "python " + os.path.join( sourceDir, "preprocess_list.py" ) + " all " + args.sourceDir ) |
| if args.want_set: |
| if args.verbose: |
| print "Pre-process headers for Boost.MPL numbered sets." |
| os.system( "python " + os.path.join( sourceDir, "preprocess_set.py" ) + " all " + args.sourceDir ) |
| if args.want_map: |
| if args.verbose: |
| print "Pre-process headers for Boost.MPL numbered maps." |
| os.system( "python " + os.path.join( sourceDir, "preprocess_map.py" ) + " all " + args.sourceDir ) |
| if args.seqType == "both" or args.seqType == "variadic": |
| if args.verbose: |
| print "Pre-process headers for Boost.MPL variadic containers." |
| os.system( "python " + os.path.join( sourceDir, "preprocess.py" ) + " all " + args.sourceDir ) |
| |
| |
| if __name__ == '__main__': |
| main() |