123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188 |
- #
- # Metrix++, Copyright 2009-2013, Metrix++ Project
- # Link: http://metrixplusplus.sourceforge.net
- #
- # This file is a part of Metrix++ Tool.
- #
- # Metrix++ is free software: you can redistribute it and/or modify
- # it under the terms of the GNU General Public License as published by
- # the Free Software Foundation, version 3 of the License.
- #
- # Metrix++ is distributed in the hope that it will be useful,
- # but WITHOUT ANY WARRANTY; without even the implied warranty of
- # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- # GNU General Public License for more details.
- #
- # You should have received a copy of the GNU General Public License
- # along with Metrix++. If not, see <http://www.gnu.org/licenses/>.
- #
- import logging
- import time
- import re
- import core.log
- import core.db.loader
- import core.db.post
- import core.db.utils
- import core.cmdparser
- import core.export.convert
- def main():
-
- exit_code = 0
- log_plugin = core.log.Plugin()
- db_plugin = core.db.post.Plugin()
- parser = core.cmdparser.MultiOptionParser(usage="Usage: %prog [options] -- [path 1] ... [path N]")
- log_plugin.declare_configuration(parser)
- db_plugin.declare_configuration(parser)
- parser.add_option("--general.format", default='xml', choices=['txt', 'xml', 'python'], help="Format of the output data. "
- "Possible values are 'xml', 'txt' or 'python' [default: %default]")
- parser.add_option("--general.namespaces", default=None, help="Allows to enumerate namespaces of interest."
- " If not defined all namespaces available in database file will be processed."
- " Separate several namespaces by comma, for example 'general,std.code.complexity'"
- " [default: %default]")
- (options, args) = parser.parse_args()
- log_plugin.configure(options)
- db_plugin.configure(options)
- out_format = options.__dict__['general.format']
- namespaces = None
- if options.__dict__['general.namespaces'] != None:
- namespaces = re.split(',', options.__dict__['general.namespaces'])
- loader_prev = core.db.loader.Loader()
- if db_plugin.dbfile_prev != None:
- loader_prev.open_database(db_plugin.dbfile_prev)
- loader = core.db.loader.Loader()
- loader.open_database(db_plugin.dbfile)
-
- paths = None
- if len(args) == 0:
- paths = [""]
- else:
- paths = args
-
- if out_format == 'txt':
- print "=" * 80 + "\n" + "Export" + "\n" + "_" * 80 + "\n"
- elif out_format == 'xml':
- print "<export>"
- elif out_format == 'python':
- print "{'export': ["
- for (ind, path) in enumerate(paths):
- logging.info("Processing: " + re.sub(r'''[\\]''', "/", path))
-
- aggregated_data = loader.load_aggregated_data(path, namespaces=namespaces)
- aggregated_data_tree = {}
- subdirs = []
- subfiles = []
- if aggregated_data != None:
- aggregated_data_tree = aggregated_data.get_data_tree(namespaces=namespaces)
- subdirs = aggregated_data.get_subdirs()
- subfiles = aggregated_data.get_subfiles()
- else:
- logging.error("Specified path '" + path + "' is invalid (not found in the database records)")
- exit_code += 1
- aggregated_data_prev = loader_prev.load_aggregated_data(path, namespaces=namespaces)
- if aggregated_data_prev != None:
- aggregated_data_tree = append_diff(aggregated_data_tree,
- aggregated_data_prev.get_data_tree(namespaces=namespaces))
-
- file_data = loader.load_file_data(path)
- file_data_tree = {}
- if file_data != None:
- file_data_tree = file_data.get_data_tree(namespaces=namespaces)
- file_data_prev = loader_prev.load_file_data(path)
- regions_matcher = None
- if file_data_prev != None:
- file_data_tree = append_diff(file_data_tree,
- file_data_prev.get_data_tree(namespaces=namespaces))
- regions_matcher = core.db.utils.FileRegionsMatcher(file_data, file_data_prev)
-
- regions = []
- for each in file_data.iterate_regions():
- region_data_tree = each.get_data_tree(namespaces=namespaces)
- if regions_matcher != None and regions_matcher.is_matched(each.get_id()):
- region_data_prev = file_data_prev.get_region(regions_matcher.get_prev_id(each.get_id()))
- region_data_tree = append_diff(region_data_tree,
- region_data_prev.get_data_tree(namespaces=namespaces))
- regions.append({"info": {"name" : each.name,
- 'type' : file_data.get_region_types()().to_str(each.get_type()),
- "cursor" : each.cursor,
- 'line_begin': each.line_begin,
- 'line_end': each.line_end,
- 'offset_begin': each.begin,
- 'offset_end': each.end},
- "data": region_data_tree})
-
- file_data_tree['regions'] = regions
-
- data = {"info": {"path": path, "id": ind + 1},
- "aggregated-data": aggregated_data_tree,
- "file-data": file_data_tree,
- "subdirs": subdirs,
- "subfiles": subfiles}
- if out_format == 'txt':
- print core.export.convert.to_txt(data, root_name = "data")
- elif out_format == 'xml':
- print core.export.convert.to_xml(data, root_name = "data")
- elif out_format == 'python':
- postfix = ""
- if ind < len(paths) - 1:
- postfix = ", "
- print core.export.convert.to_python(data, root_name = "data") + postfix
- if out_format == 'txt':
- print "\n"
- elif out_format == 'xml':
- print "</export>"
- elif out_format == 'python':
- print "]}"
- return exit_code
- def append_diff(main_tree, prev_tree):
- assert(main_tree != None)
- assert(prev_tree != None)
-
- for name in main_tree.keys():
- if name not in prev_tree.keys():
- continue
- for field in main_tree[name].keys():
- if field not in prev_tree[name].keys():
- continue
- if isinstance(main_tree[name][field], dict) and isinstance(prev_tree[name][field], dict):
- diff = {}
- for key in main_tree[name][field].keys():
- if key not in prev_tree[name][field].keys():
- continue
- main_val = main_tree[name][field][key]
- prev_val = prev_tree[name][field][key]
- if main_val == None:
- main_val = 0
- if prev_val == None:
- prev_val = 0
- diff[key] = main_val - prev_val
- main_tree[name][field]['__diff__'] = diff
- elif (not isinstance(main_tree[name][field], dict)) and (not isinstance(prev_tree[name][field], dict)):
- if '__diff__' not in main_tree[name]:
- main_tree[name]['__diff__'] = {}
- main_tree[name]['__diff__'][field] = main_tree[name][field] - prev_tree[name][field]
- return main_tree
- if __name__ == '__main__':
- ts = time.time()
- core.log.set_default_format()
- exit_code = main()
- logging.warning("Exit code: " + str(exit_code) + ". Time spent: " + str(round((time.time() - ts), 2)) + " seconds. Done")
- exit(exit_code)
-
-
-
|