def __init__(self, command_targets, load_targets, blade_path, working_dir, build_path, blade_root_dir, blade_options, command): """init method. """ self.__command_targets = command_targets self.__load_targets = load_targets self.__blade_path = blade_path self.__working_dir = working_dir self.__build_path = build_path self.__root_dir = blade_root_dir self.__options = blade_options self.__command = command # Source dir of current loading BUILD file self.__current_source_path = blade_root_dir # The direct targets that are used for analyzing self.__direct_targets = [] # All command targets, make sure that all targets specified with ... # are all in the list now self.__all_command_targets = [] # Given some targets specified in the command line, Blade will load # BUILD files containing these command line targets; global target # functions, i.e., cc_library, cc_binary and etc, in these BUILD # files will register targets into target_database, which then becomes # the input to dependency analyzer and SCons rules generator. It is # notable that not all targets in target_database are dependencies of # command line targets. self.__target_database = {} # targets to build after loading the build files. self.__build_targets = {} # The targets keys list after sorting by topological sorting method. # Used to generate build rules in correct order. self.__sorted_targets_keys = [] # The depended targets dict after topological sorting self.__depended_targets = {} # Indicate whether the deps list is expanded by expander or not self.__targets_expanded = False self.__build_time = time.time() self.__build_platform = BuildPlatform() self.build_environment = BuildEnvironment(self.__root_dir) self.svn_root_dirs = [] self._verify_history_path = os.path.join(build_path, '.blade_verify.json') self._verify_history = { 'header_inclusion_dependencies': {}, # path(.H) -> mtime(modification time) }
def __init__(self, command_targets, load_targets, blade_path, working_dir, build_dir, blade_root_dir, blade_options, command): """init method. Args: command_targets: List[str], target patterns are specified in command line. load_targets: List[str], target patterns should be loaded from workspace. It usually should be same as the command_targets, but in query dependents mode, all targets should be loaded. blade_path: str, the path of the `blade` python module, used to be called by builtin tools. """ self.__command_targets = command_targets self.__load_targets = load_targets self.__blade_path = blade_path self.__working_dir = working_dir self.__build_dir = build_dir self.__root_dir = blade_root_dir self.__options = blade_options self.__command = command # Source dir of current loading BUILD file self.__current_source_path = blade_root_dir # The targets which are specified in command line explicitly, not pattern expanded. self.__direct_targets = [] # All command targets, includes direct targets and expanded target patterns. self.__expanded_command_targets = [] # Given some targets specified in the command line, Blade will load # BUILD files containing these command line targets; global target # functions, i.e., cc_library, cc_binary and etc, in these BUILD # files will register targets into target_database, which then becomes # the input to dependency analyzer and rules generator. It is # notable that not all targets in target_database are dependencies of # command line targets. self.__target_database = {} # The targets to be build after loading the build files. self.__build_targets = {} # The targets keys list after sorting by topological sorting method. # Used to generate build rules in correct order. self.__sorted_targets_keys = [] # The depended targets dict after topological sorting self.__depended_targets = {} # Indicate whether the deps list is expanded by expander or not self.__targets_expanded = False self.__build_time = time.time() self.__build_platform = BuildPlatform() self.build_environment = BuildEnvironment(self.__root_dir) self.__build_jobs_num = 0 self.__test_jobs_num = 0 self.svn_root_dirs = [] self._verify_history_path = os.path.join(build_dir, '.blade_verify.json') self._verify_history = { 'header_inclusion_dependencies': {}, # path(.H) -> mtime(modification time) } self.__build_script = os.path.join(self.__build_dir, 'build.ninja')
class Blade(object): """Blade. A blade manager class. """ # pylint: disable=too-many-public-methods def __init__(self, command_targets, load_targets, blade_path, working_dir, build_dir, blade_root_dir, blade_options, command): """init method. Args: command_targets: List[str], target patterns are specified in command line. load_targets: List[str], target patterns should be loaded from workspace. It usually should be same as the command_targets, but in query dependents mode, all targets should be loaded. blade_path: str, the path of the `blade` python module, used to be called by builtin tools. """ self.__command_targets = command_targets self.__load_targets = load_targets self.__blade_path = blade_path self.__working_dir = working_dir self.__build_dir = build_dir self.__root_dir = blade_root_dir self.__options = blade_options self.__command = command # Source dir of current loading BUILD file self.__current_source_path = blade_root_dir # The targets which are specified in command line explicitly, not pattern expanded. self.__direct_targets = [] # All command targets, includes direct targets and expanded target patterns. self.__expanded_command_targets = [] # Given some targets specified in the command line, Blade will load # BUILD files containing these command line targets; global target # functions, i.e., cc_library, cc_binary and etc, in these BUILD # files will register targets into target_database, which then becomes # the input to dependency analyzer and rules generator. It is # notable that not all targets in target_database are dependencies of # command line targets. self.__target_database = {} # The targets to be build after loading the build files. self.__build_targets = {} # The targets keys list after sorting by topological sorting method. # Used to generate build rules in correct order. self.__sorted_targets_keys = [] # The depended targets dict after topological sorting self.__depended_targets = {} # Indicate whether the deps list is expanded by expander or not self.__targets_expanded = False self.__build_time = time.time() self.__build_platform = BuildPlatform() self.build_environment = BuildEnvironment(self.__root_dir) self.__build_jobs_num = 0 self.__test_jobs_num = 0 self.svn_root_dirs = [] self._verify_history_path = os.path.join(build_dir, '.blade_verify.json') self._verify_history = { 'header_inclusion_dependencies': {}, # path(.H) -> mtime(modification time) } self.__build_script = os.path.join(self.__build_dir, 'build.ninja') def load_targets(self): """Load the targets. """ console.info('Loading BUILD files...') (self.__direct_targets, self.__expanded_command_targets, self.__build_targets) = load_targets(self.__load_targets, self.__root_dir, self) if self.__command_targets != self.__load_targets: # In query dependents mode, we must use command targets to execute query self.__expanded_command_targets = self._expand_command_targets() console.info('Loading done.') return self.__direct_targets, self.__expanded_command_targets # For test def _expand_command_targets(self): """Expand command line targets to targets list""" all_targets = self.__build_targets all_command_targets = [] for t in self.__command_targets: t_path, name = t.split(':') if name == '...': for tkey in all_targets: if tkey[0].startswith(t_path): all_command_targets.append(tkey) elif name == '*': for tkey in all_targets: if tkey[0] == t_path: all_command_targets.append(tkey) else: all_command_targets.append((t_path, name)) return all_command_targets def analyze_targets(self): """Expand the targets. """ console.info('Analyzing dependency graph...') (self.__sorted_targets_keys, self.__depended_targets) = analyze_deps(self.__build_targets) self.__targets_expanded = True console.info('Analyzing done.') return self.__build_targets # For test def new_build_rules_generator(self): return NinjaRulesGenerator(self.__build_script, self.__blade_path, self) def build_script(self): """Return build script file name""" return self.__build_script def generate_build_rules(self): """Generate the constructing rules. """ console.info('Generating build rules...') generator = self.new_build_rules_generator() rules = generator.generate_build_script() self.__all_rule_names = generator.get_all_rule_names() console.info('Generating done.') return rules def generate(self): """Generate the build script. """ if self.__command != 'query': self.generate_build_rules() def verify(self): """Verify specific targets after build is complete. """ verify_history = self._load_verify_history() header_inclusion_history = verify_history[ 'header_inclusion_dependencies'] error = 0 # Sorting helps reduce jumps between BUILD files when fixng reported problems for k in sorted(self.__expanded_command_targets): target = self.__build_targets[k] if target.type.startswith('cc_') and target.srcs: if not target.verify_hdr_dep_missing(header_inclusion_history): error += 1 self._dump_verify_history() return error == 0 def _load_verify_history(self): if os.path.exists(self._verify_history_path): with open(self._verify_history_path) as f: try: self._verify_history = json.load(f) except Exception as e: console.warning('Error loading %s, ignore. Reason: %s' % (self._verify_history_path, str(e))) pass return self._verify_history def _dump_verify_history(self): with open(self._verify_history_path, 'w') as f: json.dump(self._verify_history, f, indent=4) def run(self, target): """Run the target. """ runner = BinaryRunner(self.__options, self.__target_database, self.__build_targets) return runner.run_target(target) def test(self): """Run tests. """ exclude_tests = [] if self.__options.exclude_tests: exclude_tests = target.normalize( self.__options.exclude_tests.split(','), self.__working_dir) test_runner = TestRunner(self.__options, self.__target_database, self.__direct_targets, self.__expanded_command_targets, self.__build_targets, exclude_tests, self.test_jobs_num()) return test_runner.run() def query(self): """Query the targets. """ output_file_name = self.__options.output_file if output_file_name: output_file_name = os.path.join(self.__working_dir, output_file_name) output_file = open(output_file_name, 'w') console.info('Query result will be written to file "%s"' % self.__options.output_file) else: output_file = sys.stdout console.info('Query result:') output_format = self.__options.output_format if output_format == 'dot': self.query_dependency_dot(output_file) elif output_format == 'tree': self.query_dependency_tree(output_file) else: self.query_dependency_plain(output_file) if output_file_name: output_file.close() return 0 def query_dependency_plain(self, output_file): result_map = self.query_helper() if self.__options.deps: for key in result_map: print(file=output_file) deps = result_map[key][0] print('//%s:%s depends on the following targets:' % (key[0], key[1]), file=output_file) for d in deps: print('%s:%s' % (d[0], d[1]), file=output_file) if self.__options.dependents: for key in result_map: print(file=output_file) depended_by = result_map[key][1] print('//%s:%s is depended by the following targets:' % (key[0], key[1]), file=output_file) for d in depended_by: print('%s:%s' % (d[0], d[1]), file=output_file) def print_dot_node(self, output_file, node): print('"%s:%s" [label = "%s:%s"]' % (node[0], node[1], node[0], node[1]), file=output_file) def print_dot_deps(self, output_file, node, target_set): targets = self.__build_targets deps = targets[node].deps for i in deps: if not i in target_set: continue print('"%s:%s" -> "%s:%s"' % (node[0], node[1], i[0], i[1]), file=output_file) def __print_dot_graph(self, result_map, name, print_mode, output_file): # print_mode = 0: deps, 1: dependents targets = result_map.keys() nodes = set(targets) for key in targets: nodes |= set(result_map[key][print_mode]) print('digraph %s {' % name, file=output_file) for i in nodes: self.print_dot_node(output_file, i) for i in nodes: self.print_dot_deps(output_file, i, nodes) print('}', file=output_file) pass def query_dependency_dot(self, output_file): result_map = self.query_helper() if self.__options.deps: self.__print_dot_graph(result_map, 'deps', 0, output_file) if self.__options.dependents: self.__print_dot_graph(result_map, 'dependents', 1, output_file) def query_helper(self): """Query the targets helper method. """ all_targets = self.__build_targets query_list = self.__expanded_command_targets result_map = {} for key in query_list: deps = all_targets[key].expanded_deps # depended_by = [k for k in all_targets if key in all_targets[k].expanded_deps] depended_by = self.__depended_targets[key] result_map[key] = (sorted(deps), sorted(depended_by)) return result_map def query_dependency_tree(self, output_file): """Query the dependency tree of the specified targets. """ if self.__options.dependents: console.error_exit( 'Only query --deps can be output as tree format') print(file=output_file) for key in self.__expanded_command_targets: self._query_dependency_tree(key, 0, self.__build_targets, output_file) print(file=output_file) def _query_dependency_tree(self, key, level, build_targets, output_file): """Query the dependency tree of the specified target recursively. """ path, name = key if level == 0: output = '%s:%s' % (path, name) elif level == 1: output = '%s %s:%s' % ('+-', path, name) else: output = '%s%s %s:%s' % ('| ' * (level - 1), '+-', path, name) print(output, file=output_file) for dkey in build_targets[key].deps: self._query_dependency_tree(dkey, level + 1, build_targets, output_file) def dump_targets(self, output_file_name): result = [] with open(output_file_name, 'w') as f: for target_key in self.__expanded_command_targets: target = self.__target_database[target_key] result.append(target.dump()) json.dump(result, fp=f, indent=2) print(file=f) def get_build_time(self): return self.__build_time def get_build_dir(self): """The current building dir. """ return self.__build_dir def get_root_dir(self): """Return the blade root path. """ return self.__root_dir def get_command(self): """Get the blade command. """ return self.__command def set_current_source_path(self, current_source_path): """Set the current source path. """ self.__current_source_path = current_source_path def get_current_source_path(self): """Get the current source path. """ return self.__current_source_path def get_target_database(self): """Get the whole target database that haven't been expanded. """ return self.__target_database def get_direct_targets(self): """Return the direct targets. """ return self.__direct_targets def get_build_targets(self): """Get all the targets to be build. """ return self.__build_targets def get_depended_target_database(self): """Get depended target database that query dependent targets directly. """ return self.__depended_targets def get_options(self): """Get the global command options. """ return self.__options def is_expanded(self): """Whether the targets are expanded. """ return self.__targets_expanded def register_target(self, target): """Register a target into blade target database. It is used to do quick looking. """ key = target.key # Check whether there is already a key in database if key in self.__target_database: console.error_exit('Target %s is duplicate in //%s/BUILD' % (target.name, target.path)) self.__target_database[key] = target def _is_real_target_type(self, target_type): """The types that shouldn't be registered into blade manager. Sholdn't invoke ninja_rule method when it is not a real target which could not be registered into blade manager, like system library. 1. system_library """ return target_type != 'system_library' def gen_targets_rules(self): """Get the build rules and return to the object who queries this. """ rules_buf = [] skip_test = getattr(self.__options, 'no_test', False) skip_package = not getattr(self.__options, 'generate_package', False) for k in self.__sorted_targets_keys: target = self.__build_targets[k] if not self._is_real_target_type(target.type): continue blade_object = self.__target_database.get(k, None) if not blade_object: console.warning('"%s" is not a registered blade object' % str(k)) continue if skip_test and target.type.endswith( '_test') and k not in self.__direct_targets: continue if skip_package and target.type == 'package' and k not in self.__direct_targets: continue blade_object.ninja_rules() rules = blade_object.get_rules() if rules: rules_buf += rules return rules_buf def get_build_platform(self): """Return build platform instance. """ return self.__build_platform def get_sources_keyword_list(self): """This keywords list is used to check the source files path. Ex, when users specifies warning=no, it could be used to check that the source files is under thirdparty or not. If not, it will warn users that this flag is used incorrectly. """ keywords = ['thirdparty'] return keywords def _build_jobs_num(self): """Calculate build jobs num.""" # User has the highest priority jobs_num = config.get_item('global_config', 'build_jobs') if jobs_num > 0: return jobs_num # Calculate job numbers smartly distcc_enabled = config.get_item('distcc_config', 'enabled') if distcc_enabled and self.build_environment.distcc_env_prepared: # Distcc doesn't cost much local cpu, jobs can be quite large. distcc_num = len(self.build_environment.get_distcc_hosts_list()) jobs_num = min(max(int(1.5 * distcc_num), 1), 20) else: cpu_core_num = cpu_count() # machines with cpu_core_num > 8 is usually shared by multiple users, # set an upper bound to avoid interfering other users jobs_num = min(cpu_core_num, 8) console.info('Adjust build jobs number(-j N) to be %d' % jobs_num) return jobs_num def build_jobs_num(self): """The number of build jobs""" if self.__build_jobs_num == 0: self.__build_jobs_num = self._build_jobs_num() return self.__build_jobs_num def test_jobs_num(self): """Calculate the number of test jobs""" # User has the highest priority jobs_num = config.get_item('global_config', 'test_jobs') if jobs_num > 0: return jobs_num # In distcc enabled mode, the build_jobs_num may be quiet large, but we # only support run test locally, so the test_jobs_num should be limited # by local cpu mumber. # WE limit the test_jobs_num to be half of build job number because test # may be heavier than build (may be not, perhaps). build_jobs_num = self.build_jobs_num() cpu_core_num = cpu_count() jobs_num = max(min(build_jobs_num, cpu_core_num) / 2, 1) console.info('Adjust build jobs number(-j N) to be %d' % jobs_num) return jobs_num def get_all_rule_names(self): return self.__all_rule_names
class Blade(object): """Blade. A blade manager class. """ # pylint: disable=too-many-public-methods def __init__(self, command_targets, load_targets, blade_path, working_dir, build_path, blade_root_dir, blade_options, command): """init method. """ self.__command_targets = command_targets self.__load_targets = load_targets self.__blade_path = blade_path self.__working_dir = working_dir self.__build_path = build_path self.__root_dir = blade_root_dir self.__options = blade_options self.__command = command # Source dir of current loading BUILD file self.__current_source_path = blade_root_dir # The direct targets that are used for analyzing self.__direct_targets = [] # All command targets, make sure that all targets specified with ... # are all in the list now self.__all_command_targets = [] # Given some targets specified in the command line, Blade will load # BUILD files containing these command line targets; global target # functions, i.e., cc_library, cc_binary and etc, in these BUILD # files will register targets into target_database, which then becomes # the input to dependency analyzer and SCons rules generator. It is # notable that not all targets in target_database are dependencies of # command line targets. self.__target_database = {} # targets to build after loading the build files. self.__build_targets = {} # The targets keys list after sorting by topological sorting method. # Used to generate build rules in correct order. self.__sorted_targets_keys = [] # The depended targets dict after topological sorting self.__depended_targets = {} # Indicate whether the deps list is expanded by expander or not self.__targets_expanded = False self.__build_time = time.time() self.__build_platform = BuildPlatform() self.build_environment = BuildEnvironment(self.__root_dir) self.svn_root_dirs = [] self._verify_history_path = os.path.join(build_path, '.blade_verify.json') self._verify_history = { 'header_inclusion_dependencies': {}, # path(.H) -> mtime(modification time) } def load_targets(self): """Load the targets. """ console.info('loading BUILDs...') (self.__direct_targets, self.__all_command_targets, self.__build_targets) = load_targets(self.__load_targets, self.__root_dir, self) if self.__command_targets != self.__load_targets: # In query dependents mode, we must use command targets to execute query self.__all_command_targets = self._expand_command_targets() console.info('loading done.') return self.__direct_targets, self.__all_command_targets # For test def _expand_command_targets(self): """Expand command line targets to targets list""" all_targets = self.__build_targets all_command_targets = [] for t in self.__command_targets: t_path, name = t.split(':') if name == '...': for tkey in all_targets: if tkey[0].startswith(t_path): all_command_targets.append(tkey) elif name == '*': for tkey in all_targets: if tkey[0] == t_path: all_command_targets.append(tkey) else: all_command_targets.append((t_path, name)) return all_command_targets def analyze_targets(self): """Expand the targets. """ console.info('analyzing dependency graph...') (self.__sorted_targets_keys, self.__depended_targets) = analyze_deps(self.__build_targets) self.__targets_expanded = True console.info('analyzing done.') return self.__build_targets # For test def new_build_rules_generator(self): if config.get_item('global_config', 'native_builder') == 'ninja': return NinjaRulesGenerator('build.ninja', self.__blade_path, self) else: return SconsRulesGenerator('SConstruct', self.__blade_path, self) def generate_build_rules(self): """Generate the constructing rules. """ console.info('generating build rules...') generator = self.new_build_rules_generator() rules = generator.generate_build_script() self.__all_rule_names = generator.get_all_rule_names() console.info('generating done.') return rules def generate(self): """Generate the build script. """ if self.__command != 'query': self.generate_build_rules() def verify(self): """Verify specific targets after build is complete. """ verify_history = self._load_verify_history() error = 0 header_inclusion_dependencies = config.get_item('cc_config', 'header_inclusion_dependencies') header_inclusion_history = verify_history['header_inclusion_dependencies'] for k in self.__sorted_targets_keys: target = self.__build_targets[k] if (header_inclusion_dependencies and target.type == 'cc_library' and target.srcs): if not target.verify_header_inclusion_dependencies(header_inclusion_history): error += 1 self._dump_verify_history() return error == 0 def run(self, target): """Run the target. """ runner = BinaryRunner(self.__build_targets, self.__options, self.__target_database) return runner.run_target(target) def test(self): """Run tests. """ skip_tests = [] if self.__options.skip_tests: skip_tests = target.normalize(self.__options.skip_tests.split(','), self.__working_dir) test_runner = TestRunner(self.__build_targets, self.__options, self.__target_database, self.__direct_targets, skip_tests) return test_runner.run() def query(self): """Query the targets. """ output_file_name = self.__options.output_file if output_file_name: output_file_name = os.path.join(self.__working_dir, output_file_name) output_file = open(output_file_name, 'w') console.info('query result will be written to file "%s"' % self.__options.output_file) else: output_file = sys.stdout console.info('query result:') output_format = self.__options.output_format if output_format == 'dot': self.query_dependency_dot(output_file) elif output_format == 'tree': self.query_dependency_tree(output_file) else: self.query_dependency_plain(output_file) if output_file_name: output_file.close() return 0 def query_dependency_plain(self, output_file): result_map = self.query_helper() if self.__options.deps: for key in result_map: print(file=output_file) deps = result_map[key][0] print('//%s:%s depends on the following targets:' % (key[0], key[1]), file=output_file) for d in deps: print('%s:%s' % (d[0], d[1]), file=output_file) if self.__options.dependents: for key in result_map: print(file=output_file) depended_by = result_map[key][1] print('//%s:%s is depended by the following targets:' % (key[0], key[1]), file=output_file) for d in depended_by: print('%s:%s' % (d[0], d[1]), file=output_file) def print_dot_node(self, output_file, node): print('"%s:%s" [label = "%s:%s"]' % (node[0], node[1], node[0], node[1]), file=output_file) def print_dot_deps(self, output_file, node, target_set): targets = self.__build_targets deps = targets[node].deps for i in deps: if not i in target_set: continue print('"%s:%s" -> "%s:%s"' % (node[0], node[1], i[0], i[1]), file=output_file) def __print_dot_graph(self, result_map, name, print_mode, output_file): # print_mode = 0: deps, 1: dependents targets = result_map.keys() nodes = set(targets) for key in targets: nodes |= set(result_map[key][print_mode]) print('digraph %s {' % name, file=output_file) for i in nodes: self.print_dot_node(output_file, i) for i in nodes: self.print_dot_deps(output_file, i, nodes) print('}', file=output_file) pass def query_dependency_dot(self, output_file): result_map = self.query_helper() if self.__options.deps: self.__print_dot_graph(result_map, 'deps', 0, output_file) if self.__options.dependents: self.__print_dot_graph(result_map, 'dependents', 1, output_file) def query_helper(self): """Query the targets helper method. """ all_targets = self.__build_targets query_list = self.__all_command_targets result_map = {} for key in query_list: deps = all_targets[key].expanded_deps # depended_by = [k for k in all_targets if key in all_targets[k].expanded_deps] depended_by = self.__depended_targets[key] result_map[key] = (sorted(deps), sorted(depended_by)) return result_map def query_dependency_tree(self, output_file): """Query the dependency tree of the specified targets. """ if self.__options.dependents: console.error_exit('only query --deps can be output as tree format') print(file=output_file) for key in self.__all_command_targets: self._query_dependency_tree(key, 0, self.__build_targets, output_file) print(file=output_file) def _query_dependency_tree(self, key, level, build_targets, output_file): """Query the dependency tree of the specified target recursively. """ path, name = key if level == 0: output = '%s:%s' % (path, name) elif level == 1: output = '%s %s:%s' % ('+-', path, name) else: output = '%s%s %s:%s' % ('| ' * (level - 1), '+-', path, name) print(output, file=output_file) for dkey in build_targets[key].deps: self._query_dependency_tree(dkey, level + 1, build_targets, output_file) def dump_targets(self, output_file_name): result = [] with open(output_file_name, 'w') as f: for target_key in self.__all_command_targets: target = self.__target_database[target_key] result.append(target.dump()) json.dump(result, fp=f, indent=2) print(file=f) def get_build_time(self): return self.__build_time def get_build_path(self): """The current building path. """ return self.__build_path def get_root_dir(self): """Return the blade root path. """ return self.__root_dir def get_command(self): """Get the blade command. """ return self.__command def set_current_source_path(self, current_source_path): """Set the current source path. """ self.__current_source_path = current_source_path def get_current_source_path(self): """Get the current source path. """ return self.__current_source_path def get_target_database(self): """Get the whole target database that haven't been expanded. """ return self.__target_database def get_direct_targets(self): """Return the direct targets. """ return self.__direct_targets def get_build_targets(self): """Get all the targets to be build. """ return self.__build_targets def get_depended_target_database(self): """Get depended target database that query dependent targets directly. """ return self.__depended_targets def get_options(self): """Get the global command options. """ return self.__options def is_expanded(self): """Whether the targets are expanded. """ return self.__targets_expanded def register_target(self, target): """Register a target into blade target database. It is used to do quick looking. """ key = target.key # Check whether there is already a key in database if key in self.__target_database: console.error_exit('Target %s is duplicate in //%s/BUILD' % ( target.name, target.path)) self.__target_database[key] = target def _is_scons_object_type(self, target_type): """The types that shouldn't be registered into blade manager. Sholdn't invoke scons_rule method when it is not a scons target which could not be registered into blade manager, like system library. 1. system_library """ return target_type != 'system_library' def gen_targets_rules(self): """Get the build rules and return to the object who queries this. """ rules_buf = [] skip_test = getattr(self.__options, 'no_test', False) skip_package = not getattr(self.__options, 'generate_package', False) native_builder = config.get_item('global_config', 'native_builder') for k in self.__sorted_targets_keys: target = self.__build_targets[k] if not self._is_scons_object_type(target.type): continue blade_object = self.__target_database.get(k, None) if not blade_object: console.warning('not registered blade object, key %s' % str(k)) continue if (skip_test and target.type.endswith('_test') and k not in self.__direct_targets): continue if (skip_package and target.type == 'package' and k not in self.__direct_targets): continue if native_builder == 'ninja': blade_object.ninja_rules() else: blade_object.scons_rules() rules = blade_object.get_rules() if rules: rules_buf.append('\n') rules_buf += rules return rules_buf def get_scons_platform(self): """Return handle of the platform class. """ return self.__build_platform def get_sources_keyword_list(self): """This keywords list is used to check the source files path. Ex, when users specifies warning=no, it could be used to check that the source files is under thirdparty or not. If not, it will warn users that this flag is used incorrectly. """ keywords = ['thirdparty'] return keywords def _load_verify_history(self): if os.path.exists(self._verify_history_path): with open(self._verify_history_path) as f: self._verify_history = json.load(f) return self._verify_history def _dump_verify_history(self): with open(self._verify_history_path, 'w') as f: json.dump(self._verify_history, f) def parallel_jobs_num(self): """Tune the jobs num. """ # User has the highest priority user_jobs_num = self.__options.jobs if user_jobs_num > 0: return user_jobs_num # Calculate job numbers smartly distcc_enabled = config.get_item('distcc_config', 'enabled') if distcc_enabled and self.build_environment.distcc_env_prepared: # Distcc doesn't cost much local cpu, jobs can be quite large. distcc_num = len(self.build_environment.get_distcc_hosts_list()) jobs_num = min(max(int(1.5 * distcc_num), 1), 20) else: cpu_core_num = cpu_count() # machines with cpu_core_num > 4 is usually shared by multiple users, # set an upper bound to avoid interfering other users jobs_num = min(2 * cpu_core_num, 8) console.info('tunes the parallel jobs number(-j N) to be %d' % jobs_num) return jobs_num def get_all_rule_names(self): return self.__all_rule_names