X-Git-Url: https://codewiz.org/gitweb?a=blobdiff_plain;f=wizard%2Fbertos_utils.py;h=925774ba387fefe0761b66364763fb40c9a8bc9d;hb=9d4807e10b3ffccbee160aab1b67b13779cf6766;hp=0f0dc709bcd5e4041509fedd0a7d8a258479769b;hpb=89d8669f1af6548342b6137479da16fdc71917ba;p=bertos.git diff --git a/wizard/bertos_utils.py b/wizard/bertos_utils.py index 0f0dc709..925774ba 100644 --- a/wizard/bertos_utils.py +++ b/wizard/bertos_utils.py @@ -11,6 +11,12 @@ import os import fnmatch +import glob +import re +import shutil + +import const +import DefineException def isBertosDir(directory): return os.path.exists(directory + "/VERSION") @@ -18,26 +24,317 @@ def isBertosDir(directory): def bertosVersion(directory): return open(directory + "/VERSION").readline().strip() -def createBertosProject(directory): +def createBertosProject(projectInfos): + directory = projectInfos.info("PROJECT_PATH") + sourcesDir = projectInfos.info("SOURCES_PATH") if not os.path.isdir(directory): os.mkdir(directory) - open(directory + "/project.bertos", "w") + f = open(directory + "/project.bertos", "w") + f.write(repr(projectInfos)) + f.close() + ## Destination source dir + srcdir = directory + "/bertos" + shutil.rmtree(srcdir, True) + shutil.copytree(sourcesDir + "/bertos", srcdir) + ## Destination makefile + makefile = directory + "/Makefile" + if os.path.exists(makefile): + os.remove(makefile) + makefile = open("mktemplates/Makefile").read() + makefile = makefileGenerator(projectInfos, makefile) + open(directory + "/Makefile", "w").write(makefile) + ## Destination project dir + prjdir = directory + "/" + os.path.basename(directory) + shutil.rmtree(prjdir, True) + os.mkdir(prjdir) + ## Destination configurations files + cfgdir = prjdir + "/cfg" + shutil.rmtree(cfgdir, True) + os.mkdir(cfgdir) + for key, value in projectInfos.info("CONFIGURATIONS").items(): + string = open(sourcesDir + "/" + key, "r").read() + for parameter, infos in value.items(): + value = infos["value"] + if "unsigned" in infos["informations"].keys() and infos["informations"]["unsigned"]: + value += "U" + if "long" in infos["informations"].keys() and infos["informations"]["long"]: + value += "L" + string = sub(string, parameter, value) + f = open(cfgdir + "/" + os.path.basename(key), "w") + f.write(string) + f.close() + ## Destinatio mk file + makefile = open("mktemplates/template.mk", "r").read() + makefile = mkGenerator(projectInfos, makefile) + open(prjdir + "/" + os.path.basename(prjdir) + ".mk", "w").write(makefile) + +def mkGenerator(projectInfos, makefile): + """ + Generates the mk file for the current project. + """ + mkData = {} + mkData["pname"] = os.path.basename(projectInfos.info("PROJECT_PATH")) + mkData["cpuname"] = projectInfos.info("CPU_INFOS")["CPU_NAME"] + mkData["cflags"] = " ".join(projectInfos.info("CPU_INFOS")["C_FLAGS"]) + mkData["ldflags"] = " ".join(projectInfos.info("CPU_INFOS")["LD_FLAGS"]) + for key in mkData: + while makefile.find(key) != -1: + makefile = makefile.replace(key, mkData[key]) + return makefile + +def makefileGenerator(projectInfos, makefile): + """ + Generate the Makefile for the current project. + """ + # TODO: write a general function that works for both the mk file and the Makefile + while makefile.find("project_name") != -1: + makefile = makefile.replace("project_name", os.path.basename(projectInfos.info("PROJECT_PATH"))) + return makefile + +def getSystemPath(): + path = os.environ["PATH"] + if os.name == "nt": + path = path.split(";") + else: + path = path.split(":") + return path + +def findToolchains(pathList): + toolchains = [] + for element in pathList: + for toolchain in glob.glob(element+ "/" + const.GCC_NAME): + if not os.path.islink(toolchain): + toolchains.append(toolchain) + return list(set(toolchains)) + +def getToolchainInfo(output): + info = {} + expr = re.compile("Target: .*") + target = expr.findall(output) + if len(target) == 1: + info["target"] = target[0].split("Target: ")[1] + expr = re.compile("gcc version [0-9,.]*") + version = expr.findall(output) + if len(version) == 1: + info["version"] = version[0].split("gcc version ")[1] + expr = re.compile("gcc version [0-9,.]* \(.*\)") + build = expr.findall(output) + if len(build) == 1: + build = build[0].split("gcc version ")[1] + build = build[build.find("(") + 1 : build.find(")")] + info["build"] = build + expr = re.compile("Configured with: .*") + configured = expr.findall(output) + if len(configured) == 1: + info["configured"] = configured[0].split("Configured with: ")[1] + expr = re.compile("Thread model: .*") + thread = expr.findall(output) + if len(thread) == 1: + info["thread"] = thread[0].split("Thread model: ")[1] + return info def findDefinitions(ftype, path): L = os.walk(path) for element in L: for filename in element[2]: - if fnmatch.fnmatch(filename, "*." + ftype): + if fnmatch.fnmatch(filename, ftype): yield (filename, element[0]) def loadCpuInfos(path): cpuInfos = [] - for definition in findDefinitions("cdef", path): - D = {} - def include(filename, dict = D, directory=definition[1]): - execfile(directory + "/" + filename, {}, D) - D["include"] = include - include(definition[0], D) - D["CPU_NAME"] = definition[0].split(".")[0] - cpuInfos.append(D) + for definition in findDefinitions(const.CPU_DEFINITION, path): + cpuInfos.append(getInfos(definition)) return cpuInfos + +def getInfos(definition): + D = {} + D.update(const.CPU_DEF) + def include(filename, dict = D, directory=definition[1]): + execfile(directory + "/" + filename, {}, D) + D["include"] = include + include(definition[0], D) + D["CPU_NAME"] = definition[0].split(".")[0] + D["DEFINITION_PATH"] = definition[1] + "/" + definition[0] + del D["include"] + return D + +def getDefinitionBlocks(text): + """ + Take a text and return a list of tuple (description, name-value). + """ + block = [] + block_tmp = re.findall(r"/\*{2}\s*([^*]*\*(?:[^/*][^*]*\*+)*)/\s*#define\s+((?:[^/]*?/?)+)\s*?(?:/{2,3}[^<].*?)?$", text, re.MULTILINE) + for comment, define in block_tmp: + block.append((" ".join(re.findall(r"^\s*\*?\s*(.*?)\s*?(?:/{2}.*?)?$", comment, re.MULTILINE)).strip(), define)) + block += re.findall(r"/{3}\s*([^<].*?)\s*#define\s+((?:[^/]*?/?)+)\s*?(?:/{2,3}[^<].*?)?$", text, re.MULTILINE) + block += [(comment, define) for define, comment in re.findall(r"#define\s*(.*?)\s*/{3}<\s*(.+?)\s*?(?:/{2,3}[^<].*?)?$", text, re.MULTILINE)] + return block + +def formatParamNameValue(text): + """ + Take the given string and return a tuple with the name of the parameter in the first position + and the value in the second. + """ + block = re.findall("\s*([^\s]+)\s*(.+?)\s*$", text, re.MULTILINE) + return block[0] + +def getDescriptionInformations(text): + """ + Take the doxygen comment and strip the wizard informations, returning the tuple + (comment, wizard_informations) + """ + index = text.find("$WIZARD") + if index != -1: + exec(text[index + 1:]) + informations = WIZARD + return text[:index].strip(), informations + else: + return text.strip(), {} + +def loadConfigurationInfos(path): + """ + Return the module configurations found in the given file as a dict with the + parameter name as key and a dict containig the fields above as value: + "value": the value of the parameter + "description": the description of the parameter + "informations": a dict containig optional informations: + "type": "int" | "boolean" | "enum" + "min": the minimum value for integer parameters + "max": the maximum value for integer parameters + "long": boolean indicating if the num is a long + "value_list": the name of the enum for enum parameters + """ + try: + configurationInfos = {} + for comment, define in getDefinitionBlocks(open(path, "r").read()): + name, value = formatParamNameValue(define) + description, informations = getDescriptionInformations(comment) + configurationInfos[name] = {} + configurationInfos[name]["value"] = value + configurationInfos[name]["informations"] = informations + if ("type" in configurationInfos[name]["informations"].keys() and + configurationInfos[name]["informations"]["type"] == "int" and + configurationInfos[name]["value"].find("L") != -1): + configurationInfos[name]["informations"]["long"] = True + configurationInfos[name]["value"] = configurationInfos[name]["value"].replace("L", "") + if ("type" in configurationInfos[name]["informations"].keys() and + configurationInfos[name]["informations"]["type"] == "int" and + configurationInfos[name]["value"].find("U") != -1): + configurationInfos[name]["informations"]["unsigned"] = True + configurationInfos[name]["value"] = configurationInfos[name]["value"].replace("U", "") + configurationInfos[name]["description"] = description + return configurationInfos + except SyntaxError: + raise DefineException.ConfigurationDefineException(path) + +def loadModuleInfos(path): + """ + Return the module infos found in the given file as a dict with the module + name as key and a dict containig the fields above as value or an empty dict + if the given file is not a BeRTOS module: + "depends": a list of modules needed by this module + "configuration": the cfg_*.h with the module configurations + "description": a string containing the brief description of doxygen + "enabled": contains False but the wizard will change if the user select + the module + """ + try: + moduleInfos = {} + string = open(path, "r").read() + commentList = re.findall(r"/\*{2}\s*([^*]*\*(?:[^/*][^*]*\*+)*)/", string) + commentList = [" ".join(re.findall(r"^\s*\*?\s*(.*?)\s*?(?:/{2}.*?)?$", comment, re.MULTILINE)).strip() for comment in commentList] + for comment in commentList: + index = comment.find("$WIZARD_MODULE") + if index != -1: + exec(comment[index + 1:]) + moduleInfos[WIZARD_MODULE["name"]] = {"depends": WIZARD_MODULE["depends"], + "configuration": WIZARD_MODULE["configuration"], + "description": "", + "enabled": False} + index = comment.find("\\brief") + if index != -1: + description = comment[index + 7:] + description = description[:description.find(" * ")] + moduleInfos[WIZARD_MODULE["name"]]["description"] = description + return moduleInfos + return {} + except SyntaxError: + raise DefineException.ModuleDefineException(path) + +def loadModuleInfosDict(path): + """ + Return the dict containig all the modules + """ + moduleInfosDict = {} + for filename, path in findDefinitions("*.h", path): + moduleInfosDict.update(loadModuleInfos(path + "/" + filename)) + return moduleInfosDict + +def loadDefineLists(path): + """ + Return a dict with the name of the list as key and a list of string as value + """ + try: + string = open(path, "r").read() + commentList = re.findall(r"/\*{2}\s*([^*]*\*(?:[^/*][^*]*\*+)*)/", string) + commentList = [" ".join(re.findall(r"^\s*\*?\s*(.*?)\s*?(?:/{2}.*?)?$", comment, re.MULTILINE)).strip() for comment in commentList] + listDict = {} + for comment in commentList: + index = comment.find("$WIZARD_LIST") + if index != -1: + exec(comment[index + 1:]) + listDict.update(WIZARD_LIST) + return listDict + except SyntaxError: + raise DefineException.EnumDefineException(path) + +def loadDefineListsDict(path): + """ + Return the dict containing all the define lists + """ + defineListsDict = {} + for filename, path in findDefinitions("*.h", path): + defineListsDict.update(loadDefineLists(path + "/" + filename)) + return defineListsDict + +def sub(string, parameter, value): + """ + Substitute the given value at the given parameter define in the given string + """ + return re.sub(r"(?P#define\s+" + parameter + r"\s+)([^\s]+)", r"\g" + value, string) + +def isInt(informations): + """ + Return True if the value is a simple int. + """ + if ("long" not in informatios.keys() or not informations["long"]) and ("unsigned" not in informations.keys() or informations["unsigned"]): + return True + else: + return False + +def isLong(informations): + """ + Return True if the value is a long. + """ + if "long" in informations.keys() and informations["long"] and "unsigned" not in informations.keys(): + return True + else: + return False + +def isUnsigned(informations): + """ + Return True if the value is an unsigned. + """ + if "unsigned" in informations.keys() and informations["unsigned"] and "long" not in informations.keys(): + return True + else: + return False + +def isUnsignedLong(informations): + """ + Return True if the value is an unsigned long. + """ + if "unsigned" in informations.keys() and "long" in informations.keys() and informations["unsigned"] and informations["long"]: + return True + else: + return False \ No newline at end of file