306 lines
13 KiB
Python
306 lines
13 KiB
Python
#! /usr/bin/python3.8
|
|
"""
|
|
@file packet_content_parser.py
|
|
@brief Parses the Service Packet Definition files for all variables
|
|
@details Used by the Mib Exporter, inherits generic File Parser
|
|
"""
|
|
import re
|
|
|
|
from fsfwgen.parserbase.file_list_parser import FileListParser
|
|
from fsfwgen.parserbase.parser import FileParser
|
|
from fsfwgen.utility.csv_writer import CsvWriter
|
|
from fsfwgen.utility.printer import Printer
|
|
|
|
PACKET_CONTENT_DEFINITION_DESTINATION = ["../../mission/pus/servicepackets/",
|
|
"../../fsfw/pus/servicepackets/"]
|
|
PACKET_CONTENT_CSV_NAME = "mib_packet_data_content.csv"
|
|
PACKET_CONTENT_HEADER_COLUMN = ["Service", "Subservice", "Packet Name", "Datatype", "Name",
|
|
"Size [Bytes]", "Comment"]
|
|
|
|
SQL_DELETE_PACKET_DATA_CONTENT_CMD = """
|
|
DROP TABLE IF EXISTS PacketContent;
|
|
"""
|
|
|
|
SQL_CREATE_PACKET_DATA_CONTENT_CMD = """
|
|
CREATE TABLE IF NOT EXISTS PacketContent (
|
|
id INTEGER PRIMARY KEY,
|
|
service INTEGER,
|
|
subsvc INTEGER,
|
|
packetName TEXT ,
|
|
dataType TEXT,
|
|
name TEXT,
|
|
size INTEGER,
|
|
comment TEXT
|
|
)
|
|
"""
|
|
|
|
SQL_INSERT_PACKET_DATA_CMD = """
|
|
INSERT INTO PacketContent(service,subsvc,packetName,dataType,name,size,comment)
|
|
VALUES(?,?,?,?,?,?,?)
|
|
"""
|
|
|
|
|
|
def main():
|
|
print("PacketContentParser: Parsing for header files.")
|
|
header_file_parser = FileListParser(PACKET_CONTENT_DEFINITION_DESTINATION)
|
|
header_file_list = header_file_parser.parse_header_files(False, "Parsing packet data files: ")
|
|
packet_content_parser = PacketContentParser(header_file_list)
|
|
subservice_table = packet_content_parser.parse_files(True)
|
|
Printer.print_content(subservice_table, "PacketContentParser: Printing packet data table:")
|
|
subservice_writer = CsvWriter(PACKET_CONTENT_CSV_NAME,
|
|
subservice_table, PACKET_CONTENT_HEADER_COLUMN)
|
|
subservice_writer.write_to_csv()
|
|
subservice_writer.move_csv("..")
|
|
|
|
|
|
# noinspection PyTypeChecker
|
|
class PacketContentParser(FileParser):
|
|
# Initialize all needed columns
|
|
def __init__(self, file_list):
|
|
super().__init__(file_list)
|
|
self.serviceColumn = 0
|
|
self.subserviceColumn = 1
|
|
self.classNameColumn = 2
|
|
self.datatypeColumn = 3
|
|
self.nameColumn = 4
|
|
self.sizeColumn = 5
|
|
self.commentColumn = 6
|
|
self.lastEntryColumn = 7
|
|
self.columnListLength = 8
|
|
self.dictEntryList = list(range(self.columnListLength-1))
|
|
|
|
self.datatypeMatch = False
|
|
self.ignoreFlag = False
|
|
|
|
def _handle_file_parsing(self, file_name: str, *args: any):
|
|
self_print_parsing_info = False
|
|
if len(args) == 1 and isinstance(args[0], bool):
|
|
self_print_parsing_info = args[0]
|
|
|
|
# Read service from file name
|
|
self.dictEntryList[self.serviceColumn] = re.search('[0-9]{1,3}', file_name).group(0)
|
|
self.dictEntryList[self.subserviceColumn] = " "
|
|
file = open(file_name, "r")
|
|
if self_print_parsing_info:
|
|
print("Parsing " + file_name + " ...")
|
|
# Scans each line for possible variables
|
|
for line in file.readlines():
|
|
# Looks for class and struct definitions which mark a PUS packet
|
|
self.scan_for_class_and_struct_match_and_handle_it(line)
|
|
# Looks for variables
|
|
self.scan_for_variable_match_and_handle_it(line)
|
|
|
|
# Operation taken when file parsing is complete
|
|
# All packet content sizes are set by analysing the datatype
|
|
def _post_parsing_operation(self):
|
|
self.update_packet_content_sizes()
|
|
|
|
def scan_for_class_and_struct_match_and_handle_it(self, line):
|
|
class_or_struct_match = re.search('[\s]*class[\s]*([\w]*)[\s]*.*[\s]*{[\s]*([^\n]*)', line)
|
|
if not class_or_struct_match:
|
|
class_or_struct_match = re.search(
|
|
'[\s]*struct[\s]*([\w]*)[\s]*.*[\s]*{[\s]*([^\n]*)', line)
|
|
if class_or_struct_match:
|
|
self.dictEntryList[self.classNameColumn] = class_or_struct_match.group(1)
|
|
if class_or_struct_match.group(2):
|
|
self.dictEntryList[self.subserviceColumn] = \
|
|
self.check_for_subservice_string(class_or_struct_match.group(2))
|
|
|
|
def scan_for_variable_match_and_handle_it(self, line):
|
|
# Look for datatype definitions
|
|
var_match = self.packet_content_matcher(line)
|
|
if var_match:
|
|
# Attempts to find variable definition inside that packet
|
|
self.update_packet_content_table()
|
|
|
|
def packet_content_matcher(self, line):
|
|
# First step: Search for possible parameter definitions
|
|
# Generic serialize element or datatypes
|
|
var_match = re.search(
|
|
r'[\w]*(?:<)?[\s]*(uint32_t|uint8_t|uint16_t|ReturnValue_t|Mode_t|Submode_t|'
|
|
r'object_id_t|float|double|bool|ActionId_t|EventId_t|sid_t|ParameterId_t)'
|
|
r'(?:>)?[\s]*([\w]*)[\s]*(?:[= 0-9]*)?[;](?:[\/!< ]*([^\n]*))?', line)
|
|
if var_match:
|
|
# Debug printout
|
|
# print(var_match.group(0))
|
|
self.handle_generic_variable_match(var_match)
|
|
# Serial Fixed Array List with Size Header
|
|
else:
|
|
var_match = re.search(r'[ \w]*<SerialFixedArrayListAdapter<([\w_, ()]*)>>'
|
|
r'[\s]*([\w]*)[\s]*[;](?:[/!< ]*([^\n]*))?', line)
|
|
if var_match:
|
|
self.handle_serial_fixed_array_match(var_match)
|
|
# Serial Buffer, No length field
|
|
if not var_match:
|
|
var_match = re.search(r'[ \w]*<SerialBufferAdapter<([\w_,]*)>>'
|
|
r'[\s]*([\w]*)[\s]*[;](?:[/!< ]*([^\n]*))?', line)
|
|
if not var_match:
|
|
var_match = re.search(r'[\w ]*(?:<)?(uint32_t|uint8_t|uint16_t)[\s]*\*'
|
|
r'(?:>)?[\s]*([\w]*)[\s]*[;](?:[/!< ]*([^\n]*))?', line)
|
|
if var_match:
|
|
self.handle_serial_buffer_match(var_match)
|
|
# exclude size definition in serialize adapter or any definitions which are not parameter initializations
|
|
# or typedefs
|
|
if var_match and re.search("typedef", var_match.group(0)):
|
|
var_match = False
|
|
return var_match
|
|
|
|
def update_packet_content_table(self):
|
|
self.index = self.index + 1
|
|
dict_entry_tuple = tuple(self.dictEntryList[:self.columnListLength])
|
|
if not self.ignoreFlag:
|
|
self.mib_table.update({self.index: dict_entry_tuple})
|
|
else:
|
|
self.ignoreFlag = False
|
|
|
|
def handle_generic_variable_match(self, var_match):
|
|
self.handle_var_match(var_match)
|
|
self.handle_exporter_string(var_match.group(3))
|
|
|
|
def handle_serial_fixed_array_match(self, var_match):
|
|
if self.check_for_ignore_string(var_match.group(0)):
|
|
pass
|
|
else:
|
|
fixed_array_properties = re.search(
|
|
'([\w_]*)[\s]*,[\s]*([\w_()]*)[\s]*,[\s]*([\w_()]*)[\s]*', var_match.group(1))
|
|
if fixed_array_properties:
|
|
type_of_next_buffer_size = fixed_array_properties.group(3)
|
|
self.index = self.index + 1
|
|
self.dictEntryList[self.datatypeColumn] = type_of_next_buffer_size
|
|
self.dictEntryList[self.nameColumn] = "Size of following buffer"
|
|
dict_entry_tuple = tuple(self.dictEntryList[:self.columnListLength])
|
|
self.mib_table.update({self.index: dict_entry_tuple})
|
|
self.handle_var_match(var_match)
|
|
self.dictEntryList[self.datatypeColumn] = fixed_array_properties.group(1) + " *"
|
|
self.handle_exporter_string(var_match.group(3))
|
|
|
|
def handle_serial_buffer_match(self, var_match):
|
|
self.handle_var_match(var_match)
|
|
self.dictEntryList[self.datatypeColumn] = var_match.group(1) + " *"
|
|
self.dictEntryList[self.sizeColumn] = "deduced"
|
|
self.handle_exporter_string(var_match.group(3))
|
|
|
|
def handle_var_match(self, var_match):
|
|
self.dictEntryList[self.commentColumn] = ""
|
|
self.dictEntryList[self.sizeColumn] = ""
|
|
self.dictEntryList[self.datatypeColumn] = var_match.group(1)
|
|
self.dictEntryList[self.nameColumn] = var_match.group(2)
|
|
|
|
def update_packet_content_sizes(self):
|
|
self.dictEntryList[self.sizeColumn] = " "
|
|
for key, content in self.mib_table.items():
|
|
content = self.attempt_uint_match(content)
|
|
if not self.datatypeMatch:
|
|
content = self.attempt_eight_byte_match(content)
|
|
if not self.datatypeMatch:
|
|
content = self.attempt_four_byte_match(content)
|
|
if not self.datatypeMatch:
|
|
content = self.attempt_two_byte_match(content)
|
|
if not self.datatypeMatch:
|
|
content = self.attempt_one_byte_match(content)
|
|
content = self.handle_uint_buffer_type(content)
|
|
self.mib_table.update({key: content})
|
|
|
|
def attempt_uint_match(self, content):
|
|
self.datatypeMatch = re.search('uint([\d]{1,2})_t', content[self.datatypeColumn])
|
|
if self.datatypeMatch:
|
|
content = list(content)
|
|
content[self.sizeColumn] = round(int(self.datatypeMatch.group(1)) / 8)
|
|
content = tuple(content)
|
|
return content
|
|
|
|
def attempt_four_byte_match(self, content):
|
|
self.datatypeMatch = re.search(
|
|
r'object_id_t|ActionId_t|Mode_t|float|sid_t|ParameterId_t',
|
|
content[self.datatypeColumn])
|
|
if self.datatypeMatch:
|
|
content = list(content)
|
|
content[self.sizeColumn] = 4
|
|
content = tuple(content)
|
|
return content
|
|
|
|
def attempt_eight_byte_match(self, content):
|
|
self.datatypeMatch = re.search('double', content[self.datatypeColumn])
|
|
if self.datatypeMatch:
|
|
content = list(content)
|
|
content[self.sizeColumn] = 8
|
|
content = tuple(content)
|
|
return content
|
|
|
|
def attempt_two_byte_match(self, content):
|
|
self.datatypeMatch = re.search('ReturnValue_t|EventId_t', content[self.datatypeColumn])
|
|
if self.datatypeMatch:
|
|
content = list(content)
|
|
content[self.sizeColumn] = 2
|
|
content = tuple(content)
|
|
return content
|
|
|
|
def attempt_one_byte_match(self, content):
|
|
self.datatypeMatch = re.search('Submode_t|bool', content[self.datatypeColumn])
|
|
if self.datatypeMatch:
|
|
content = list(content)
|
|
content[self.sizeColumn] = 1
|
|
content = tuple(content)
|
|
return content
|
|
|
|
def handle_uint_buffer_type(self, content):
|
|
if re.search('\*', content[self.datatypeColumn]):
|
|
content = list(content)
|
|
content[self.sizeColumn] = "deduced"
|
|
content = tuple(content)
|
|
return content
|
|
|
|
# Used to scan exporter string for ignore flag or store any comments
|
|
def handle_exporter_string(self, match):
|
|
exporter_string = re.search('[ /!<]*\[EXPORT[\w]*\][\s]*:[\s]*([^\n]*)', match)
|
|
if exporter_string:
|
|
type_string = re.search("\[TYPE|BUFFERTYPE\][\s]*([\w]*)[^\n|\[]*", exporter_string.group(0),
|
|
re.IGNORECASE)
|
|
if type_string:
|
|
self.dictEntryList[self.datatypeColumn] = str(type_string.group(1)) + " *"
|
|
comment_string = re.search("\[COMMENT\][\s]*([\w]*)[^\n|\[]*", exporter_string.group(0),
|
|
re.IGNORECASE)
|
|
if comment_string:
|
|
self.dictEntryList[self.commentColumn] = comment_string.group(1)
|
|
self.check_for_ignore_string(exporter_string.group(0))
|
|
if not comment_string:
|
|
self.dictEntryList[self.commentColumn] = exporter_string.group(1)
|
|
|
|
# Used to transform comma separated subservice numbers into specific subservice numbers
|
|
def check_for_subservice_string(self, full_description):
|
|
subservice_info = re.search(
|
|
r'^.*//[\s]*[!<]*[\s]*\[EXPORT[\w]*\][\s]*:[\s]*\[SUBSERVICE\][\s]*([^\n]*)',
|
|
full_description, re.IGNORECASE)
|
|
description = ' '
|
|
if subservice_info:
|
|
description = self.handle_subservice_string(subservice_info)
|
|
if full_description == '':
|
|
description = ' '
|
|
return description
|
|
|
|
def check_for_ignore_string(self, string):
|
|
ignore_string = re.search("IGNORE", string, re.IGNORECASE)
|
|
if ignore_string:
|
|
self.ignoreFlag = True
|
|
return True
|
|
|
|
@staticmethod
|
|
def handle_subservice_string(subservice_info):
|
|
description = ' '
|
|
subservice_list = [int(x) for x in subservice_info.group(1).split(',')]
|
|
subservice_number = len(subservice_list)
|
|
for i in range(subservice_number):
|
|
description = description + str(subservice_list[i])
|
|
if i == subservice_number - 2:
|
|
description = description + " and "
|
|
elif i < subservice_number - 1:
|
|
description = description + ", "
|
|
return description
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|
|
|
|
|
|
|