esp-idf/tools/mass_mfg/mfg_gen.py
Djordje Nedic b654c14d0e tools: Introduce support for blank lines in config and value files for mfg utility
Some users have requested this feature.

In order to avoid complete refactoring or introducing more code complexity, a design choice to create temporary files without blank lines is made.

Additionally, an extension check is added and there are multiple smaller code style and structure improvements.

Closes https://github.com/espressif/esp-idf/issues/8421
2022-07-26 23:17:33 +02:00

467 lines
19 KiB
Python

#!/usr/bin/env python
#
# SPDX-FileCopyrightText: 2018-2022 Espressif Systems (Shanghai) CO LTD
# SPDX-License-Identifier: Apache-2.0
#
from __future__ import print_function
import argparse
import csv
import distutils.dir_util
import os
import sys
from future.moves.itertools import zip_longest
try:
idf_path = os.environ['IDF_PATH']
sys.path.insert(0, idf_path + '/components/nvs_flash/nvs_partition_generator/')
import nvs_partition_gen
except Exception as e:
print(e)
sys.exit('Please check IDF_PATH')
def create_temp_files(args):
new_filenames = []
for filename in [args.conf, args.values]:
name, ext = os.path.splitext(filename)
new_filename = name + '_tmp' + ext
strip_blank_lines(filename, new_filename)
new_filenames.append(new_filename)
return new_filenames
def strip_blank_lines(input_filename, output_filename):
with open(input_filename, 'r') as read_from, open(output_filename,'w', newline='') as write_to:
for line in read_from:
if not line.isspace():
write_to.write(line)
def verify_values_exist(input_values_file, keys_in_values_file):
""" Verify all keys have corresponding values in values file
"""
with open(input_values_file, 'r') as values_file:
values_file_reader = csv.reader(values_file, delimiter=',')
next(values_file_reader)
for line_num, line in enumerate(values_file_reader, start=2):
key_count_in_values_file = len(keys_in_values_file)
if len(line) != key_count_in_values_file:
raise SystemExit('\nError: Number of values is not equal to number of keys in file: %s at line No:%s\n'
% (str(input_values_file), str(line_num)))
def verify_keys_exist(values_file_keys, input_config_file):
""" Verify all keys from config file are present in values file
"""
keys_missing = []
with open(input_config_file,'r') as config_file:
config_file_reader = csv.reader(config_file, delimiter=',')
for line_num, line in enumerate(config_file_reader, start=1):
if 'namespace' not in line:
if values_file_keys:
if line[0] == values_file_keys[0]:
del values_file_keys[0]
else:
keys_missing.append([line_num, line[0]])
else:
keys_missing.append([line_num, line[0]])
if keys_missing:
for line_num, key in keys_missing:
print('Key:`', str(key), '` at line no:', str(line_num),
' in config file is not found in values file.')
raise SystemExit(1)
def verify_datatype_encoding(input_config_file):
""" Verify datatype and encodings from config file is valid
"""
valid_encodings = {'string', 'binary', 'hex2bin','u8', 'i8', 'u16', 'u32', 'i32','base64'}
valid_datatypes = {'file','data','namespace'}
with open(input_config_file,'r') as config_file:
config_file_reader = csv.reader(config_file, delimiter=',')
for line_num, line in enumerate(config_file_reader, start=1):
if line[1] not in valid_datatypes:
raise SystemExit('Error: config file: %s has invalid datatype at line no:%s\n`'
% (str(input_config_file), str(line_num)))
if 'namespace' not in line and line[2] not in valid_encodings:
raise SystemExit('Error: config file: %s has invalid encoding at line no:%s\n`'
% (str(input_config_file), str(line_num)))
def verify_file_data_count(input_config_file, keys_repeat):
""" Verify count of data on each line in config file is equal to 3
(as format must be: <key,type and encoding>)
"""
with open(input_config_file, 'r') as config_file:
config_file_reader = csv.reader(config_file, delimiter=',')
for line_num, line in enumerate(config_file_reader, start=1):
if len(line) != 3 and line[0] not in keys_repeat:
raise SystemExit('Error: data missing in config file at line no:%s <format needed:key,type,encoding>\n'
% str(line_num))
def verify_data_in_file(input_config_file, input_values_file, config_file_keys, keys_in_values_file, keys_repeat):
""" Verify count of data on each line in config file is equal to 3 \
(as format must be: <key,type and encoding>)
Verify datatype and encodings from config file is valid
Verify all keys from config file are present in values file and \
Verify each key has corresponding value in values file
"""
try:
verify_file_data_count(input_config_file, keys_repeat)
verify_datatype_encoding(input_config_file)
# Get keys from values file present in config files
values_file_keys = get_keys(keys_in_values_file, config_file_keys)
verify_keys_exist(values_file_keys, input_config_file)
verify_values_exist(input_values_file, keys_in_values_file)
except Exception as err:
print(err)
exit(1)
def get_keys(keys_in_values_file, config_file_keys):
""" Get keys from values file present in config file
"""
values_file_keys = [key for key in keys_in_values_file if key in config_file_keys]
return values_file_keys
def add_config_data_per_namespace(input_config_file):
""" Add config data per namespace to `config_data_to_write` list
"""
config_data_to_write = []
config_data_per_namespace = []
with open(input_config_file,'r') as csv_config_file:
config_file_reader = csv.reader(csv_config_file, delimiter=',')
# `config_data_per_namespace` is added to `config_data_to_write` list after reading next namespace
for config_data in config_file_reader:
if 'REPEAT' in config_data:
config_data.remove('REPEAT')
if 'namespace' in config_data:
if config_data_per_namespace:
config_data_to_write.append(config_data_per_namespace)
config_data_per_namespace = []
config_data_per_namespace.append(config_data)
else:
config_data_per_namespace.append(config_data)
else:
config_data_per_namespace.append(config_data)
# `config_data_per_namespace` is added to `config_data_to_write` list as EOF is reached
if (not config_data_to_write) or (config_data_to_write and config_data_per_namespace):
config_data_to_write.append(config_data_per_namespace)
return config_data_to_write
def get_fileid_val(file_identifier, key_value_data, fileid_value):
""" Get file identifier value
"""
file_id_found = False
for key in key_value_data:
if file_identifier and not file_id_found and file_identifier in key:
fileid_value = key[1]
file_id_found = True
if not file_id_found:
fileid_value = str(int(fileid_value) + 1)
return fileid_value
def add_data_to_file(config_data_to_write, key_value_pair, output_csv_file):
""" Add data to csv target file
"""
header = ['key', 'type', 'encoding', 'value']
data_to_write = []
with open(output_csv_file, 'w', newline='') as target_csv_file:
output_file_writer = csv.writer(target_csv_file, delimiter=',')
output_file_writer.writerow(header)
for namespace_config_data in config_data_to_write:
for data in namespace_config_data:
data_to_write = data[:]
if 'namespace' in data:
data_to_write.append('')
output_file_writer.writerow(data_to_write)
else:
key = data[0]
while key not in key_value_pair[0]:
del key_value_pair[0]
if key in key_value_pair[0]:
value = key_value_pair[0][1]
data_to_write.append(value)
del key_value_pair[0]
output_file_writer.writerow(data_to_write)
def create_dir(filetype, output_dir_path):
""" Create new directory(if doesn't exist) to store file generated
"""
output_target_dir = os.path.join(output_dir_path,filetype,'')
if not os.path.isdir(output_target_dir):
distutils.dir_util.mkpath(output_target_dir)
return output_target_dir
def set_repeat_value(total_keys_repeat, keys, csv_file, target_filename):
with open(csv_file, 'r') as read_from, open(target_filename,'w', newline='') as write_to:
csv_file_reader = csv.reader(read_from, delimiter=',')
headers = next(csv_file_reader)
values = next(csv_file_reader)
csv_file_writer = csv.writer(write_to, delimiter=',')
csv_file_writer.writerow(headers)
csv_file_writer.writerow(values)
# read new data, add value if key has repeat tag, write to new file
for row in csv_file_reader:
key_val_new = list(zip_longest(keys, row))
key_val_pair = list(zip_longest(keys, values))
key_repeated = total_keys_repeat[:]
index = 0
while key_val_new and key_repeated:
# if key has repeat tag, get its corresponding value, write to file
if key_val_new[0][0] == key_repeated[0]:
row[index] = key_val_pair[0][1]
del key_repeated[0]
del key_val_new[0]
del key_val_pair[0]
index += 1
csv_file_writer.writerow(row)
return target_filename
def create_intermediate_csv(args, keys_in_values_file, keys_repeat, is_encr=False):
file_identifier_value = '0'
# Add config data per namespace to `config_data_to_write` list
config_data_to_write = add_config_data_per_namespace(args.conf)
try:
with open(args.values, 'r') as csv_values_file:
values_file_reader = csv.reader(csv_values_file, delimiter=',')
keys = next(values_file_reader)
filename, file_ext = os.path.splitext(args.values)
target_filename = filename + '_created' + file_ext
if keys_repeat:
target_values_file = set_repeat_value(keys_repeat, keys, args.values, target_filename)
else:
target_values_file = args.values
with open(target_values_file, 'r') as csv_values_file:
values_file_reader = csv.reader(csv_values_file, delimiter=',')
next(values_file_reader)
# Create new directory(if doesn't exist) to store csv file generated
output_csv_target_dir = create_dir('csv', args.outdir)
# Create new directory(if doesn't exist) to store bin file generated
output_bin_target_dir = create_dir('bin', args.outdir)
for values_data_line in values_file_reader:
key_value_data = list(zip_longest(keys_in_values_file, values_data_line))
# Get file identifier value from values file
file_identifier_value = get_fileid_val(args.fileid, key_value_data, file_identifier_value)
key_value_pair = key_value_data[:]
# Verify if output csv file does not exist
csv_filename = args.prefix + '-' + file_identifier_value + '.' + 'csv'
output_csv_file = output_csv_target_dir + csv_filename
if os.path.isfile(output_csv_file):
raise SystemExit('Target csv file: %s already exists.`' % output_csv_file)
# Add values corresponding to each key to csv intermediate file
add_data_to_file(config_data_to_write, key_value_pair, output_csv_file)
print('\nCreated CSV file: ===>', output_csv_file)
# Verify if output bin file does not exist
bin_filename = args.prefix + '-' + file_identifier_value + '.' + 'bin'
output_bin_file = output_bin_target_dir + bin_filename
if os.path.isfile(output_bin_file):
raise SystemExit('Target binary file: %s already exists.`' % output_bin_file)
args.input = output_csv_file
args.output = os.path.join('bin', bin_filename)
if args.keygen:
args.keyfile = 'keys-' + args.prefix + '-' + file_identifier_value
if is_encr:
nvs_partition_gen.encrypt(args)
else:
nvs_partition_gen.generate(args)
print('\nFiles generated in %s ...' % args.outdir)
except Exception as e:
print(e)
exit(1)
def verify_file_format(args):
keys_in_config_file = []
keys_in_values_file = []
keys_repeat = []
# Verify files have csv extension
conf_name, conf_extension = os.path.splitext(args.conf)
if conf_extension != '.csv':
raise SystemExit('Error: config file: %s does not have the .csv extension.' % args.conf)
values_name, values_extension = os.path.splitext(args.values)
if values_extension != '.csv':
raise SystemExit('Error: values file: %s does not have the .csv extension.' % args.values)
# Verify files are not empty
if os.stat(args.conf).st_size == 0:
raise SystemExit('Error: config file: %s is empty.' % args.conf)
if os.stat(args.values).st_size == 0:
raise SystemExit('Error: values file: %s is empty.' % args.values)
# Extract keys from config file
with open(args.conf, 'r') as config_file:
config_file_reader = csv.reader(config_file, delimiter=',')
for config_data in config_file_reader:
if 'namespace' not in config_data:
keys_in_config_file.append(config_data[0])
if 'REPEAT' in config_data:
keys_repeat.append(config_data[0])
# Extract keys from values file
with open(args.values, 'r') as values_file:
values_file_reader = csv.reader(values_file, delimiter=',')
keys_in_values_file = next(values_file_reader)
# Verify file identifier exists in values file
if args.fileid:
if args.fileid not in keys_in_values_file:
raise SystemExit('Error: target_file_identifier: %s does not exist in values file.\n' % args.fileid)
else:
args.fileid = 1
return keys_in_config_file, keys_in_values_file, keys_repeat
def generate(args):
# Create work files with no blank lines
args.conf, args.values = create_temp_files(args)
# Verify input config and values file format
keys_in_config_file, keys_in_values_file, keys_repeat = verify_file_format(args)
# Verify data in the input_config_file and input_values_file
verify_data_in_file(args.conf, args.values, keys_in_config_file,
keys_in_values_file, keys_repeat)
encryption_enabled = False
if (args.keygen or args.inputkey):
encryption_enabled = True
print('\nGenerating encrypted NVS binary images...')
# Create intermediate csv file
create_intermediate_csv(args, keys_in_values_file, keys_repeat, is_encr=encryption_enabled)
def generate_key(args):
nvs_partition_gen.generate_key(args)
def main():
try:
parser = argparse.ArgumentParser(description='\nESP Manufacturing Utility', formatter_class=argparse.RawTextHelpFormatter)
subparser = parser.add_subparsers(title='Commands',
dest='command',
help='\nRun mfg_gen.py {command} -h for additional help\n\n')
parser_gen = subparser.add_parser('generate',
help='Generate NVS partition',
formatter_class=argparse.RawTextHelpFormatter)
parser_gen.set_defaults(func=generate)
parser_gen.add_argument('conf',
default=None,
help='Path to configuration csv file to parse')
parser_gen.add_argument('values',
default=None,
help='Path to values csv file to parse')
parser_gen.add_argument('prefix',
default=None,
help='Unique name for each output filename prefix')
parser_gen.add_argument('size',
default=None,
help='Size of NVS partition in bytes\
\n(must be multiple of 4096)')
parser_gen.add_argument('--fileid',
default=None,
help='''Unique file identifier(any key in values file) \
\nfor each filename suffix (Default: numeric value(1,2,3...)''')
parser_gen.add_argument('--version',
choices=[1, 2],
default=2,
type=int,
help='''Set multipage blob version.\
\nVersion 1 - Multipage blob support disabled.\
\nVersion 2 - Multipage blob support enabled.\
\nDefault: Version 2 ''')
parser_gen.add_argument('--keygen',
action='store_true',
default=False,
help='Generates key for encrypting NVS partition')
parser_gen.add_argument('--keyfile',
default=None,
help=argparse.SUPPRESS)
parser_gen.add_argument('--inputkey',
default=None,
help='File having key for encrypting NVS partition')
parser_gen.add_argument('--outdir',
default=os.getcwd(),
help='Output directory to store files created\
\n(Default: current directory)')
parser_gen.add_argument('--input',
default=None,
help=argparse.SUPPRESS)
parser_gen.add_argument('--output',
default=None,
help=argparse.SUPPRESS)
parser_gen_key = subparser.add_parser('generate-key',
help='Generate keys for encryption',
formatter_class=argparse.RawTextHelpFormatter)
parser_gen_key.set_defaults(func=generate_key)
parser_gen_key.add_argument('--keyfile',
default=None,
help='Path to output encryption keys file')
parser_gen_key.add_argument('--outdir',
default=os.getcwd(),
help='Output directory to store files created.\
\n(Default: current directory)')
args = parser.parse_args()
args.func(args)
except ValueError as err:
print(err)
except Exception as e:
print(e)
if __name__ == '__main__':
main()