192 lines
7.1 KiB
Executable File

# Copyright 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018 Max H. Parke KA1RBI
# This file is part of OP25
# OP25 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; either version 3, or (at your option)
# any later version.
# OP25 is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
# License for more details.
# You should have received a copy of the GNU General Public License
# along with OP25; see the file COPYING. If not, write to the Free
# Software Foundation, Inc., 51 Franklin Street, Boston, MA
# 02110-1301, USA.
import sys
import os
import csv
def get_frequency(f): # return frequency in Hz
if f.find('.') == -1: # assume in Hz
return int(f)
else: # assume in MHz due to '.'
return int(float(f) * 1000000)
def get_int_dict(s):
# parameter string s is the file name to be read,
# except when the first character of s is a digit,
# in which case s itself is the comma-separated list of ints
if s[0].isdigit():
return dict.fromkeys([int(d) for d in s.split(',')])
# create dict by reading from file
d = {} # this is the dict
with open(s,"r") as f:
for v in f:
v = v.split("\t",1) # split on tab
v0 = int(v[0]) # first parameter is tgid or start of tgid range
v1 = v0
if (len(v) > 1) and (int(v[1]) > v0): # second parameter if present is end of tgid range
v1 = int(v[1])
for tg in range(v0, (v1 + 1)):
if tg not in d: # is this a new tg?
d[tg] = [] # if so, add to dict (key only, value null)
sys.stderr.write('added talkgroup %d from %s\n' % (tg,s))
except (IndexError, ValueError) as ex:
return dict.fromkeys(d)
def utf_ascii(ustr):
return (ustr.decode("utf-8")).encode("ascii", "ignore")
class id_registry:
def __init__(self):
self.cache = {}
self.wildcards = {}
def add(self, id_str, tag, color):
if '-' in id_str:
a = [int(x) for x in id_str.split('-')]
if a[0] >= a[1]:
raise AssertionError('invalid ID range in %s' % id_str)
for i in range(a[0], a[1]+1):
self.cache[i] = {'tag': tag, 'color': color}
elif '.' in id_str:
rc = id_str.find('.')
self.wildcards[id_str] = {'prefix': id_str[:rc], 'len': len(id_str), 'tag': tag, 'color': color, 'type': '.'}
elif '*' in id_str:
rc = id_str.find('*')
self.wildcards[id_str] = {'prefix': id_str[:rc], 'len': 0, 'tag': tag, 'color': color, 'type': '*'}
self.cache[int(id_str)] = {'tag': tag, 'color': color}
def lookup(self, id):
if id in self.cache.keys():
return self.cache[id]
id_str = '%d' % id
for w in self.wildcards.keys():
if not id_str.startswith(self.wildcards[w]['prefix']):
if self.wildcards[w]['type'] == '.' and len(id_str) != self.wildcards[w]['len']:
# wildcard matched
self.cache[id] = {'tag': self.wildcards[w]['tag'], 'color': self.wildcards[w]['color']}
return self.cache[id]
# not found in cache, and no wildcard matched
self.cache[id] = None
return None
def get_color(self, id):
d = self.lookup(id)
if not d:
return 0
return d['color']
def get_tag(self, id):
d = self.lookup(id)
if not d:
return ""
return d['tag']
def load_tsv(tsv_filename):
hdrmap = []
configs = {}
with open(tsv_filename, 'r') as csvfile:
sreader = csv.reader(csvfile, delimiter='\t', quotechar='"', quoting=csv.QUOTE_ALL)
for row in sreader:
if not hdrmap:
# process first line of tsv file - header line
for hdr in row:
hdr = hdr.replace(' ', '_')
hdr = hdr.lower()
fields = {}
for i in range(len(row)):
if row[i]:
fields[hdrmap[i]] = row[i]
if hdrmap[i] != 'sysname':
fields[hdrmap[i]] = fields[hdrmap[i]].lower()
nac = int(fields['nac'], 0)
configs[nac] = fields
return configs
def read_tags_file(tags_file, result):
import csv
with open(tags_file, 'r') as csvfile:
sreader = csv.reader(csvfile, delimiter='\t', quotechar='"', quoting=csv.QUOTE_ALL)
for row in sreader:
id_str = row[0]
txt = row[1]
if len(row) >= 3:
color = int(row[2])
except ValueError as ex:
color = 0
color = 0
result.add(id_str, txt, color)
def make_config(configs):
result_config = {}
for nac in configs:
result_config[nac] = {'cclist':[], 'offset':0, 'whitelist':None, 'blacklist':{}, 'tgid_map':id_registry(), 'unit_id_map': id_registry(), 'sysname': configs[nac]['sysname'], 'center_frequency': None}
for f in configs[nac]['control_channel_list'].split(','):
if 'offset' in configs[nac]:
result_config[nac]['offset'] = int(configs[nac]['offset'])
if 'modulation' in configs[nac]:
result_config[nac]['modulation'] = configs[nac]['modulation']
result_config[nac]['modulation'] = 'cqpsk'
for k in ['whitelist', 'blacklist']:
if k in configs[nac]:
result_config[nac][k] = get_int_dict(configs[nac][k])
if 'tgid_tags_file' in configs[nac]:
fname = configs[nac]['tgid_tags_file']
if ',' in fname:
l = fname.split(',')
tgid_tags_file = l[0]
unit_id_tags_file = l[1]
tgid_tags_file = fname
unit_id_tags_file = None
result_config[nac]['tgid_tags_file'] = tgid_tags_file
result_config[nac]['unit_id_tags_file'] = unit_id_tags_file
read_tags_file(tgid_tags_file, result_config[nac]['tgid_map'])
if unit_id_tags_file is not None and os.access(unit_id_tags_file, os.R_OK):
read_tags_file(unit_id_tags_file, result_config[nac]['unit_id_map'])
if 'center_frequency' in configs[nac]:
result_config[nac]['center_frequency'] = get_frequency(configs[nac]['center_frequency'])
return result_config
def main():
import json
result = make_config(load_tsv(sys.argv[1]))
print (json.dumps(result, indent=4, separators=[',',':'], sort_keys=True))
if __name__ == '__main__':