From 885c5835a002b67656174e09a68952aa2fa3c4de Mon Sep 17 00:00:00 2001 From: va7eex Date: Tue, 21 Feb 2017 00:31:17 -0800 Subject: [PATCH] Commenting --- get_tle.py | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-) diff --git a/get_tle.py b/get_tle.py index 588a84c..fe0d402 100644 --- a/get_tle.py +++ b/get_tle.py @@ -2,15 +2,11 @@ import requests import json import array import sys - +#Where we pull the data from. web_tle = [ 'http://www.celestrak.com/NORAD/elements/weather.txt', 'https://www.celestrak.com/NORAD/elements/stations.txt' ] +#celestrak uses a different key value than I do, find all instances of their keys and replace with mine. sats_of_interest = { 'NOAA-15':'NOAA 15', 'NOAA-18':'NOAA 18', 'NOAA-19':'NOAA 19', 'METEOR-M1':'METEOR-M 1', 'METEOR-M2':'METEOR-M 2', 'ISS':'ISS (ZARYA)' } -#sats_of_interest = { 'NOAA-15':'NOAA 15' } - - -#page = urllib2.urlopen(web_noaa_tle) -#page = requests.get(web_noaa_tle, stream=True) #NOAA 15 [B] #1 25338U 98030A 16173.50363904 .00000079 00000-0 52195-4 0 9991 @@ -21,6 +17,7 @@ line_num = 0 cur_sat_line = -1 cur_sat = "" +#This was originally intended to check that the data I was receiving was legitimate, very half-assed. def check_line1(str): if( len(str) == 69 ): fields = {} @@ -63,12 +60,13 @@ def check_line2(str): for tle in web_tle: page = requests.get(tle, stream=True) for line in page.iter_lines(): - #print '>>', line + #I don't actually remember why I made this var 'hsat' for hsat, sat in sats_of_interest.iteritems(): if( line.startswith(sat) ): cur_sat_line = line_num cur_sat = hsat + #Enable this for data integrity check if you want. if( line_num == cur_sat_line + 1 and cur_sat_line >= 0 ): if( check_line1(line) ): data[cur_sat] = line