2013-01-21 00:21:53 +01:00
|
|
|
import glob
|
|
|
|
import os
|
|
|
|
|
|
|
|
paths = ['src/*.cpp', 'src/kademlia/*.cpp', 'include/libtorrent/*.hpp', 'include/libtorrent/kademlia/*.hpp', 'include/libtorrent/aux_/*.hpp', 'include/libtorrent/extensions/*.hpp']
|
|
|
|
|
|
|
|
os.system('ctags %s 2>/dev/null' % ' '.join(paths))
|
|
|
|
|
|
|
|
files = []
|
|
|
|
|
|
|
|
for p in paths:
|
|
|
|
files.extend(glob.glob(p))
|
|
|
|
|
|
|
|
items = []
|
|
|
|
|
|
|
|
# keeps 20 non-comment lines, used to get more context around
|
|
|
|
# todo-items
|
|
|
|
context = []
|
|
|
|
|
2013-01-21 02:40:59 +01:00
|
|
|
priority_count = [0, 0, 0, 0]
|
|
|
|
|
2013-01-21 00:28:06 +01:00
|
|
|
def html_sanitize(s):
|
|
|
|
ret = ''
|
|
|
|
for i in s:
|
|
|
|
if i == '<': ret += '<'
|
|
|
|
elif i == '>': ret += '>'
|
|
|
|
elif i == '&': ret += '&'
|
|
|
|
else: ret += i
|
|
|
|
return ret
|
|
|
|
|
2013-01-21 00:21:53 +01:00
|
|
|
for f in files:
|
|
|
|
h = open(f)
|
|
|
|
|
|
|
|
state = ''
|
|
|
|
line_no = 0
|
|
|
|
context_lines = 0
|
|
|
|
|
|
|
|
for l in h:
|
|
|
|
line_no += 1
|
|
|
|
line = l.strip()
|
|
|
|
if 'TODO:' in line and line.startswith('//'):
|
|
|
|
line = line.split('TODO:')[1].strip()
|
|
|
|
state = 'todo'
|
|
|
|
items.append({})
|
|
|
|
items[-1]['location'] = '%s:%d' % (f, line_no)
|
|
|
|
items[-1]['priority'] = 0
|
|
|
|
if line[0] in '0123456789':
|
|
|
|
items[-1]['priority'] = int(line[0])
|
|
|
|
line = line[1:].strip()
|
|
|
|
items[-1]['todo'] = line
|
2013-01-21 02:40:59 +01:00
|
|
|
prio = items[-1]['priority']
|
|
|
|
if prio >= 0 and prio <= 3: priority_count[prio] += 1
|
2013-01-21 00:21:53 +01:00
|
|
|
continue
|
|
|
|
|
|
|
|
if state == '':
|
2013-01-21 00:28:06 +01:00
|
|
|
context.append(html_sanitize(l))
|
2013-01-21 00:21:53 +01:00
|
|
|
if len(context) > 20: context.pop(0)
|
|
|
|
continue
|
|
|
|
|
|
|
|
if state == 'todo':
|
|
|
|
if line.strip().startswith('//'):
|
|
|
|
items[-1]['todo'] += '\n'
|
|
|
|
items[-1]['todo'] += line[2:].strip()
|
|
|
|
else:
|
|
|
|
state = 'context'
|
2013-01-21 00:28:06 +01:00
|
|
|
items[-1]['context'] = ''.join(context) + '<div style="background: #ffff00" width="100%">' + html_sanitize(l) + '</div>';
|
2013-01-21 00:21:53 +01:00
|
|
|
context_lines = 1
|
2013-01-21 17:25:43 +01:00
|
|
|
|
|
|
|
context.append(html_sanitize(l))
|
|
|
|
if len(context) > 20: context.pop(0)
|
2013-01-21 00:21:53 +01:00
|
|
|
continue
|
|
|
|
|
|
|
|
if state == 'context':
|
2013-01-21 00:28:06 +01:00
|
|
|
items[-1]['context'] += html_sanitize(l)
|
2013-01-21 00:21:53 +01:00
|
|
|
context_lines += 1
|
2013-01-21 17:25:43 +01:00
|
|
|
|
|
|
|
context.append(html_sanitize(l))
|
|
|
|
if len(context) > 20: context.pop(0)
|
2013-01-21 00:21:53 +01:00
|
|
|
if context_lines > 30: state = ''
|
|
|
|
|
|
|
|
h.close()
|
|
|
|
|
|
|
|
items.sort(key = lambda x: x['priority'], reverse = True)
|
|
|
|
|
|
|
|
#for i in items:
|
|
|
|
# print '\n\n', i['todo'], '\n'
|
|
|
|
# print i['location'], '\n'
|
|
|
|
# print 'prio: ', i['priority'], '\n'
|
|
|
|
# if 'context' in i:
|
|
|
|
# print i['context'], '\n'
|
|
|
|
|
|
|
|
out = open('docs/todo.html', 'w+')
|
|
|
|
out.write('''<html><head>
|
|
|
|
<script type="text/javascript">
|
|
|
|
/* <![CDATA[ */
|
|
|
|
var expanded = -1
|
|
|
|
function expand(id) {
|
|
|
|
if (expanded != -1) {
|
|
|
|
var ctx = document.getElementById(expanded);
|
|
|
|
ctx.style.display = "none";
|
|
|
|
// if we're expanding the field that's already
|
|
|
|
// expanded, just collapse it
|
|
|
|
var no_expand = id == expanded;
|
|
|
|
expanded = -1;
|
|
|
|
if (no_expand) return;
|
|
|
|
}
|
|
|
|
var ctx = document.getElementById(id);
|
|
|
|
ctx.style.display = "table-row";
|
|
|
|
expanded = id;
|
|
|
|
}
|
|
|
|
/* ]]> */
|
|
|
|
</script>
|
|
|
|
|
|
|
|
</head><body>
|
|
|
|
<h1>libtorrent todo-list</h1>
|
2013-01-21 02:40:59 +01:00
|
|
|
<span style="color: #f77">%d important</span>
|
|
|
|
<span style="color: #3c3">%d relevant</span>
|
|
|
|
<span style="color: #77f">%d feasible</span>
|
|
|
|
<span style="color: #999">%d notes</span>
|
|
|
|
<table width="100%%" border="1" style="border-collapse: collapse;">''' % \
|
|
|
|
(priority_count[3], priority_count[2], priority_count[1], priority_count[0]))
|
|
|
|
|
|
|
|
prio_colors = [ '#ccc', '#ccf', '#cfc', '#fcc']
|
2013-01-21 00:21:53 +01:00
|
|
|
|
|
|
|
index = 0
|
|
|
|
for i in items:
|
|
|
|
if not 'context' in i: i['context'] = ''
|
2013-01-21 02:40:59 +01:00
|
|
|
out.write('<tr style="background: %s"><td>relevance %d</td><td><a href="javascript:expand(%d)">%s</a></td><td>%s</td></tr>' \
|
|
|
|
% (prio_colors[i['priority']], i['priority'], index, i['location'], i['todo'].replace('\n', ' ')))
|
2013-01-21 00:21:53 +01:00
|
|
|
|
|
|
|
out.write('<tr id="%d" style="display: none;" colspan="3"><td colspan="3"><h2>%s</h2><h4>%s</h4><pre style="background: #f6f6f6; border: solid 1px #ddd;">%s</pre></td></tr>' \
|
|
|
|
% (index, i['todo'], i['location'], i['context']))
|
|
|
|
index += 1
|
|
|
|
|
|
|
|
out.write('</table></body></html>')
|
|
|
|
out.close()
|
|
|
|
|