godot/doc/tools/doc_status.py
Rémi Verschelde fc8ccd5b8c SCsub: Add python shebang as a hint for syntax highlighting
Also switch existing shebangs to "better" /usr/bin/env python.
2016-10-17 20:10:46 +02:00

431 lines
13 KiB
Python
Executable file

#!/usr/bin/env python3
import sys
import re
import math
import platform
import xml.etree.ElementTree as ET
################################################################################
# Config #
################################################################################
flags = {
'c': platform.platform() != 'Windows', # Disable by default on windows, since we use ANSI escape codes
'b': False,
'g': False,
's': False,
'u': False,
'h': False,
'p': False,
'o': True,
'i': False,
}
flag_descriptions = {
'c': 'Toggle colors when outputting.',
'b': 'Toggle showing only not fully described classes.',
'g': 'Toggle showing only completed classes.',
's': 'Toggle showing comments about the status.',
'u': 'Toggle URLs to docs.',
'h': 'Show help and exit.',
'p': 'Toggle showing percentage as well as counts.',
'o': 'Toggle overall column.',
'i': 'Toggle collapse of class items columns.',
}
long_flags = {
'colors': 'c',
'use-colors': 'c',
'bad': 'b',
'only-bad': 'b',
'good': 'g',
'only-good': 'g',
'comments': 's',
'status': 's',
'urls': 'u',
'gen-url': 'u',
'help': 'h',
'percent': 'p',
'use-percentages': 'p',
'overall': 'o',
'use-overall': 'o',
'items': 'i',
'collapse': 'i',
}
table_columns = ['name', 'brief_description', 'description', 'methods', 'constants', 'members', 'signals']
table_column_names = ['Name', 'Brief Desc.', 'Desc.', 'Methods', 'Constants', 'Members', 'Signals']
colors = {
'name': [36], # cyan
'part_big_problem': [4, 31], # underline, red
'part_problem': [31], # red
'part_mostly_good': [33], # yellow
'part_good': [32], # green
'url': [4, 34], # underline, blue
'section': [1, 4], # bold, underline
'state_off': [36], # cyan
'state_on': [1, 35], # bold, magenta/plum
}
overall_progress_description_weigth = 10
################################################################################
# Utils #
################################################################################
def validate_tag(elem, tag):
if elem.tag != tag:
print('Tag mismatch, expected "' + tag + '", got ' + elem.tag)
sys.exit(255)
def color(color, string):
if flags['c']:
color_format = ''
for code in colors[color]:
color_format += '\033[' + str(code) + 'm'
return color_format + string + '\033[0m'
else:
return string
ansi_escape = re.compile(r'\x1b[^m]*m')
def nonescape_len(s):
return len(ansi_escape.sub('', s))
################################################################################
# Classes #
################################################################################
class ClassStatusProgress:
def __init__(self, described = 0, total = 0):
self.described = described
self.total = total
def __add__(self, other):
return ClassStatusProgress(self.described + other.described, self.total + other.total)
def increment(self, described):
if described:
self.described += 1
self.total += 1
def is_ok(self):
return self.described >= self.total
def to_configured_colored_string(self):
if flags['p']:
return self.to_colored_string('{percent}% ({has}/{total})', '{pad_percent}{pad_described}{s}{pad_total}')
else:
return self.to_colored_string()
def to_colored_string(self, format='{has}/{total}', pad_format='{pad_described}{s}{pad_total}'):
ratio = self.described/self.total if self.total != 0 else 1
percent = round(100*ratio)
s = format.format(has = str(self.described), total = str(self.total), percent = str(percent))
if self.described >= self.total:
s = color('part_good', s)
elif self.described >= self.total/4*3:
s = color('part_mostly_good', s)
elif self.described > 0:
s = color('part_problem', s)
else:
s = color('part_big_problem', s)
pad_size = max(len(str(self.described)), len(str(self.total)))
pad_described = ''.ljust(pad_size - len(str(self.described)))
pad_percent = ''.ljust(3 - len(str(percent)))
pad_total = ''.ljust(pad_size - len(str(self.total)))
return pad_format.format(pad_described = pad_described, pad_total = pad_total, pad_percent = pad_percent, s = s)
class ClassStatus:
def __init__(self, name=''):
self.name = name
self.has_brief_description = True
self.has_description = True
self.progresses = {
'methods': ClassStatusProgress(),
'constants': ClassStatusProgress(),
'members': ClassStatusProgress(),
'signals': ClassStatusProgress()
}
def __add__(self, other):
new_status = ClassStatus()
new_status.name = self.name
new_status.has_brief_description = self.has_brief_description and other.has_brief_description
new_status.has_description = self.has_description and other.has_description
for k in self.progresses:
new_status.progresses[k] = self.progresses[k] + other.progresses[k]
return new_status
def is_ok(self):
ok = True
ok = ok and self.has_brief_description
ok = ok and self.has_description
for k in self.progresses:
ok = ok and self.progresses[k].is_ok()
return ok
def make_output(self):
output = {}
output['name'] = color('name', self.name)
ok_string = color('part_good', 'OK')
missing_string = color('part_big_problem', 'MISSING')
output['brief_description'] = ok_string if self.has_brief_description else missing_string
output['description'] = ok_string if self.has_description else missing_string
description_progress = ClassStatusProgress(
(self.has_brief_description + self.has_description) * overall_progress_description_weigth,
2 * overall_progress_description_weigth
)
items_progress = ClassStatusProgress()
for k in ['methods', 'constants', 'members', 'signals']:
items_progress += self.progresses[k]
output[k] = self.progresses[k].to_configured_colored_string()
output['items'] = items_progress.to_configured_colored_string()
output['overall'] = (description_progress + items_progress).to_colored_string('{percent}%', '{pad_percent}{s}')
if self.name.startswith('Total'):
output['url'] = color('url', 'http://docs.godotengine.org/en/latest/classes/_classes.html')
if flags['s']:
output['comment'] = color('part_good', 'ALL OK')
else:
output['url'] = color('url', 'http://docs.godotengine.org/en/latest/classes/class_{name}.html'.format(name=self.name.lower()))
if flags['s'] and not flags['g'] and self.is_ok():
output['comment'] = color('part_good', 'ALL OK')
return output
def generate_for_class(c):
status = ClassStatus()
status.name = c.attrib['name']
for tag in list(c):
if tag.tag == 'brief_description':
status.has_brief_description = len(tag.text.strip()) > 0
elif tag.tag == 'description':
status.has_description = len(tag.text.strip()) > 0
elif tag.tag in ['methods', 'signals']:
for sub_tag in list(tag):
descr = sub_tag.find('description')
status.progresses[tag.tag].increment(len(descr.text.strip()) > 0)
elif tag.tag in ['constants', 'members']:
for sub_tag in list(tag):
status.progresses[tag.tag].increment(len(sub_tag.text.strip()) > 0)
elif tag.tag in ['theme_items']:
pass #Ignore those tags, since they seem to lack description at all
else:
print(tag.tag, tag.attrib)
return status
################################################################################
# Arguments #
################################################################################
input_file_list = []
input_class_list = []
for arg in sys.argv[1:]:
if arg.startswith('--'):
flags[long_flags[arg[2:]]] = not flags[long_flags[arg[2:]]]
elif arg.startswith('-'):
for f in arg[1:]:
flags[f] = not flags[f]
elif arg.endswith('.xml'):
input_file_list.append(arg)
else:
input_class_list.append(arg)
if flags['i']:
for r in ['methods', 'constants', 'members', 'signals']:
index = table_columns.index(r)
del table_column_names[index]
del table_columns[index]
table_column_names.append('Items')
table_columns.append('items')
if flags['o'] == (not flags['i']):
table_column_names.append('Overall')
table_columns.append('overall')
if flags['u']:
table_column_names.append('Docs URL')
table_columns.append('url')
################################################################################
# Help #
################################################################################
if len(input_file_list) < 1 or flags['h']:
if not flags['h']:
print(color('section', 'Invalid usage') + ': At least one classes.xml file is required')
print(color('section', 'Usage') + ': doc_status.py [flags] <classes.xml> [class names]')
print('\t< and > signify required parameters, while [ and ] signify optional parameters.')
print('\tNote that you can give more than one classes file, in which case they will be merged on-the-fly.')
print(color('section', 'Available flags') + ':')
possible_synonym_list = list(long_flags)
possible_synonym_list.sort()
flag_list = list(flags)
flag_list.sort()
for flag in flag_list:
synonyms = [color('name', '-' + flag)]
for synonym in possible_synonym_list:
if long_flags[synonym] == flag:
synonyms.append(color('name', '--' + synonym))
print(('{synonyms} (Currently '+color('state_'+('on' if flags[flag] else 'off'), '{value}')+')\n\t{description}').format(
synonyms = ', '.join(synonyms),
value = ('on' if flags[flag] else 'off'),
description = flag_descriptions[flag]
))
sys.exit(0)
################################################################################
# Parse class list #
################################################################################
class_names = []
classes = {}
for file in input_file_list:
tree = ET.parse(file)
doc = tree.getroot()
if 'version' not in doc.attrib:
print('Version missing from "doc"')
sys.exit(255)
version = doc.attrib['version']
for c in list(doc):
if c.attrib['name'] in class_names:
continue
class_names.append(c.attrib['name'])
classes[c.attrib['name']] = c
class_names.sort()
if len(input_class_list) < 1:
input_class_list = class_names
################################################################################
# Make output table #
################################################################################
table = [table_column_names]
table_row_chars = '+- '
table_column_chars = '|'
total_status = ClassStatus('Total')
for cn in input_class_list:
if not cn in classes:
print('Cannot find class ' + cn + '!')
sys.exit(255)
c = classes[cn]
validate_tag(c, 'class')
status = ClassStatus.generate_for_class(c)
if flags['b'] and status.is_ok():
continue
if flags['g'] and not status.is_ok():
continue
total_status = total_status + status
out = status.make_output()
row = []
for column in table_columns:
if column in out:
row.append(out[column])
else:
row.append('')
if 'comment' in out and out['comment'] != '':
row.append(out['comment'])
table.append(row)
################################################################################
# Print output table #
################################################################################
if len(table) == 1:
print(color('part_big_problem', 'No classes suitable for printing!'))
sys.exit(0)
if len(table) > 2:
total_status.name = 'Total = {0}'.format(len(table) - 1)
out = total_status.make_output()
row = []
for column in table_columns:
if column in out:
row.append(out[column])
else:
row.append('')
table.append(row)
table_column_sizes = []
for row in table:
for cell_i, cell in enumerate(row):
if cell_i >= len(table_column_sizes):
table_column_sizes.append(0)
table_column_sizes[cell_i] = max(nonescape_len(cell), table_column_sizes[cell_i])
divider_string = table_row_chars[0]
for cell_i in range(len(table[0])):
divider_string += table_row_chars[1] * (table_column_sizes[cell_i] + 2) + table_row_chars[0]
print(divider_string)
for row_i, row in enumerate(table):
row_string = table_column_chars
for cell_i, cell in enumerate(row):
padding_needed = table_column_sizes[cell_i] - nonescape_len(cell) + 2
if cell_i == 0:
row_string += table_row_chars[2] + cell + table_row_chars[2]*(padding_needed-1)
else:
row_string += table_row_chars[2]*math.floor(padding_needed/2) + cell + table_row_chars[2]*math.ceil((padding_needed/2))
row_string += table_column_chars
print(row_string)
if row_i == 0 or row_i == len(table) - 2:
print(divider_string)
print(divider_string)
if total_status.is_ok() and not flags['g']:
print('All listed classes are ' + color('part_good', 'OK') + '!')