import urllib
from math import ceil, floor
import copy
+import wtforms
from babel.localedata import exists
import jinja2
return HTML_CLEANER.clean_html(html)
-MARKDOWN_INSTANCE = markdown.Markdown(safe_mode='escape')
+def convert_to_tag_list_of_dicts(tag_string):
+ """
+ Filter input from incoming string containing user tags,
+
+ Strips trailing, leading, and internal whitespace, and also converts
+ the "tags" text into an array of tags
+ """
+ taglist = []
+ if tag_string:
+
+ # Strip out internal, trailing, and leading whitespace
+ stripped_tag_string = u' '.join(tag_string.strip().split())
+
+ # Split the tag string into a list of tags
+ for tag in stripped_tag_string.split(
+ mg_globals.app_config['tags_delimiter']):
+
+ # Ignore empty or duplicate tags
+ if tag.strip() and tag.strip() not in [t['name'] for t in taglist]:
+
+ if mg_globals.app_config['tags_case_sensitive']:
+ taglist.append({'name': tag.strip(),
+ 'slug': slugify(tag.strip())})
+ else:
+ taglist.append({'name': tag.strip().lower(),
+ 'slug': slugify(tag.strip().lower())})
+ return taglist
+
+
+def media_tags_as_string(media_entry_tags):
+ """
+ Generate a string from a media item's tags, stored as a list of dicts
+
+ This is the opposite of convert_to_tag_list_of_dicts
+ """
+ media_tag_string = ''
+ if media_entry_tags:
+ media_tag_string = mg_globals.app_config['tags_delimiter'].join(
+ [tag['name'] for tag in media_entry_tags])
+ return media_tag_string
+
+TOO_LONG_TAG_WARNING = \
+ u'Tags must be shorter than %s characters. Tags that are too long: %s'
+
+def tag_length_validator(form, field):
+ """
+ Make sure tags do not exceed the maximum tag length.
+ """
+ tags = convert_to_tag_list_of_dicts(field.data)
+ too_long_tags = [
+ tag['name'] for tag in tags
+ if len(tag['name']) > mg_globals.app_config['tags_max_length']]
+
+ if too_long_tags:
+ raise wtforms.ValidationError(
+ TOO_LONG_TAG_WARNING % (mg_globals.app_config['tags_max_length'], \
+ ', '.join(too_long_tags)))
+MARKDOWN_INSTANCE = markdown.Markdown(safe_mode='escape')
+
def cleaned_markdown_conversion(text):
"""
Take a block of text, run it through MarkDown, and clean its HTML.