| | |
| | import logging |
| | import re |
| |
|
| | log = logging.getLogger(__name__) |
| |
|
| |
|
| | def is_email(element): |
| | """ |
| | >>> is_email('username@example.com') |
| | True |
| | >>> is_email('example.com') |
| | False |
| | >>> is_email('firstname.lastname@domain.co.uk') |
| | True |
| | """ |
| | email_regex = r"^[A-Za-z0-9\.\+_-]+@[A-Za-z0-9\._-]+\.[a-zA-Z]*$" |
| | return re.match(email_regex, str(element)) |
| |
|
| |
|
| | def html_email(email, title=None): |
| | """ |
| | >>> html_email('username@example.com') |
| | '<a href="mailto:username@example.com">username@example.com</a>' |
| | """ |
| |
|
| | if not title: |
| | title = email |
| |
|
| | return '<a href="mailto:{email}">{title}</a>'.format(email=email, title=title) |
| |
|
| |
|
| | def html_list(data): |
| | """ |
| | >>> html_list(['example.com', 'admin1@example.com', 'admin2@example.com']) |
| | '<ul> |
| | <li>example.com</li> |
| | <li><a href="mailto:admin1@example.com">admin1@example.com</a></li> |
| | <li><a href="mailto:admin2@example.com">admin2@example.com</a></li> |
| | </ul>' |
| | """ |
| | html = "<ul>" |
| |
|
| | for item in data: |
| |
|
| | if isinstance(item, dict): |
| | if item.get("email"): |
| | item = html_email(item.get("email"), item.get("name", None)) |
| | elif item.get("name"): |
| | item = item.get("name") |
| |
|
| | if is_email(item): |
| | item = html_email(item, item) |
| |
|
| | html += "<li>{}</li>".format(item) |
| |
|
| | return html + "</ul>" |
| |
|
| |
|
| | def html_table_header_row(data): |
| | """ |
| | >>> html_table_header_row(['administrators', 'key', 'leader', 'project']) |
| | '\\n\\t<tr><th>Administrators</th><th>Key</th><th>Leader</th><th>Project</th></tr>' |
| | >>> html_table_header_row(['key', 'project', 'leader', 'administrators']) |
| | '\\n\\t<tr><th>Key</th><th>Project</th><th>Leader</th><th>Administrators</th></tr>' |
| | """ |
| | html = "\n\t<tr>" |
| |
|
| | for th in data: |
| | title = th.replace("_", " ").title() |
| | html += "<th>{}</th>".format(title) |
| |
|
| | return html + "</tr>" |
| |
|
| |
|
| | def html_row_with_ordered_headers(data, col_headers, row_header=None): |
| | """ |
| | >>> headers = ['administrators', 'key', 'leader', 'project'] |
| | >>> data = {'key': 'DEMO', 'project': 'Demonstration', |
| | 'leader': 'leader@example.com', |
| | 'administrators': ['admin1@example.com', 'admin2@example.com']} |
| | >>> html_row_with_ordered_headers(data, headers) |
| | '\\n\\t<tr><td><ul> |
| | <li><a href="mailto:admin1@example.com">admin1@example.com</a></li> |
| | <li><a href="mailto:admin2@example.com">admin2@example.com</a></li> |
| | </ul></td><td>DEMO</td><td>leader@example.com</td><td>Demonstration</td></tr>' |
| | >>> headers = ['key', 'project', 'leader', 'administrators'] |
| | >>> html_row_with_ordered_headers(data, headers) |
| | '\\n\\t<tr><td>DEMO</td><td>Demonstration</td> |
| | <td>leader@example.com</td><td> |
| | <ul> |
| | <li><a href="mailto:admin1@example.com">admin1@example.com</a></li> |
| | <li><a href="mailto:admin2@example.com">admin2@example.com</a></li> |
| | </ul></td></tr>' |
| | """ |
| | html = "\n\t<tr>" |
| |
|
| | if row_header: |
| | html += "<th>{}</th>".format(row_header.replace("_", " ").title()) |
| | for header in col_headers: |
| | element = data[header] |
| |
|
| | if isinstance(element, list): |
| | element = html_list(element) |
| |
|
| | if is_email(element): |
| | element = html_email(element) |
| |
|
| | html += "<td>{}</td>".format(element) |
| |
|
| | return html + "</tr>" |
| |
|
| |
|
| | def html_table_from_dict(data, ordering): |
| | """ |
| | >>> ordering = ['administrators', 'key', 'leader', 'project'] |
| | >>> data = [ \ |
| | {'key': 'DEMO', 'project': 'Demo project', 'leader': 'lead@example.com', \ |
| | 'administrators': ['admin@example.com', 'root@example.com']},] |
| | >>> html_table_from_dict(data, ordering) |
| | '<table> |
| | <tbody>\\n |
| | <tr> |
| | <th>Administrators</th> |
| | <th>Key</th> |
| | <th>Leader</th> |
| | <th>Project</th> |
| | </tr>\\n |
| | <tr> |
| | <td> |
| | <ul> |
| | <li><a href="mailto:admin@example.com">admin@example.com</a></li> |
| | <li><a href="mailto:root@example.com">root@example.com</a></li> |
| | </ul> |
| | </td> |
| | <td>DEMO</td> |
| | <td>lead@example.com</td> |
| | <td>Demo project</td> |
| | </tr>\\n |
| | </tbody> |
| | </table>' |
| | >>> ordering = ['key', 'project', 'leader', 'administrators'] |
| | >>> html_table_from_dict(data, ordering) |
| | '<table> |
| | <tbody>\\n |
| | <tr> |
| | <th>Key</th> |
| | <th>Project</th> |
| | <th>Leader</th> |
| | <th>Administrators</th> |
| | </tr>\\n |
| | <tr> |
| | <td>DEMO</td> |
| | <td>Demo project</td> |
| | <td>lead@example.com</td> |
| | <td> |
| | <ul> |
| | <li><a href="mailto:admin@example.com">admin@example.com</a></li> |
| | <li><a href="mailto:root@example.com">root@example.com</a></li> |
| | </ul> |
| | </td> |
| | </tr>\\n |
| | </tbody> |
| | </table>' |
| | """ |
| | html = "<table><tbody>" |
| | html += html_table_header_row(ordering) |
| |
|
| | for row in data: |
| | html += html_row_with_ordered_headers(row, ordering) |
| |
|
| | return html + "\n</tbody></table>" |
| |
|
| |
|
| | def html_table_from_nested_dict(data, ordering): |
| | """ |
| | >>> ordering = ['manager', 'admin', 'employee_count'] |
| | >>> data = { |
| | 'project_A': {'manager': 'John', 'admin': 'admin1@example.com', 'employee_count': '4'}, |
| | 'project_B': {'manager': 'Jane', 'admin': 'admin2@example.com', 'employee_count': '7'} |
| | } |
| | >>> html_table_from_nested_dict(data, ordering) |
| | <table> |
| | <tbody> |
| | <tr> |
| | <th></th> |
| | <th>Manager</th> |
| | <th>Admin</th> |
| | <th>Employee Count</th> |
| | </tr> |
| | <tr> |
| | <th>Project A</th> |
| | <td>John</td> |
| | <td><a href="mailto:admin1@example.com">admin1@example.com</a></td> |
| | <td>4</td> |
| | </tr> |
| | <tr> |
| | <th>Project B</th> |
| | <td>Jane</td> |
| | <td><a href="mailto:admin2@example.com">admin2@example.com</a></td> |
| | <td>7</td> |
| | </tr> |
| | </tbody> |
| | </table> |
| | """ |
| |
|
| | html = "<table><tbody>" |
| | |
| | header_row = [""] |
| | header_row.extend(ordering) |
| | html += html_table_header_row(header_row) |
| |
|
| | for row_header, row in data.items(): |
| | html += html_row_with_ordered_headers(row, ordering, row_header) |
| |
|
| | return html + "\n</tbody></table>" |
| |
|
| |
|
| | def block_code_macro_confluence(code, lang=None): |
| | """ |
| | Wrap into code block macro |
| | :param code: |
| | :param lang: |
| | :return: |
| | """ |
| | if not lang: |
| | lang = "" |
| | return ( |
| | """\ |
| | <ac:structured-macro ac:name="code" ac:schema-version="1"> |
| | <ac:parameter ac:name="language">{lang}</ac:parameter> |
| | <ac:plain-text-body><![CDATA[{code}]]></ac:plain-text-body> |
| | </ac:structured-macro> |
| | """ |
| | ).format(lang=lang, code=code) |
| |
|
| |
|
| | def html_code__macro_confluence(text): |
| | """ |
| | Wrap into html macro |
| | :param text: |
| | :return: |
| | """ |
| | return ( |
| | """\ |
| | <ac:structured-macro ac:name="html" ac:schema-version="1"> |
| | <ac:plain-text-body><![CDATA[{text}]]></ac:plain-text-body> |
| | </ac:structured-macro> |
| | """ |
| | ).format(text=text) |
| |
|
| |
|
| | def noformat_code_macro_confluence(text, nopanel=None): |
| | """ |
| | Wrap into code block macro |
| | :param text: |
| | :param nopanel: (bool) True or False Removes the panel around the content. |
| | :return: |
| | """ |
| | if not nopanel: |
| | nopanel = False |
| | return ( |
| | """\ |
| | <ac:structured-macro ac:name="noformat" ac:schema-version="1"> |
| | <ac:parameter ac:name="nopanel">{nopanel}</ac:parameter> |
| | <ac:plain-text-body><![CDATA[{text}]]></ac:plain-text-body> |
| | </ac:structured-macro> |
| | """ |
| | ).format(nopanel=nopanel, text=text) |
| |
|
| |
|
| | def symbol_normalizer(text): |
| | if not text: |
| | return "" |
| | result = text |
| | result = result.replace("Ä", u"Ä") |
| | result = result.replace("ä", u"ä") |
| | result = result.replace("Ë", u"Ë") |
| | result = result.replace("ë", u"ë") |
| | result = result.replace("Ï", u"Ï") |
| | result = result.replace("ï", u"ï") |
| | result = result.replace("Ö", u"Ö") |
| | result = result.replace("ö", u"ö") |
| | result = result.replace("Ü", u"Ü") |
| | result = result.replace("ü", u"ü") |
| | result = result.replace("Á", u"Á") |
| | result = result.replace("á", u"á") |
| | result = result.replace("É", u"É") |
| | result = result.replace("é", u"é") |
| | result = result.replace("Í", u"Í") |
| | result = result.replace("í", u"í") |
| | result = result.replace("Ó", u"Ó") |
| | result = result.replace("ó", u"ó") |
| | result = result.replace("Ú", u"Ú") |
| | result = result.replace("ú", u"ú") |
| | result = result.replace("À", u"À") |
| | result = result.replace("à", u"à") |
| | result = result.replace("È", u"È") |
| | result = result.replace("è", u"è") |
| | result = result.replace("Ì", u"Ì") |
| | result = result.replace("ì", u"ì") |
| | result = result.replace("Ò", u"Ò") |
| | result = result.replace("ò", u"ò") |
| | result = result.replace("Ù", u"Ù") |
| | result = result.replace("ù", u"ù") |
| | result = result.replace("Â", u"Â") |
| | result = result.replace("â", u"â") |
| | result = result.replace("Ê", u"Ê") |
| | result = result.replace("ê", u"ê") |
| | result = result.replace("Î", u"Î") |
| | result = result.replace("î", u"î") |
| | result = result.replace("Ô", u"Ô") |
| | result = result.replace("ô", u"ô") |
| | result = result.replace("Û", u"Û") |
| | result = result.replace("û", u"û") |
| | result = result.replace("Å", u"Å") |
| | result = result.replace("å", u"å") |
| | result = result.replace("°", u"°") |
| | return result |
| |
|
| |
|
| | def parse_cookie_file(cookie_file): |
| | """ |
| | Parse a cookies.txt file (Netscape HTTP Cookie File) |
| | return a dictionary of key value pairs |
| | compatible with requests. |
| | :param cookie_file: a cookie file |
| | :return dict of cookies pair |
| | """ |
| | cookies = {} |
| | with open(cookie_file, "r") as fp: |
| | for line in fp: |
| | if not re.match(r"^(#|$)", line): |
| | line_fields = line.strip().split("\t") |
| | try: |
| | cookies[line_fields[5]] = line_fields[6] |
| | except IndexError as e: |
| | log.error(e) |
| | return cookies |
| |
|