forked from friendica/friendica-addons
Discourse addon created
This commit is contained in:
parent
43503abc15
commit
da663db1cb
97
discourse/README
Normal file
97
discourse/README
Normal file
|
@ -0,0 +1,97 @@
|
|||
NSFW
|
||||
|
||||
"Not safe for work"
|
||||
|
||||
Scans the message content for the string 'nsfw'
|
||||
(case insensitive) and if found replaces the content
|
||||
with a "click to open/close" link, default is closed.
|
||||
|
||||
If you click on the 'Not safe for work' addon under
|
||||
/settings/addon a text field appears, where you can
|
||||
extend the list of search terms. The terms must be
|
||||
seperated by commas.
|
||||
|
||||
It is also possible to enter profile URLs as values.
|
||||
This is quite useful for the case, that you perhaps
|
||||
don't want to see postings by person_A, but person_B
|
||||
is one of your contacts and person_B used to reshare
|
||||
postings by person_A.
|
||||
|
||||
You can also make use of regular expressions.
|
||||
They also have to be seperated by commas and the
|
||||
regex itself has to be enclosed with slashes:
|
||||
|
||||
... nsfw, /<REGEX>/, politics,...
|
||||
|
||||
---------------
|
||||
A few examples:
|
||||
---------------
|
||||
|
||||
1)
|
||||
Let's say you don't want to see postings which contain
|
||||
the term 'fake news'
|
||||
|
||||
The term could appear in several ways:
|
||||
|
||||
fakenews, fake news, fake_news, fake-news, f@ke news,
|
||||
f4ke news, f4k3 n3ws, and so on and so on and so on.
|
||||
|
||||
You could write every possible version of it as single
|
||||
item into your NSFW-filter list, but this can also be
|
||||
done with a single regex, which matches all of them:
|
||||
|
||||
/f[@4a]k[3e][-_ ]n[3e]w[sz]/
|
||||
|
||||
|
||||
2)
|
||||
Another use case could be, that you are simply not
|
||||
interested in postings about christmas.
|
||||
|
||||
/christmas(?:[-_ ]?(?:tree|time|eve|pudding))?/
|
||||
|
||||
|
||||
ATTENTION:
|
||||
|
||||
It is absolutely important, that you use grouping
|
||||
parentheses instead of capturing parentheses!!
|
||||
|
||||
Grouping parentheses are:
|
||||
|
||||
(?: )
|
||||
|
||||
If you use capturing parentheses, which are
|
||||
|
||||
( )
|
||||
|
||||
it will produce errors and the regex won't work and
|
||||
at least your targets will not get collapsed.
|
||||
|
||||
|
||||
|
||||
3)
|
||||
Another possibility is the usage of a so called
|
||||
'lookbehind' construct. I'll give an example followed
|
||||
by a descripton:
|
||||
|
||||
/(?<!the )\badvent\b/
|
||||
|
||||
The \b is a word boundary, what matches the beginning
|
||||
and the end of a word. The simple pattern of 'advent'
|
||||
would match advent iteself, but also adventure.
|
||||
This can be prevented by
|
||||
|
||||
/\badvent\b/
|
||||
|
||||
The first part of the regex above
|
||||
|
||||
(?<!the )
|
||||
|
||||
is a negative lookbehind. It makes \badvent\b only
|
||||
match, if there is no 'the ' before \badvent\b or in
|
||||
words:
|
||||
|
||||
It looks for 'advent', but doesn't match 'the advent'.
|
||||
|
||||
|
||||
For more informations take a look at the PCRE regex
|
||||
dialect.
|
224
discourse/discourse.php
Normal file
224
discourse/discourse.php
Normal file
|
@ -0,0 +1,224 @@
|
|||
<?php
|
||||
|
||||
/**
|
||||
* Name: Discourse Mail Connector
|
||||
* Description: Improves mails from Discourse in mailing list mode
|
||||
* Version: 0.1
|
||||
* Author: Michael Vogel <http://pirati.ca/profile/heluecht>
|
||||
*
|
||||
*/
|
||||
//use DOMDocument;
|
||||
//use DOMXPath;
|
||||
use Friendica\App;
|
||||
use Friendica\Core\Hook;
|
||||
use Friendica\Core\L10n;
|
||||
use Friendica\Core\Logger;
|
||||
use Friendica\Core\PConfig;
|
||||
use Friendica\Util\XML;
|
||||
use Friendica\Content\Text\Markdown;
|
||||
use Friendica\Util\Network;
|
||||
Use Friendica\Util\DateTimeFormat;
|
||||
|
||||
function discourse_install()
|
||||
{
|
||||
Hook::register('email_getmessage', __FILE__, 'discourse_email_getmessage');
|
||||
Hook::register('email_getmessage_end', __FILE__, 'discourse_email_getmessage_end');
|
||||
Hook::register('addon_settings', __FILE__, 'discourse_addon_settings');
|
||||
Hook::register('addon_settings_post', __FILE__, 'discourse_addon_settings_post');
|
||||
}
|
||||
|
||||
function discourse_uninstall()
|
||||
{
|
||||
Hook::unregister('email_getmessage', __FILE__, 'discourse_email_getmessage');
|
||||
Hook::unregister('email_getmessage_end', __FILE__, 'discourse_email_getmessage_end');
|
||||
Hook::unregister('addon_settings', __FILE__, 'discourse_addon_settings');
|
||||
Hook::unregister('addon_settings_post', __FILE__, 'discourse_addon_settings_post');
|
||||
}
|
||||
|
||||
function discourse_addon_settings(App $a, &$s)
|
||||
{
|
||||
}
|
||||
|
||||
function discourse_addon_settings_post(App $a)
|
||||
{
|
||||
}
|
||||
|
||||
function discourse_email_getmessage(App $a, &$message)
|
||||
{
|
||||
// Logger::info('Got raw message', $message);
|
||||
// Remove the title on comments, they don't serve any purpose there
|
||||
if ($message['item']['parent-uri'] != $message['item']['uri']) {
|
||||
unset($message['item']['title']);
|
||||
}
|
||||
|
||||
if (preg_match('=topic/(.*)/(.*)@(.*)=', $message['item']['uri'], $matches)) {
|
||||
Logger::info('Got post data', ['topic' => $matches[1], 'post' => $matches[2], 'host' => $matches[3]]);
|
||||
if (discourse_fetch_post_from_api($message, $matches[2], $matches[3])) {
|
||||
return;
|
||||
}
|
||||
}
|
||||
|
||||
// Search in the text part for the link to the discourse entry and the text body
|
||||
// The text body is used as alternative, if the fetched HTML isn't working
|
||||
if (!empty($message['text'])) {
|
||||
discourse_get_text($message);
|
||||
}
|
||||
|
||||
if (!empty($message['item']['plink'])) {
|
||||
if (preg_match('=(http.*)/t/.*/(.*\d)/(.*\d)=', $message['item']['plink'], $matches)) {
|
||||
if (discourse_fetch_topic_from_api($message, $matches[1], $matches[1], $matches[1])) {
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Search in the HTML part for the discourse entry and the author profile
|
||||
if (!empty($message['html'])) {
|
||||
discourse_get_html($message);
|
||||
}
|
||||
}
|
||||
|
||||
function discourse_fetch_topic_from_api(&$message, $host, $topic, $pid)
|
||||
{
|
||||
$url = $host . '/t/' . $topic . '/posts.json?posts_ids[]=' . $pid;
|
||||
$curlResult = Network::curl($url);
|
||||
if (!$curlResult->isSuccess()) {
|
||||
return false;
|
||||
}
|
||||
$raw = $curlResult->getBody();
|
||||
$data = json_decode($raw, true);
|
||||
$posts = $data['post_stream']['posts'];
|
||||
foreach($posts as $post) {
|
||||
if ($post['post_number'] != $pid) {
|
||||
continue;
|
||||
}
|
||||
Logger::info('Got post data from topic', $post);
|
||||
discourse_process_post($message, $post);
|
||||
return true;
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
function discourse_fetch_post_from_api(&$message, $post, $host)
|
||||
{
|
||||
$url = "https://" . $host . '/posts/' . $post . '.json';
|
||||
$curlResult = Network::curl($url);
|
||||
if (!$curlResult->isSuccess()) {
|
||||
return false;
|
||||
}
|
||||
|
||||
$raw = $curlResult->getBody();
|
||||
$data = json_decode($raw, true);
|
||||
if (empty($data)) {
|
||||
return false;
|
||||
}
|
||||
|
||||
discourse_process_post($message, $data);
|
||||
|
||||
Logger::info('Got API data', $message);
|
||||
return true;
|
||||
}
|
||||
|
||||
function discourse_process_post(&$message, $post)
|
||||
{
|
||||
if ($post['post_number'] == 1) {
|
||||
// Thread information
|
||||
}
|
||||
|
||||
$nick = $post['username'];
|
||||
$name = $post['name'];
|
||||
// User information
|
||||
|
||||
$message['html'] = $post['cooked'];
|
||||
$message['text'] = $post['raw'];
|
||||
$message['item']['created'] = DateTimeFormat::utc($post['created_at']);
|
||||
}
|
||||
|
||||
function discourse_get_html(&$message)
|
||||
{
|
||||
$doc = new DOMDocument();
|
||||
$doc2 = new DOMDocument();
|
||||
$doc->preserveWhiteSpace = false;
|
||||
|
||||
$html = mb_convert_encoding($message['html'], 'HTML-ENTITIES', "UTF-8");
|
||||
@$doc->loadHTML($html, LIBXML_HTML_NODEFDTD);
|
||||
|
||||
$xpath = new DomXPath($doc);
|
||||
|
||||
// Fetch the first 'div' before the 'hr' -hopefully this fits for all systems
|
||||
$result = $xpath->query("//hr//preceding::div[1]");
|
||||
$div = $doc2->importNode($result->item(0), true);
|
||||
$doc2->appendChild($div);
|
||||
$message['html'] = $doc2->saveHTML();
|
||||
Logger::info('Found html body', ['html' => $message['html']]);
|
||||
|
||||
$profile = discourse_get_profile($xpath);
|
||||
if (!empty($profile)) {
|
||||
Logger::info('Found profile', $profile);
|
||||
/*
|
||||
$message['item']['author-avatar'] = $contact['avatar'];
|
||||
$message['item']['author-link'] = $profile['link'];
|
||||
$message['item']['author-name'] = $profile['name'];
|
||||
*/
|
||||
}
|
||||
}
|
||||
|
||||
function discourse_get_text(&$message)
|
||||
{
|
||||
$text = $message['text'];
|
||||
$text = str_replace("\r", '', $text);
|
||||
$pos = strpos($text, "\n---\n");
|
||||
if ($pos > 0) {
|
||||
$message['text'] = trim(substr($text, 0, $pos));
|
||||
Logger::info('Found text body', ['text' => $message['text']]);
|
||||
|
||||
$message['text'] = Markdown::toBBCode($message['text']);
|
||||
|
||||
$text = substr($text, $pos);
|
||||
if (preg_match('=\((http.*?)\)=', $text, $link)) {
|
||||
$message['item']['plink'] = $link[1];
|
||||
Logger::info('Found plink', ['plink' => $message['item']['plink']]);
|
||||
}
|
||||
} else {
|
||||
Logger::info('No separator found', ['text' => $text]);
|
||||
}
|
||||
}
|
||||
|
||||
function discourse_get_profile($xpath)
|
||||
{
|
||||
$profile = [];
|
||||
$list = $xpath->query("//td//following::img");
|
||||
foreach ($list as $node) {
|
||||
$attr = [];
|
||||
foreach ($node->attributes as $attribute) {
|
||||
$attr[$attribute->name] = $attribute->value;
|
||||
}
|
||||
|
||||
if (!empty($attr['src']) && !empty($attr['title'])
|
||||
&& !empty($attr['width']) && !empty($attr['height'])
|
||||
&& ($attr['width'] == $attr['height'])) {
|
||||
$profile = ['avatar' => $attr['src'], 'name' => $attr['title']];
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
$list = $xpath->query("//td//following::a");
|
||||
foreach ($list as $node) {
|
||||
if (!empty(trim($node->textContent)) && $node->attributes->length) {
|
||||
$attr = [];
|
||||
foreach ($node->attributes as $attribute) {
|
||||
$attr[$attribute->name] = $attribute->value;
|
||||
}
|
||||
if (!empty($attr['href']) && (strpos($attr['href'], '/' . $profile['name']))) {
|
||||
$profile['link'] = $attr['href'];
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
return $profile;
|
||||
}
|
||||
|
||||
function discourse_email_getmessage_end(App $a, &$message)
|
||||
{
|
||||
// Logger::info('Got converted message', $message);
|
||||
}
|
Loading…
Reference in a new issue