aboutsummaryrefslogtreecommitdiffstats
path: root/include
diff options
context:
space:
mode:
authorfriendica <info@friendica.com>2012-11-06 16:16:26 -0800
committerfriendica <info@friendica.com>2012-11-06 16:16:26 -0800
commit55df0452b60d6faf00b782a03c8cc590f08d8f94 (patch)
tree5896aa3320fbc46c223150b9f8048dd4af952d05 /include
parent9050faa4874b691321c2a752803669911584a39c (diff)
downloadvolse-hubzilla-55df0452b60d6faf00b782a03c8cc590f08d8f94.tar.gz
volse-hubzilla-55df0452b60d6faf00b782a03c8cc590f08d8f94.tar.bz2
volse-hubzilla-55df0452b60d6faf00b782a03c8cc590f08d8f94.zip
finish off like/dislike localisation except for zrl/magic_link
Diffstat (limited to 'include')
-rw-r--r--include/conversation.php77
-rw-r--r--include/text.php18
2 files changed, 59 insertions, 36 deletions
diff --git a/include/conversation.php b/include/conversation.php
index f763bb0b9..ae7b47569 100644
--- a/include/conversation.php
+++ b/include/conversation.php
@@ -88,52 +88,59 @@ function localize_item(&$item){
if($extracted['images'])
$item['body'] = item_redir_and_replace_images($extracted['body'], $extracted['images'], $item['contact-id']);
- $xmlhead="<"."?xml version='1.0' encoding='UTF-8' ?".">";
-/*
+ logger('localize');
+
if (activity_match($item['verb'],ACTIVITY_LIKE) || activity_match($item['verb'],ACTIVITY_DISLIKE)){
-// FIXME we shouldn't have a DB lookup on every item
- $r = q("SELECT * from `item`,`contact` WHERE
- `item`.`contact-id`=`contact`.`id` AND `item`.`uri`='%s';",
- dbesc($item['parent_uri']));
- if(count($r)==0) return;
- $obj=$r[0];
+ $obj= json_decode($item['object']);
- $author = '[url=' . $item['author']['xchan_url'] . ']' . $item['author']['xchan_name'] . '[/url]';
- $objauthor = '[url=' . $obj['author-link'] . ']' . $obj['author-name'] . '[/url]';
+// logger('object: ' . print_r($obj,true));
+
+ if($obj->author && $obj->author->link)
+ $author_link = get_json_rel_link($obj->author->link,'alternate');
+ else
+ $author_link = '';
+
+ $author_name = (($obj->author && $obj->author->name) ? $obj->author->name : '');
+
+ $item_url = get_json_rel_link($obj->link,'alternate');
+
+
+ // If we couldn't parse something useful, don't bother translating.
+ // We need something better than zrl here, probably magic_link(), but it needs writing
+
+ if($author_link && $author_name && $item_url) {
+
+ $author = '[url=' . zrl($item['author']['xchan_url']) . ']' . $item['author']['xchan_name'] . '[/url]';
+ $objauthor = '[url=' . zrl($author_link) . ']' . $author_name . '[/url]';
- switch($obj['verb']){
- case ACTIVITY_POST:
- switch ($obj['obj_type']){
- case ACTIVITY_OBJ_EVENT:
- $post_type = t('event');
- break;
- default:
- $post_type = t('status');
- }
- break;
- default:
- if($obj['resource_id']){
+ switch($obj->type) {
+ case ACTIVITY_OBJ_PHOTO:
$post_type = t('photo');
- $m=array(); preg_match("/\[url=([^]]*)\]/", $obj['body'], $m);
- $rr['plink'] = $m[1];
- } else {
+ break;
+ case ACTIVITY_OBJ_EVENT:
+ $post_type = t('event');
+ break;
+ case ACTIVITY_OBJ_NOTE:
+ default:
$post_type = t('status');
- }
- }
+ break;
+ }
- $plink = '[url=' . $obj['plink'] . ']' . $post_type . '[/url]';
+ $plink = '[url=' . zrl($item_url) . ']' . $post_type . '[/url]';
+
+ if(activity_match($item['verb'],ACTIVITY_LIKE)) {
+ $bodyverb = t('%1$s likes %2$s\'s %3$s');
+ }
+ elseif(activity_match($item['verb'],ACTIVITY_DISLIKE)) {
+ $bodyverb = t('%1$s doesn\'t like %2$s\'s %3$s');
+ }
+ $item['body'] = sprintf($bodyverb, $author, $objauthor, $plink);
- if(activity_match($item['verb'],ACTIVITY_LIKE)) {
- $bodyverb = t('%1$s likes %2$s\'s %3$s');
- }
- elseif(activity_match($item['verb'],ACTIVITY_DISLIKE)) {
- $bodyverb = t('%1$s doesn\'t like %2$s\'s %3$s');
}
- $item['body'] = sprintf($bodyverb, $author, $objauthor, $plink);
}
-*/
+
if (activity_match($item['verb'],ACTIVITY_FRIEND)) {
if ($item['obj_type']=="" || $item['obj_type']!== ACTIVITY_OBJ_PERSON) return;
diff --git a/include/text.php b/include/text.php
index 464396d62..0d92c6875 100644
--- a/include/text.php
+++ b/include/text.php
@@ -1720,4 +1720,20 @@ function find_xchan_in_array($xchan,$arr) {
}
}
return array();
-} \ No newline at end of file
+}
+
+function get_json_rel_link($j,$rel) {
+ if(count($j))
+ foreach($j as $l)
+ if($l->rel === $rel)
+ return $l->href;
+ return '';
+}
+
+
+// Lots of code to write here
+
+function magic_link($s) {
+ return $s;
+}
+ \ No newline at end of file