aboutsummaryrefslogtreecommitdiffstats
path: root/include/feedutils.php
diff options
context:
space:
mode:
authorzotlabs <mike@macgirvin.com>2017-04-11 20:17:22 -0700
committerMario Vavti <mario@mariovavti.com>2017-04-12 07:04:22 +0200
commit7b173a75e47cfdf9f46b7d635df84b2be286b0e6 (patch)
tree240f259b9cdff0a495b1bc5abf7c74a9105a6beb /include/feedutils.php
parentcb5a047e5d111a6504bfb355f8494a6a8abe9118 (diff)
downloadvolse-hubzilla-7b173a75e47cfdf9f46b7d635df84b2be286b0e6.tar.gz
volse-hubzilla-7b173a75e47cfdf9f46b7d635df84b2be286b0e6.tar.bz2
volse-hubzilla-7b173a75e47cfdf9f46b7d635df84b2be286b0e6.zip
correct the mastodon "boost" (aka 'share') author attribution by checking for share activities and pulling the original author info from the activity:object
Diffstat (limited to 'include/feedutils.php')
-rw-r--r--include/feedutils.php139
1 files changed, 85 insertions, 54 deletions
diff --git a/include/feedutils.php b/include/feedutils.php
index d1985649d..477685f9a 100644
--- a/include/feedutils.php
+++ b/include/feedutils.php
@@ -254,81 +254,124 @@ function get_atom_elements($feed, $item, &$author) {
else
$base_url = '';
- // look for a photo. We should check media size and find the best one,
- // but for now let's just find any author photo
- $rawauthor = $item->get_item_tags(SIMPLEPIE_NAMESPACE_ATOM_10,'author');
+ $rawverb = $item->get_item_tags(NAMESPACE_ACTIVITY, 'verb');
- if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link']) {
- $base = $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
- foreach($base as $link) {
- if(!x($author, 'author_photo') || ! $author['author_photo']) {
- if($link['attribs']['']['rel'] === 'photo' || $link['attribs']['']['rel'] === 'avatar')
- $author['author_photo'] = unxmlify($link['attribs']['']['href']);
- }
- }
+ // select between supported verbs
+
+ if($rawverb) {
+ $res['verb'] = unxmlify($rawverb[0]['data']);
}
- $rawactor = $item->get_item_tags(NAMESPACE_ACTIVITY, 'actor');
+ // translate OStatus unfollow to activity streams if it happened to get selected
+
+ if((x($res,'verb')) && ($res['verb'] === 'http://ostatus.org/schema/1.0/unfollow'))
+ $res['verb'] = ACTIVITY_UNFOLLOW;
- if($rawactor && activity_match($rawactor[0]['child'][NAMESPACE_ACTIVITY]['obj_type'][0]['data'],ACTIVITY_OBJ_PERSON)) {
- $base = $rawactor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
- if($base && count($base)) {
- foreach($base as $link) {
- if($link['attribs']['']['rel'] === 'alternate' && (! $res['author_link']))
- $author['author_link'] = unxmlify($link['attribs']['']['href']);
- if(!x($author, 'author_photo') || ! $author['author_photo']) {
- if($link['attribs']['']['rel'] === 'avatar' || $link['attribs']['']['rel'] === 'photo')
- $author['author_photo'] = unxmlify($link['attribs']['']['href']);
- }
- }
- }
- }
- // check for a yahoo media element (github etc.)
- if(! $author['author_photo']) {
- $rawmedia = $item->get_item_tags(NAMESPACE_YMEDIA,'thumbnail');
- if($rawmedia && $rawmedia[0]['attribs']['']['url']) {
- $author['author_photo'] = strip_tags(unxmlify($rawmedia[0]['attribs']['']['url']));
+ if(array_key_exists('verb',$res) && $res['verb'] === ACTIVITY_SHARE) {
+ // For Mastodon shares ("boosts"), we need to parse the original author information
+ // from the activity:object -> author structure
+ $rawobj = $item->get_item_tags(NAMESPACE_ACTIVITY, 'object');
+
+ if($rawobj) {
+ $rawauthor = $rawobj->get_item_tags(SIMPLEPIE_NAMESPACE_ATOM_10,'author');
+ if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['name']) {
+ $author['author_name'] = unxmlify($rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['name']);
+ }
+
+ if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['uri']) {
+ $author['author_link'] = unxmlify($rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['uri']);
+ }
+ if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link']) {
+ $base = $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
+ foreach($base as $link) {
+ if(!x($author, 'author_photo') || ! $author['author_photo']) {
+ if($link['attribs']['']['rel'] === 'photo' || $link['attribs']['']['rel'] === 'avatar')
+ $author['author_photo'] = unxmlify($link['attribs']['']['href']);
+ }
+ }
+ }
}
}
+ else {
+ // look for a photo. We should check media size and find the best one,
+ // but for now let's just find any author photo
+ $rawauthor = $item->get_item_tags(SIMPLEPIE_NAMESPACE_ATOM_10,'author');
- // No photo/profile-link on the item - look at the feed level
-
- if((! (x($author,'author_link'))) || (! (x($author,'author_photo')))) {
- $rawauthor = $feed->get_feed_tags(SIMPLEPIE_NAMESPACE_ATOM_10,'author');
if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link']) {
$base = $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
foreach($base as $link) {
- if($link['attribs']['']['rel'] === 'alternate' && (! $author['author_link'])) {
- $author['author_link'] = unxmlify($link['attribs']['']['href']);
- $author['author_is_feed'] = true;
- }
- if(! $author['author_photo']) {
+ if(!x($author, 'author_photo') || ! $author['author_photo']) {
if($link['attribs']['']['rel'] === 'photo' || $link['attribs']['']['rel'] === 'avatar')
$author['author_photo'] = unxmlify($link['attribs']['']['href']);
}
}
}
- $rawactor = $feed->get_feed_tags(NAMESPACE_ACTIVITY, 'subject');
+ $rawactor = $item->get_item_tags(NAMESPACE_ACTIVITY, 'actor');
if($rawactor && activity_match($rawactor[0]['child'][NAMESPACE_ACTIVITY]['obj_type'][0]['data'],ACTIVITY_OBJ_PERSON)) {
$base = $rawactor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
-
if($base && count($base)) {
foreach($base as $link) {
if($link['attribs']['']['rel'] === 'alternate' && (! $res['author_link']))
$author['author_link'] = unxmlify($link['attribs']['']['href']);
- if(! (x($author,'author_photo'))) {
+ if(!x($author, 'author_photo') || ! $author['author_photo']) {
if($link['attribs']['']['rel'] === 'avatar' || $link['attribs']['']['rel'] === 'photo')
$author['author_photo'] = unxmlify($link['attribs']['']['href']);
}
}
}
}
+
+ // check for a yahoo media element (github etc.)
+
+ if(! $author['author_photo']) {
+ $rawmedia = $item->get_item_tags(NAMESPACE_YMEDIA,'thumbnail');
+ if($rawmedia && $rawmedia[0]['attribs']['']['url']) {
+ $author['author_photo'] = strip_tags(unxmlify($rawmedia[0]['attribs']['']['url']));
+ }
+ }
+
+
+ // No photo/profile-link on the item - look at the feed level
+
+ if((! (x($author,'author_link'))) || (! (x($author,'author_photo')))) {
+ $rawauthor = $feed->get_feed_tags(SIMPLEPIE_NAMESPACE_ATOM_10,'author');
+ if($rawauthor && $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link']) {
+ $base = $rawauthor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
+ foreach($base as $link) {
+ if($link['attribs']['']['rel'] === 'alternate' && (! $author['author_link'])) {
+ $author['author_link'] = unxmlify($link['attribs']['']['href']);
+ $author['author_is_feed'] = true;
+ }
+ if(! $author['author_photo']) {
+ if($link['attribs']['']['rel'] === 'photo' || $link['attribs']['']['rel'] === 'avatar')
+ $author['author_photo'] = unxmlify($link['attribs']['']['href']);
+ }
+ }
+ }
+
+ $rawactor = $feed->get_feed_tags(NAMESPACE_ACTIVITY, 'subject');
+
+ if($rawactor && activity_match($rawactor[0]['child'][NAMESPACE_ACTIVITY]['obj_type'][0]['data'],ACTIVITY_OBJ_PERSON)) {
+ $base = $rawactor[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['link'];
+
+ if($base && count($base)) {
+ foreach($base as $link) {
+ if($link['attribs']['']['rel'] === 'alternate' && (! $res['author_link']))
+ $author['author_link'] = unxmlify($link['attribs']['']['href']);
+ if(! (x($author,'author_photo'))) {
+ if($link['attribs']['']['rel'] === 'avatar' || $link['attribs']['']['rel'] === 'photo')
+ $author['author_photo'] = unxmlify($link['attribs']['']['href']);
+ }
+ }
+ }
+ }
+ }
}
$ostatus_protocol = (($item->get_item_tags(NAMESPACE_OSTATUS,'conversation')) ? true : false);
@@ -492,18 +535,6 @@ function get_atom_elements($feed, $item, &$author) {
$res['coord'] = unxmlify($rawgeo[0]['data']);
- $rawverb = $item->get_item_tags(NAMESPACE_ACTIVITY, 'verb');
-
- // select between supported verbs
-
- if($rawverb) {
- $res['verb'] = unxmlify($rawverb[0]['data']);
- }
-
- // translate OStatus unfollow to activity streams if it happened to get selected
-
- if((x($res,'verb')) && ($res['verb'] === 'http://ostatus.org/schema/1.0/unfollow'))
- $res['verb'] = ACTIVITY_UNFOLLOW;
$cats = $item->get_categories();
if($cats) {