diff options
author | marijus <mario@mariovavti.com> | 2014-11-03 11:19:59 +0100 |
---|---|---|
committer | marijus <mario@mariovavti.com> | 2014-11-03 11:19:59 +0100 |
commit | 1f1b6785cd1e3b03c3de6277646d4fa2f9018c5a (patch) | |
tree | 3fd4ff213586e089d8fd7c0c9f77ef8c47ea5f54 /include | |
parent | 287d5e0f106eed3783ee120bafc338e7763e36e9 (diff) | |
parent | ac858f4bd4fcb9c65cbc7b19288cdad0076aa072 (diff) | |
download | volse-hubzilla-1f1b6785cd1e3b03c3de6277646d4fa2f9018c5a.tar.gz volse-hubzilla-1f1b6785cd1e3b03c3de6277646d4fa2f9018c5a.tar.bz2 volse-hubzilla-1f1b6785cd1e3b03c3de6277646d4fa2f9018c5a.zip |
Merge branch 'master' of https://github.com/friendica/red
Diffstat (limited to 'include')
-rw-r--r-- | include/Import/Importer.php | 94 | ||||
-rw-r--r-- | include/Import/refimport.php | 282 |
2 files changed, 376 insertions, 0 deletions
diff --git a/include/Import/Importer.php b/include/Import/Importer.php new file mode 100644 index 000000000..c42344236 --- /dev/null +++ b/include/Import/Importer.php @@ -0,0 +1,94 @@ +<?php /** @file */ + +namespace Redmatrix\Import; + + +class Import { + + $credentials = null; + $itemlist = null; + $src_items = null; + + $items = null; + + function get_credentials() { + + } + + function get_itemlist() { + + + } + + + function get_item_ident($item) { + + } + + function get_item($item_ident) { + + } + + function get_taxonomy($item_ident) { + + } + + function get_children($item_ident) { + + } + + function convert_item($item_ident) { + + } + + function convert_taxonomy($item_ident) { + + + } + + function convert_child($child) { + + } + + function store($item,$update = false) { + + } + + function run() { + + $this->credentials = $this->get_credentials(); + $this->itemlist = $this->get_itemlist(); + if($this->itemlist) { + $this->src_items = array(); + $this->items = array(); + $cnt = 0; + foreach($this->itemlist as $item) { + $ident = $item->get_item_ident($item); + $this->src_items[$ident]['item'] = $this->get_item($ident); + $this->src_items[$ident]['taxonomy'] = $this->get_taxonomy($ident); + $this->src_items[$ident]['children'] = $this->get_children($ident); + $this->items[$cnt]['item'] = $this->convert_item($ident); + $this->items[$cnt]['item']['term'] = $this->convert_taxonomy($ident); + if($this->src_items[$ident]['children']) { + $this->items[$cnt]['children'] = array(); + foreach($this->src_items[$ident]['children'] as $child) { + $this[$cnt]['children'][] = $this->convert_child($child); + } + } + $cnt ++; + } + + + + + } + + + + + + + } + + +}
\ No newline at end of file diff --git a/include/Import/refimport.php b/include/Import/refimport.php new file mode 100644 index 000000000..4f2572660 --- /dev/null +++ b/include/Import/refimport.php @@ -0,0 +1,282 @@ +<?php + +require_once('include/html2bbcode.php'); +require_once('include/hubloc.php'); + +// Sample module for importing conversation data from Reflection CMS. Some preparation was used to +// dump relevant posts, categories and comments into individual JSON files, and also JSON dump of +// the user table to search for avatars. Importation was also batched in sets of 20 posts per page +// visit so as to survive shared hosting process limits. This provides some clues as how to handle +// WordPress imports, which use a somewhat similar DB structure. The batching and individual files +// might not be needed in VPS environments. As such this could be considered an extreme test case, but +// the importation was successful in all regards using this code. The module URL was visited repeatedly +// with a browser until all the posts had been imported. + + + + +define('REDMATRIX_IMPORTCHANNEL','mike'); +define('REFLECT_EXPORTUSERNAME','mike'); +define('REFLECT_BLOGNAME','Diary and Other Rantings'); +define('REFLECT_BASEURL','http://example.com/'); +define('REFLECT_USERFILE','user.json'); + +// set to true if you need to process everything again +define('REFLECT_OVERWRITE',false); + +// we'll only process a small number of posts at a time on a shared host. + +define('REFLECT_MAXPERRUN',30); + +function reflect_get_channel() { + + // this will be the channel_address or nickname of the red channel + + $c = q("select * from channel left join xchan on channel_hash = xchan_hash + where channel_address = '%s' limit 1", + dbesc(REDMATRIX_IMPORTCHANNEL) + ); + return $c[0]; +} + + +function refimport_content(&$a) { + + $channel = reflect_get_channel(); + + // load the user file. We need that to find the commenter's avatars + + $u = file_get_contents(REFLECT_USERFILE); + if($u) { + $users = json_decode($u,true); + } + + $ignored = 0; + $processed = 0; + + $files = glob('article/*'); + if(! $files) + return; + + foreach($files as $f) { + $s = file_get_contents($f); + $j = json_decode($s,true); + + if(! $j) + continue; + + $arr = array(); + + // see if this article was already processed + $r = q("select * from item where mid = '%s' and uid = %d limit 1", + dbesc($j['guid']), + intval($channel['channel_id']) + ); + if($r) { + if(REFLECT_OVERWRITE) + $arr['id'] = $r[0]['id']; + else { + $ignored ++; + rename($f,str_replace('article','done',$f)); + continue; + } + } + + $arr['uid'] = $channel['channel_account_id']; + $arr['aid'] = $channel['channel_id']; + $arr['mid'] = $arr['parent_mid'] = $j['guid']; + $arr['created'] = $j['created']; + $arr['edited'] = $j['edited']; + $arr['author_xchan'] = $channel['channel_hash']; + $arr['owner_xchan'] = $channel['channel_hash']; + $arr['app'] = REFLECT_BLOGNAME; + $arr['item_flags'] = ITEM_ORIGIN|ITEM_WALL|ITEM_THREAD_TOP; + $arr['verb'] = ACTIVITY_POST; + + // this is an assumption + $arr['comment_policy'] = 'contacts'; + + + // import content. In this case the content is XHTML. + + $arr['title'] = html2bbcode($j['title']); + $arr['title'] = htmlspecialchars($arr['title'],ENT_COMPAT,'UTF-8',false); + + + $arr['body'] = html2bbcode($j['body']); + $arr['body'] = htmlspecialchars($arr['body'],ENT_COMPAT,'UTF-8',false); + + + // convert relative urls to other posts on that service to absolute url on our service. + $arr['body'] = preg_replace_callback("/\[url\=\/+article\/(.*?)\](.*?)\[url\]/",'reflect_article_callback',$arr['body']); + + // also import any photos + $arr['body'] = preg_replace_callback("/\[img(.*?)\](.*?)\[\/img\]/",'reflect_photo_callback',$arr['body']); + + + // add categories + + if($j['taxonomy'] && is_array($j['taxonomy']) && count($j['taxonomy'])) { + $arr['term'] = array(); + foreach($j['taxonomy'] as $tax) { + $arr['term'][] = array( + 'uid' => $channel['channel_id'], + 'type' => TERM_CATEGORY, + 'otype' => TERM_OBJ_POST, + 'term' => trim($tax['name']), + 'url' => $channel['xchan_url'] . '?f=&cat=' . urlencode(trim($tax['name'])) + ); + } + } + + // store the item + + if($arr['id']) + item_store_update($arr); + else + item_store($arr); + + // if there are any comments, process them + // $comment['registered'] is somebody with an account on the system. Others are mostly anonymous + + if($j['comments']) { + foreach($j['comments'] as $comment) { + $user = (($comment['registered']) ? reflect_find_user($users,$comment['author']) : null); + reflect_comment_store($channel,$arr,$comment,$user); + } + } + $processed ++; + + if(REFLECT_MAXPERRUN && $processed > REFLECT_MAXPERRUN) + break; + } + return 'processed: ' . $processed . EOL . 'completed: ' . $ignored . EOL; + +} + +function reflect_article_callback($matches) { + return '[zrl=' . z_root() . '/display/'. $matches[1] . ']' . $matches[2] . '[/zrl]'; +} + +function reflect_photo_callback($matches) { + + if(strpos($matches[2],'http') !== false) + return $matches[0]; + + $prefix = REFLECT_BASEURL; + $x = z_fetch_url($prefix.$matches[2],true); + + $hash = basename($matches[2]); + + if($x['success']) { + $channel = reflect_get_channel(); + require_once('include/photos.php'); + $p = photo_upload($channel,$channel, + array('data' => $x['body'], + 'resource_id' => str_replace('-','',$hash), + 'filename' => $hash . '.jpg', + 'type' => 'image/jpeg', + 'not_visible' => true + ) + ); + + if($p['success']) + $newlink = $p['resource_id'] . '-0.jpg'; + + + // import photo and locate the link for it. + return '[zmg]' . z_root() . '/photo/' . $newlink . '[/zmg]'; + + } + // no replacement. Leave it alone. + return $matches[0]; +} + +function reflect_find_user($users,$name) { + if($users) { + foreach($users as $x) { + if($x['name'] === $name) { + return $x; + } + } + } + + return false; + +} + +function reflect_comment_store($channel,$post,$comment,$user) { + + // if the commenter was the channel owner, use their redmatrix xchan + + if($comment['author'] === REFLECT_EXPORTUSERNAME && $comment['registered']) + $hash = $channel['xchan_hash']; + else { + // we need a unique hash for the commenter. We don't know how many may have supplied + // http://yahoo.com as their URL, so we'll use their avatar guid if they have one. + // anonymous folks may get more than one xchan_hash if they commented more than once. + + $hash = (($comment['registered'] && $user) ? $user['avatar'] : ''); + if(! $hash) + $hash = random_string() . '.unknown'; + + // create an xchan for them which will also import their profile photo + // they will have a network type 'unknown'. + + $x = array( + 'hash' => $hash, + 'guid' => $hash, + 'url' => (($comment['url']) ? $comment['url'] : z_root()), + 'photo' => (($user) ? REFLECT_BASEURL . $user['avatar'] : z_root() . '/' . get_default_profile_photo()), + 'name' => $comment['author'] + ); + xchan_store($x); + + } + + $arr = array(); + + $r = q("select * from item where mid = '%s' and uid = %d limit 1", + dbesc($comment['guid']), + intval($channel['channel_id']) + ); + if($r) { + if(REFLECT_OVERWRITE) + $arr['id'] = $r[0]['id']; + else + return; + } + + // this is a lot like storing the post except for subtle differences, like parent_mid, flags, author_xchan, + // and we don't have a comment edited field so use creation date + + $arr['uid'] = $channel['channel_account_id']; + $arr['aid'] = $channel['channel_id']; + $arr['mid'] = $comment['guid']; + $arr['parent_mid'] = $post['mid']; + $arr['created'] = $comment['created']; + $arr['edited'] = $comment['created']; + $arr['author_xchan'] = $hash; + $arr['owner_xchan'] = $channel['channel_hash']; + $arr['item_flags'] = ITEM_ORIGIN|ITEM_WALL; + $arr['verb'] = ACTIVITY_POST; + $arr['comment_policy'] = 'contacts'; + + + $arr['title'] = html2bbcode($comment['title']); + $arr['title'] = htmlspecialchars($arr['title'],ENT_COMPAT,'UTF-8',false); + + + $arr['body'] = html2bbcode($comment['body']); + $arr['body'] = htmlspecialchars($arr['body'],ENT_COMPAT,'UTF-8',false); + $arr['body'] = preg_replace_callback("/\[url\=\/+article\/(.*?)\](.*?)\[url\]/",'reflect_article_callback',$arr['body']); + $arr['body'] = preg_replace_callback("/\[img(.*?)\](.*?)\[\/img\]/",'reflect_photo_callback',$arr['body']); + + // logger('comment: ' . print_r($arr,true)); + + if($arr['id']) + item_store_update($arr); + else + item_store($arr); + +} |