This night we had a hardware upgrade which added additional CPU cores to our web server to mitigate the poor response time that has been plaguing us for some time.

Based on preliminary readings of our statistics the upgrade was a success and response time during peak hours should now be much better.

Today's update (e49fef5) brings with it the following changes:

Performance improvements

• The folder menu template (element_menu) has been rewritten to use the new fs2tree template function, which should speed it up quite a bit.
• Display of forum comments should be faster because the backend database structure has been significantly improved.

New features

• A new template function called fs2tree has been added, which is used to convert the array from filesystem template function into a tree structure, which should be useful for template designers.
• The translation languages Amharic, Oromo and Swahili has been added.

Enhancements

• On Safari (Mac/iOS) autocorrection and autocapitalization was done to the quiz type text, which was unfortunate, as it caused a lot of typing errors which could've been avoided. This behavior has been disabled.
• The HTML layout of the filesystem request handler has been improved by using the new fs2tree template function.
• Virtualhost and response time has been added to backend web server logs, allowing better monitoring.
• The commenting feature was redesigned to avoid a circular foreign-key constraint, allowing improved database backup performance.
• The database backup script is now more automation-friendly.

Bugfixes

• SoundManager was upgraded to version 2.97a.20150601, which should fix the HTML5 MP3 playback issue on iOS/Safari.
• An internal error when trying display a message without a sender was fixed.
• Tables were missing labels if user interface language was set to Arabic. This has been fixed.
• A file-not-found error was triggered if the incomplete Spanish user interface language was used. This has been fixed.

Because of an unexpected network issue with our backend network the servers were down from 14:49 to 15:52 (Europe/Oslo timezone). Our data center provider was able to resolve the issue and services are now operational again.

Today's update (4920618) brings with it the following changes:

Performance improvements

• The user search now uses an indexed search method, making it much faster. This has a side-effect that pure substring search is no longer possible. You can now only search on full tokens (like the full name or email).

New features

• The read request handler and embed template function are now history-aware. This feature enables the history request handler to show the historic content of objects that use data request handler to refer to the content, like images, videos and other binary content. The thumbnail, watermark and wav2mp3 request handlers have also been made history-aware.
  • The following object types don't support relative file references (because they use resolve_path) when viewed using history:
    • Flash apps
    • JavaScript apps and scripts
    • SVG images
    • Perl scripts
    • CSS stylesheets
    • Text files
    • XML files
  • Java apps don't support history viewing at all (because they use resolve_path which doesn't support history).
  • Zip files can't be unpacked from history, they must be restored first. You can download them, though.
• The embed template function now supports render_mode => "source" on JavaScript and CSS.

Enhancements

• It is now possible to restore a deleted object or container into the correct parent container. The parent container must exist, or you'll get a database error during restore. Be aware that the parent container must be restored before the child object or container, or you'll get a database error about missing parent. This is most likely to occur when trying to restore a recursive delete.
• Some flash info messages are now considered redundant and are not shown during normal usage.

Bugfixes

• Small change in behavior. If embed template function is called with an odd number of options the options variable is now empty instead of cut at the last missing key/value pair. This should make the error in API usage more obvious. This could potentially trigger some regressions in content that was previously working.
• The user search feature wasn't able to search for firstname lastname after a recent update. This has now been fixed.
• A regression was found in the matchbox quiz type, which could return invalid data from the database when used together with the history-aware feature.
• The links to Bokmålsordboka and Nynorskordboka have been updated.
• Whitespace is now trimmed when using the history request handler.
• Backend importing of content was not possible because the cost and copyright boolean object attributes where not sent properly to the database. This has now been fixed.

Today's update (0b662f0) brings with it the following changes:

Performance improvements

• Object content is now stored in an external service called BlobMachine, making it possible to scale to multiple web servers.
• All objects are stored named by the digest checksum of the content, improving caching because identical files only need to be read once. This feature also improves the transactional behavior during object modification.

New features

• Whenever objects are modified the old version of content and metadata is stored, allowing quick restoration of content back to a previous version. Use the history button in the toolbar to access this feature. The history is kept for 6 weeks.
• It's also possible to restore deleted containers and objects, but because of a bug they are restored to the root instead of their proper location. When restoring content on an existing object or container the parent is maintained. This issue will be fixed in a future update.

Enhancements

• Cache files are now properly named after the digest checksum, which allows them to be created only once, becaues they never change without the name also changing.
• The cache directory is regularly cleaned of old files, recovering disk space more efficiently.
• The timeout in thumbnail(r) was increased to account for the increased time it takes to fetch videos from the blob service.

Bugfixes

• The thumbnail request handler was unable to generate thumbnails for SVG images. Changed to return the SVG image directly when asked for a thumbnail of it. It should be able to scale to the needed size because it's a vector image.