Loading Google client services is a little monster, eating around
2MB (opcache enabled) of memory. As far as we instantiate the
repository instances really early, no matter they are not used
later (editor, file picker, admin...).. we are delaying the
load of the service until we know we are going to use it.
Surely applying this very same (sort of lazy load) techinque
to other repositories could lead to a nice memory reduction
in lots of pages.
MDL-50176 repository_youtube: also delay the inclusion of client.
While memory was fixed with previous commit, still some extra included
files where being reported, so go crazy and move all the stuff to
the new init delayed method.
As of April 20, 2015, the old API used by repository_youtube is no longer
accessible. This change switches over to using the standard Google Data
API v3.
Unfortunately this means that to search YouTube, a developer api key is now
*required*.
Special thanks to UNSW for sponsoring this work.
- extension was returned without leading dot which resulted in improper validation
- fixed human-readable mime type in error message
- fixed youtube repository to support video files and pass filetype validation
repositories wikimedia and youtube:
- save last search in
- process parameter 'page'
- remove buttons 'Refresh', 'Search' and 'Logout' since they don't have meaning
filepicker:
- allow unknown number of pages (['pages'] = -1), show one 'Next' page after the current (both JS and non-JS modes)
repository wikimedia:
- add non-AJAX search form to make it work with JS disabled
Eloy Lafuente (stronk7)