Can be very efficient with the right input, but probably has
degenerative cases where it increases the size. Worst-case size
increase is one character per ID in the result set, so it shouldn't be
an issue.
String lengths for a large search result that was unstorable before:
serialize: 1991874
json_encode: 756249
cooked: 244
And some with more discontinuity:
serialize: 1772238
json_encode: 674751
cooked: 37950
serialize: 164942
json_encode: 65771
cooked: 25369
http://php.net/manual/en/function.session-set-save-handler.php:
When using objects as session save handlers, it is important
to register the shutdown function with PHP to avoid unexpected
side-effects from the way PHP internally destroys objects on
shutdown and may prevent the write and close from being called.
Typically you should register 'session_write_close' using the
register_shutdown_function() function.
We're not using objects as session save handlers, but some people
(notably, Synology users) seem to be encountering a similar issue
related to the timing of object destruction. Closing the session
earlier in the shutdown process is a perfectly sane thing to do, so
let's do it.
InnoDB's performance is unacceptable out of the box, and there's no
clean way to batch these insertions into transactions, and stuff.
Benchmarks for 1000 inserts:
InnoDB: 54.826
MyISAM: 0.091
InnoDB transaction: 0.159
Fixes GH #14
Include the correct type parameter in the URL and rename the random
type parameter to random_type.
Remove unused parameters from the function call.
Should fix issue #11
Handling of unknown tags was wrong (we should just pass them through and
hope for the best, not do some weird-ass id3v2 thing). Add matroska to
the list of known tags. Switch the video types to a generic cleanup
function and move the several-times-duplicated general information
gathering into the general information gathering function where it
belongs. Treat the general information as another tag type instead of
special-casing it as a complete override through array_merge. Drop the
useless iconv stuff (getID3 already translates tags to our requested
charset). Rename some functions to more closely match their purpose.
Fix some private functions that were marked public. Fix encoding
detection, which has been completely broken for a while.
Code duplication is bad even when it's CSS, mmkay?
If a theme has templates/rtl.css, it will be loaded in addition to
default.css. This is more sane than the old behaviour, which loaded
default-rtl.css instead of default.css.
I'm not feeling ambitious enough to actually check whether the current
-rtl.css files work and change them to overrides instead of modified
copypasta, so I just dropped them for now.
Remove some of the roundabout complexity that had built up. Push people
toward using a single, flexible tool for most of their transcoding
needs. Increase backend and configuration flexibility to support
user-requested format changes (e.g. for an HTML5 player); this
functionality is not yet exposed.
Now that playback doesn't pretend it's supported, we don't need the
fragile hacks to support guessing the content size or 'seeking' within
a transcoded stream.
