UserActivityStream -- used to create a full activity stream including subscriptions, favorites, notices, etc -- normally buffers everything into memory at once. This is infeasible for accounts with long histories of serious usage; it can take tens of seconds just to pull all records from the database, and working with them all in memory is very likely to hit resource limits.
This commit adds an alternate mode for this class which avoids pulling notices until during the actual output. Instead of pre-sorting and buffering all the notices, empty spaces between the other activities are filled in with notices as we're making output. This means more smaller queries spread out during operations, and less stuff kept in memory.
Callers (backupaccount action, and backupuser.php) which can stream their output pass an $outputMode param of UserActivityStream::OUTPUT_RAW, and during getString() it'll send straight to output as well as slurping the notices in this extra funky fashion.
Other callers will let it default to the OUTPUT_STRING mode, which keeps the previous behavior.
There should be a better way to do this, swapping out the stringer output for raw output more consitently.
http://status.net/open-source/issues/2442
Notes:
* Mapstraction causes JavaScript errors in XHTML mode, breaking our code if we're run later so the link doesn't work to get back to Desktop.
* not 100% sure how safe feature detection is here?
* Currently will be useless but visible links if no JS available; need to fall back to server-side for limited browsers
Regressions caused by bad refactoring in commit 21feac3bea.
Test cases in tests/CommandInterpreterTest.php were made against the pre-refactoring code, and now check out with the fixed code.
Failures were caused by not changing logic structure when moving from multiple exit points (each if point would return directly with a null or an object) to setting a result variable and then falling through to a common exit point. Without the if statements being restructured, the result variable would just get overridden by the next case.
Our mailbox actions (inbox and outbox) were doing their own display of
messages. This was causing issues with especially showmessage, which
since the more rigourous nickname checks were added, no longer works as
a mailbox subclass.
I've taken the time to rip out the message listing code from MailboxAction
and moved it to a MessageList widget. The different mailboxes now have their
own subclasses that show the correct profile in the list.
This entry was using the strings 'true' and 'false' instead of literal booleans, which could confuse clients expecting literal booleans as in other places and on Twitter in this place.
Format's already available as a member variable, so use it!
Fixes some error reponses in api/statusnet/groups/leave.json which were coming through as XML.
May fix some others as well.
$config['site']['logperf'] = true; // to record & dump total hits of each type and the runtime to syslog
$config['site']['logperf_detail'] = true; // very verbose -- dump the individual cache keys and queries as they get used (may contain private info in some queries)
Seeing 180 cache gets on a timeline page seems not unusual currently; since these run in serial, even relatively small roundtrip times can add up heavily.
We should consider ways to reduce the number of round trips, such as more frequently storing compound objects or the output of processing in memcached.
Doing parallel multi-key lookups could also help by collapsing round-trip times, but might not be easy to fit into SN's object model. (For things like streams this should actually work pretty well -- grab the list, then when it's returned go grab all the individual items in parallel and return the list)
* dropped unnecessary join on notice table
* made the function actually static, since it makes no sense as an instance variable. The only caller (in AttachmentList) is updated.
In order to apply to PHP's POST processing, the MAX_FILE_SIZE field must appear *before* the file upload field. They were incorrectly placed after, where they had no effect on POST processing.
Part of the reported issue was previuosly fixed by dc497ed0 (smaller size images being blanked).
This commit fixes the remaining bug with original-size avatars being left as BMP (which could include the 96px size for instance, which could cause problems in browsers not supporting BMP natively)
Added ImageFile::copyTo() as a convenient alias for resizeTo() when not resizing; this performs the BMP/XPM/XBM->PNG conversion if needed, or copies the original file.
Copying instead of using move_uploaded_file() is fine here since:
a) the files are cleaned up on script completion anyway (vs moving to remove it)
b) we're already performing getimagesize() and possibly load/resize on the file before this point (vs needing to move the file into a usable area to work with open_basedir restrictions that prevent working directly with uploaded files in the temp dir; since this would fail anyway, we lose nothing)
ImageFile::preferredType() now works on $this->type instead of asking for one, to make it handier to use from outside. (This is still needed in order for calling code to generate a target filename.)
Recommended for future:
* additional consolidation between the various ways of uploading avatars (touched avatarsettings, grouplogo, and apiaccountupdateprofileimage with similar minor changes)
* consolidate type checks and file naming into Avatar class
