* put the tables in the right order
* JOIN on user_text as well for better index usage
* Force the index on the recentchanges table as well
* Don't ORDER BY rev_user_text if it's constant
* This means queries could possibly return fewer results than the limit and still set a query-continue
* Add iicontinue, rvcontinue, cicontinue, incontinue, amfrom to faciliate query-continue for these modules
* Implemented by blocking additions to the ApiResult object if they would make it too large
** Important things like query-continue values and warnings are exempt from this check
** RSS feeds and exported XML are also exempted (size-checking them would be too messy)
** Result size is checked against $wgAPIMaxResultSize, which defaults to 8 MB
For those who really care, per-file details follow:
ApiResult.php:
* Introduced ApiResult::$mSize which keeps track of the result size.
* Introduced ApiResult::size() which calculates an array's size
(which is the sum of the strlen()s of its elements).
* ApiResult::addValue() now checks that the result size stays below
$wgAPIMaxResultSize. If the item won't fit, it won't be added and addValue()
will return false. Callers should check the return value and set a
query-continue if it's false.
* Closed the back door that is ApiResult::getData(): callers can't manipulate
the data array directly anymore so they can't bypass the result size limit.
* Added ApiResult::setIndexedTagName_internal() which will call
setIndexedTagName() on an array already in the result. This is needed for the
'new' order of adding results, which means addValue()ing one result at a time
until you hit the limit or run out, then calling this function to set the tag
name.
* Added ApiResult::disableSizeCheck() and enableSizeCheck() which disable and
enable size checking in addValue(). This is used for stuff like query-continue
elements and warnings which shouldn't count towards the result size.
* Added ApiResult::unsetValue() which removes an element from the result and
decreases $mSize.
ApiBase.php:
* Like ApiResult::getData(), ApiBase::getResultData() no longer returns a
reference.
* Use ApiResult::disableSizeCheck() in ApiBase::setWarning()
ApiQueryBase.php:
* Added ApiQueryBase::addPageSubItem(), which adds page subitems one item
at a time.
* addPageSubItem() and addPageSubItems() now return whether the subitem
fit in the result.
* Use ApiResult::disableSizeCheck() in setContinueEnumParameter()
ApiMain.php:
* Use ApiResult::disableSizeCheck() in ApiMain::substituteResultWithError()
* Use getParameter() rather than $mRequest to obtain requestid
DefaultSettings.php:
* Added $wgAPIMaxResultSize, with a default value of 8 MB
ApiQuery*.php:
* Added results one at a time, and set a query-continue if the result is full.
ApiQueryLangLinks.php and friends:
* Migrated from addPageSubItems() to addPageSubItem(). This eliminates the
need for $lastId.
ApiQueryAllLinks.php, ApiQueryWatchlist.php, ApiQueryAllimages.php, ApiQuerySearch.php:
* Renamed $data to something more appropriate ($pageids, $ids or $titles)
ApiQuerySiteinfo.php:
* Abuse siprop as a query-continue parameter and set it to all props that
couldn't be processed.
ApiQueryRandom.php:
* Doesn't do continuations, because the result is supposed to be random.
* Be smart enough to not run the second query if the results of the first
didn't fit.
ApiQueryImageInfo.php, ApiQueryRevisions.php, ApiQueryCategoryInfo.php, ApiQueryInfo.php:
* Added continue parameter which basically skips the first so many items
ApiQueryBacklinks.php:
* Throw the result in a big array first and addValue() that one element at a time if necessary
** This is necessary because the results aren't retrieved in order
* Introduced $this->pageMap to map namespace and title to page ID
* Rewritten extractRowInfo() and extractRedirRowInfo() a little
* Declared all private member variables explicitly
ApiQueryDeletedrevs.php:
* Use a pagemap just like in Backlinks
* Introduce fake page IDs and keep track of them so we know where to add what
** This doesn't change the output format, because the fake page IDs start at 0 and are consecutive
ApiQueryAllmessages.php:
* Add amfrom to facilitate query-continue
ApiQueryUsers.php:
* Rewrite: put the getOtherUsersInfo() code in execute()
Doxygen documentation update:
* Changed alls @addtogroup to @ingroup. @addtogroup adds the comment to the group description, but doesn't add the file, class, function, ... to the group like @ingroup does. See for example http://svn.wikimedia.org/doc/group__SpecialPage.html where it's impossible to see related files, classes, ... that should belong to that group.
* Added @file to file description, it seems that it should be explicitely decalred for file descriptions, otherwise doxygen will think that the comment document the first class, variabled, function, ... that is in that file.
* Removed some empty comments
* Removed some ?>
Added following groups:
* ExternalStorage
* JobQueue
* MaintenanceLanguage
One more thing: there are still a lot of warnings when generating the doc.
* Adding ApiQueryBase::addJoinConds() as wrapper for Database::select()'s $join_conds parameter
* Migrating query modules to addJoinConds()
* Using implicit join rather than INNER JOIN in ApiQueryBacklinks
* Using FORCE INDEX (times) on logging table in ApiQueryLogEvents; although MySQL 4 seems to pick this index automatically (evidenced by the fact the WMF servers are still alive), MySQL 5 doesn't and filesorts
* Replacing LEFT JOIN with implicit (inner) join in ApiQueryContributions: revisions without a corresponding page table entry shouldn't be shown anyway
* Added ApiQueryBase::addJoin() which provides a cleaner interface to construct JOIN queries. Behind the scenes this still uses the old, ugly way, but it'll be easy to rewrite when/if the Database class gets its own function for JOINs
* Used addJoin() in query modules where necessary
* Removed FORCE INDEX (rc_timestamp) from ApiQueryRecentchanges: it's nigh impossible to integrate with addJoin() and it doesn't seem to be necessary anyway (my MySQL instance automatically chooses rc_timestamp)
Using the user_timestamp index means you can't do an indexed search on the name prefix...
Basically the same as scanning raw on rev_timestamp, but with logged-in edits skipped so you don't scan *quite* as huge an amount of the table.
Since there's no index for the username prefix, the more specific the prefix, the *more* rows you'll have to scan through to fill out the result set, so the worse the performance will get.
* ucuserprefix now only works for anonymous users. This allows us to add WHERE rev_user=0 (fewer rows to scan) and use the user_timestamp index (so the ORDER BY rev_user_text DESC can be killed)
* ucuserprefix must be an IP containing at least two full octets, e.g. '12.345.', to ensure the resulting LIKE is somewhat sane
* Replaced $wgAPIUCUserPrefixMinLength with the more generic $wgAPIMaxDBRows
* Added ApiBase::checkRowCount() which checks whether the amount of rows to be scanned is acceptable (i.e. <$wgAPIMaxDBRows). Not using this anywhere (yet?), but it's nice to have
* Killed a filesort in the usercontribs query, query is now indexed nicely
* Dropped the minimum length for ucuserprefix since it's no longer needed (query optimized)
* Removed drnamespace from list=deletedrevs (filesorts 8M rows for drnamespace=0)
* Support multiple orderings in ApiBase::addWhereRange()
