Mixture of the NewParserTests.php with old code present in MediaWikiParserTest.php, reverting r79184 (revert of a revert...) and even eval().
Using the iterator as dataProvider, but defering the insertion of articles, as the dataprovider is processed on load, before switching dbs.
Each parser test appear now as a phpunit test. This means that by adding 683 tests (partly from extensions) we now surpass two thousand tests.
The downside of that is that they become painfully slow.
Initialise $wgHooks to $wgHooks, not to array() (r82499) as otherwise we lose ParserFirstCallInit registrations.
There's no attempt to support extensions registering to $wgParser instead of using ParserFirstCallInit properly (Cite falls in this category).
As Setup.php assigns variables based on the cache config, bootstrap.php was late on reseting them, as some objects were already created.
So we could end up with a SqlBagOStuff created there, which when later accessed (such as trying to invalidate the cache for a user) would
-as any non-sqlite SqlBagOStuff- open a new db connection. Which is precisely what we shall not be done when dealing with temporary tables
(and would indeed fail miserably due to not finding unittest_objectcache table).
In summary, reenabling temporary tables disabled in r79411.
MediaWikiTestCase duplicates code from PHPUnit constructor. It worked
fine until PHPUnit 3.5.8 made FrameWork_TestCase attributes private,
thus the children class is no more able to update them ($data for example)
The change break any tests making use of the @dataProvider helper.
PHPUnit commit:
0b85d08c39
TESTS:
3.5.7 : OK -> OK
3.5.8 : KO -> OK
3.5.10 : KO -> OK
make databaseless
Tests: 486, Assertions: 73986, Incomplete: 1, Skipped: 3.
(skipped tests are SQLite related)
-No need to specify a new function for each argument, use PHP5 magic to automatically set it
-Create (g|s)etCliArg() functions
-Implement some features from the old parser tests
