It was accidentally passing a username where it should have passed a user ID, causing PostgreSQL's stricter comparisons to fail, while MySQL's allowed it to run without complaint but returned bad results.
Of course that bug got hidden by the test.... testing the wrong thing... :)
Now correctly loads using the user id instead of name, checks the proper return values, and actually compares the right object.
The ID might be different than the one you expect. Since r88755 the tables
are no more dropped and recreated, thus the autoincrement pointer is not
reset. A previous test could have raised that pointer, thus the id might
be above the expected value.
This patch track the block ID inserted and use it in the assertion.
Fixes regression from r84475 and friends which made Block->load() and its new front-end Block::newFromTarget() fail when an empty string was passed in as the IP / $vagueTarget parameter to indicate skipping IP-based lookups.
Added phpunit test cases to confirm that both Block->load() and Block::newFromTarget() work when given null (already ok), '' (as done from CheckUser), or false (not seen, but perfectly legit sounding).
Adjusted comparisons to work as expected.
Don't setup/destroy for databaseless tests.
Make needsDB() return based on the DocComment.
Delayed db init until the test is run, so that it isn't set for excluded tests (the exclusion is done after all the classes are instantiated).
