- now the currentVersions are queried later, because only after the
realm handler is started, can we query the current version.
- this lead to only parsing the migrations at initialize
- and thus in start querying the versions and performing the required
migrations
- If a code migration is run programmatically, then in some migrations
shouldn't fail if a realm is missing -> the realm might not be available
in a certain environment
- We have to re-think this. It does not work, throwing an exception if
commands are registered on a read-only TX as then we don't know if we
want to roll back or not - we probably need a ROLLBACK_ON_ERROR or
something, indicating that we are preparing a non-read-only TX.
- Now if you need to perform commands to carry on in your transaction,
you can simply use the tx.flush()-method.
- Should something go wrong, then even after a tx.flush() one can call
throw tx.fail("Reason") or tx.rollbackOnClose()
- So sadly just auto closing a TX using try-resource from Java7 is a bad idea.
- Doing that leads to problems when an exception is thrown, then the close is called (duh) but this leads to commit being called.
- Since the Java language does not offer a decent way to detect if the close is being called in the context of an exception i was forced to add a tx.commitOnClose() and tx.rollbackOnClose().
- The default is that when a TX is opened, then the close strategy is rollback; the API user must call tx.commitOnClose() before the TX is closed by the braces, or as late as possible, to make sure that if an exception is thrown the transaction is rolled back, and not committed.
- The API was also extended with a tx.fail(msg):StrolchTransactionException so that if the implementor detects an unrecoverable error, one can write: throw tx.fail(“my reason”);
This was sadly an unavoidable late 1.0.0 change
When a TX is closed, all locked objects are unlocked. The but originated from multiple commands being performed for the same object, thus the object was locked multiple times i.e. the lock counter was >1. Now added a releaseLock() method which is called by the TX when the TX is closed so that the lock counter is really 0 and the lock is released.
- Added tests for this situation
- documented the LockHandler
- Now we can use Version to see if current version is greater than
- Fixed bug where version checking didn't check for app
- TODO is still open to handle migration with intermediary steps
Robert von Burg
msmock