As I can't copy-paste code here as our application uses Connector/J inside a complicated self-written framework, I'll describe the reproduction steps:
- Create a table someTable with a field foo type longtext
- Use a prepared statement with a simple query like INSERT INTO someTable (foo) VALUES ( ? )
- Generate random strings with below code:
- Use this to generate strings length 8000000 (8 million) doublequotes and 10000000 (10 million) doublequotes
- Check the results (SELECT LENGTH(foo) FROM someTable)
You'll see (or at least I see) that 8000000 inserts a string length 8000000 (how it should be), but the 10000000 inserts a string half its length: 5000000.
I think I also found the problem and the solution, so I will add a pull request.
In our case (where we insert very large JSON strings with a lot of doublequotes) this makes us lose data and generate invalid JSON strings. This does not happen in 1.5.9, but does happen in 1.6.x and 2.0.x as far as I have tested.
This is a critical bug as this results in invalid and lost data.