Again : Note, when I pu t the last insert statement to first there is no error.
The sql text has the same length, only command order changes and the BLR in 1st case is >64k in 2nd case smaller? That is strange.
I've committed a bugfix. But generally, I'd avoid such long blocks.
The issue is not that its BLR is bigger or smaller. It's about wrapping the BLR offset over the 64KB boundary (thus becoming kinda random) and either occasionally matching some value inside the debug info table or not. In other words, it's about pure luck.