is it possible to execute an undecidable off-chain loop, and attack the network in this way? taking into account that the off-chain code is flat Haskell code is also complete turing.
The off-chain code is executed by a single node to construct the transaction which is then disseminated to the other nodes. I think it is indeed possible to keep this node busy with a never ending loop in the off-chain code. The thing is that - to my knowledge - there are no public entry points for the nodes to execute arbitrary off-chain code. You must always run the off-chain code on a specific node you have access to and which exposes only a defined set of operations. So basically you can only attack yourself.