I still do not understand the whole side-tangle problem after reading the article by Dom. So this side-tangle starts by referencing some milestone and then grows further by adding txs which are only referencing their own sub-tangle txs.
Then it can happen that a walk (initiated by calling getTransactionsToApprove) ends up in this side-tangle (by stepping from that initial milestone from the maintangle into the sidetangle) and consequently has to validate a lot of txs. This consumes a lot of resources and results in the node freezing and getting out of sync. I can fully understand this.
So now v1.5.2 came up with a more sophisticated walk preventing this. But why doesn't the max-depth parameter already (available in versions < v1.5.2) prevent this from happening?
Can somebody explain where my reasoning fails?