Message11273

Impressive analysis Stefan. Not much I can add.

I suppose the finally clauses fail to operate because they need to allocate objects? It's almost as if a rul of successful finally clauses was that they should have all their object allocation done ahead of time. If once you run out of memory, all contracts enforced by finally are unreliable, the condition should not be handled too locally but propagate out to where all affected objects are dereferenced. But how can anyone know where that is?

Only one thought: even after it is (mis-)handled, is there any chance of spotting that we were, for a time, out of memory?