The overcommit is not the only argument.

The second argument is that complex programs that need to handle OOM properly will be full of complex untested error paths.

How should one handle a OOM error If one can't allocate some temporary node during an algorithm and how to recover without leaving a corrupted state? That's not a question I want to answer for almost every function call.

That's why the rust standard library data structures don't handle OOM. To make it easier to use for the most common case. And for the cases where the OOM handling is required, one can simply use data structures for it, which is what Linux will do (problem solved)

Also compare that to the case of mutex poisoning which can return an error on each unlock, and which is now considered a mistake because in practice, nobody want to handle a poisoned mutex error.