[masklinn]

> The problem is that C++ doesn't have finally/unwind-protect and thus requires RAII to approximate it.

I don't agree with this statement, at least for finally v RAII: RAII is far superior a tool to deal with scope and resource management, as it requires less work on the part of the resource user and is as a consequence far more secure. It also ties in with C++'s memory semantics and therefore exists "for free", which is nice. Even more so in as messy a language as C++.

RAII has issues, one being composition (a subject on which `finally` is worse, not better) and the second one being "in-place" operations, which `finally` can handle but — at the end of the day — should probably be implemented separately.

`unwind-protect` handles both case very nicely and is therefore — as far as I'm concerned — superior. Although it has an (easily dismissed) "inconvenient" over RAII of adding a special form to the language.

Smalltalk's `BlockClosure#ensure` is also quite nice in that department, though I'd say slightly inferior to `unwind-protect` as building on it is more expensive.