Hacker News Reader

This is called reference counting, and it's nigh-impossible to do quickly in a nondeterministic system.

https://en.wikipedia.org/wiki/Reference_counting

See this for why it's slow compared to other schemes: http://www.cecs.uci.edu/~papers/ipdps06/pdfs/1568974892-IPDP...

Reference counting is a runtime operation, an alternative to mark-and-sweep as a GC algorithm that allows for deterministic deallocation at the cost of memory overhead. While the net time overhead will probably always be greater than mark-and-sweep's (although there are optimizations which can make them quite similar), mark-and-sweep has the disadvantage of causing infrequent, long pauses rather than a predictable uniform slowness. Reference counting is used in Python (which also has mark-and-sweep to detect reference cycles), C++ in the form of shared_ptr<T>, Rust as Rc<T>, and objective c/swift (and certianly many more).

Compile time memory management refers to things like escape and ownership analysis. Escape analysis finds locals that never escape the scope they're allocated in, directly or indirectly, and allocates them on the stack rather than the heap. It's used in openJDK and probably other major JVMs, and required by the Go standard. Ownership analysis verifies that there only ever exists one live reference to an object in memory, so that a deallocation can be statically inserted whenever it leaves scope, so it doesn't become garbage. I have only seen it used in languages where there are explicit ownership annotations, such as Rust and C++.

To be sure, these are not the only forms of compile time memory management, but they're probably the most versatile.

Reply on news.ycombinator.com