Use immutable pass by reference. Make a copy only if mutability is requested in the thread. This makes concurrent reads lock-free but also cuts down on memory allocations.
This is essentially what Herd does. It's only semantically a pass by value, but the same reference counting optimizations still apply.
In fact, Herd's approach is a bit more powerful than this because (in theory) it can remove the copy entirely if the caller doesn't use the old value any more after creating the thread. In practice, my optimizations aren't perfect and the language won't always detect this.
The big downside is that we have to use atomic reference counts for _everything_. From memory this was about a 5-15% performance hit versus non-atomic counters, though the number might be higher if other bottlenecks were removed.
Nothing "real", just the synthetic benchmarks in the ./benchmarks dir.
Unnecessary copies are definitely a risk, and there's certain code patterns that are much harder for my interpreter to detect and remove. E.g. the nbodies has lots of modifications to dicts stored in arrays, and is also the only benchmark that loses to Python.
The other big performance limitation with my implementation is just the cost of atomic reference counting, and that's the main tradeoff versus deep cloning to pass between threads. There would definitely be room to improve this with better reference counting optimizations though.
Also, copy by value in itself is just a semantic requirement, it doesn't say how it's implemented.
And shared mutable memory is pretty damn slow (given you are not fine with data race garbage), because atomic operations destroy caches. So it's the usual space-time tradeoff at the end of the day.
In C, you can simply mutate the underlying characters. So changing the fifth character in a string is as easy as:
str[4] = 0;
Whereas using the immutable syntax you might have to do something like: str = str.substr(0, 4) + "\0" + str.substr(4);(Ah, no, your example elsewhere in the thread suggests that the referred-to structures get implicitly copied all over the place.)
Optimization specifically for function calls, or... ?
Because if you're doing all this copy-on-write anyway, the indirection seems to be a needless cost in other contexts.
> needless cost
Are you comparing to a language with mutable references or a our functional language? A language with mutable references will of course be faster, but this is more intended as an alternative to pure functional languages (since functions are referentially transparent).
In this case, the cost of the indirection is approximately zero (relative to the cost of just doing reference counting), since passing a reference just requires a bump to the refcount. And most of the time the refcount increments are skipped by "moving" instead of copying the reference.
But I think at this point I'd have to read and analyze the code to have a proper understanding.
... Although granting that you already have paid the cost of reference-counting GC (and, I assume, per-object allocation), it probably is indeed insignificant. And special-casing where the reference count == 1 is also kinda neat. (E.g. CPython doesn't "move references" per se if I'm thinking clearly; but it does detect cases where it can safely mutate the underlying implementation of a type that's "immutable" from the perspective of language syntax.)
I've worked on systems where we spent more time reasoning about shared state than writing actual logic. The typical answer is "just make everything immutable" but then you lose convenient imperative syntax. This sits in an interesting middle ground.
Curious about performance in practice. Copy-on-write is great until you hit a hot path that triggers lots of copies. Have you benchmarked any real workloads?