Nim ARC mode suitability for low-traffic services hinges on unresolved default status

For a Nim service handling only a few requests per second, the goal is to reduce steady-state memory and tail latency by selecting the ARC garbage collector instead of the legacy mark-and-sweep default. ARC reclaims objects deterministically at the point of last reference, avoiding stop-the-world pauses and potentially lowering resident memory for infrequent workloads.

The decision is unresolved because the Nim core team has not confirmed whether ARC will replace the default GC in future releases. Hesitation appears tied to compatibility with existing C-interop code and the maturity of the ORC ownership system. Switching to ARC also changes object lifetimes: code relying on non-deterministic finalizer execution, such as resource cleanup in destroy, may need auditing. Reference-counting cycles can leak unless developers explicitly use weak references or the unchecked pragma. Practical verification on the target toolchain remains necessary before committing to ARC in production.

Assuming a Nim 2.0 or later toolchain, is ARC currently safe to adopt for a long-running low-traffic service that links against C libraries? What is the recommended strategy for identifying reference cycles before they become production leaks? How should teams weigh the current uncertainty about ARC's default status when making a compiler configuration decision?