[Feature Request] [mojo-lang] Non-destructive comptime for unrolling

Review Mojo's priorities

• I have read the roadmap and priorities and I believe this request falls within the priorities.

What is your request?

Currently elaborating a comptime for clones the entire loop body for each iteration (including nested parameter scope graphs), binds the induction variable on that clone, and elaborates by consuming the clone. Because elaboration is destructive, each iteration needs its own disposable IR copy.

I suspect a big share of the cost for large loops is this clone then consume model, not just the work that actually depends on the induction variable. Nested comptime for makes that cost multiplicative.

It would be good to explore a model where the loop body can be treated more like a reusable template, by binding parameters per iteration and elaborating without requiring a full body clone for every trip. But I don't know what shape that could/should take and how it could/should be wired into the existing infrastructure.

What is your motivation for this change?

comptime for is a frequent source of exploding compile times. Making it cheaper would help a lot of generic code.

Any other details?

CC: @dgurchenkov

Start by tracing the existing comptime for elaboration path described in the issue, especially where the loop body is cloned, induction variables are bound, and the clone is consumed. Determine whether a reusable template model fits the existing infrastructure; done means a concrete design and implementation that avoids full per-iteration body clones while preserving nested-loop behavior and improving compile-time cost.