Proposal: Add throw/catch backend option for Do notation (faster failure paths)

[taman9333]

Description

While exploring the Do notation implementation, I discovered that Ruby's throw/catch mechanism could provide a faster alternative to the current exception-based approach, particularly for failure-heavy workloads. I wanted to share my findings and discuss whether this would be a valuable addition to the library.

Context

The current Do notation uses exceptions (Halt) for control flow, which has the benefit of working seamlessly with database transactions (automatic rollback). However, I was curious whether alternative control flow mechanisms could offer better performance.

Benchmarks (Ruby 3.4.5)

I benchmarked three approaches:

• raise/rescue (current implementation)
• throw/catch (Ruby's non-local return)
• Fiber.yield/resume

Scenario	raise	throw	fiber
3-step success	554k i/s	522k i/s	321k i/s
Early failure (step 1)	571k i/s	757k i/s	366k i/s
Middle failure (step 2)	409k i/s	529k i/s	291k i/s
Late failure (step 3)	343k i/s	419k i/s	245k i/s
Realistic operation failure	405k i/s	511k i/s	277k i/s

---

Proposal

I'm not suggesting replacing the current raise backend, it's valuable because it works with transaction blocks automatically. Instead, I'd like to propose adding an optional backend to use throw instead of raise and when using the throw backend, users would handle transactions explicitly in a base class.

So I would like to know if that's something you'd consider adding to the library? If there's interest, I'd be happy to submit a PR with the implementation

[flash-gordon]

Hi there. I was aware throw/catch is faster from the very beginning, but compatibility with transaction-like blocks is paramount. The gem could have throw/catch as an option, but I don't want it because the performance gain is not worth the added complexity. If anyone really wants to have this, they can implement their own module or even publish it as a gem; it won't conflict with dry-monads' behavior as far as I can see.
In general, if someone is after performance, I advise against dry-monads because it wraps all results. These wraps are not expensive, but they are not free.