Explicit proper tail calls

[DemiMarie]

Rust should have a syntax for explicit proper tail calls. I have a pre-RFC on this Internals thread. Is it ready to be proposed as a full RFC?

[ticki]

I'd prefer using attributes, since tail calls ideally shouldn't affect the behavior.

[nrc]

cc #271

[DemiMarie]

Tail calls do have an observable effect: they can change a program that
crashes with a stack overflow into one that does not.

On Oct 3, 2016 01:19, "ticki" [email protected] wrote:

I'd prefer using attributes, since tail calls ideally shouldn't affect the
behavior.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1760 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AGGWB4w-jQLt-VtxDxylfFVbe6zjjfQ1ks5qwJB8gaJpZM4KMMef
.

[nagisa]

That's the practical outlook. In idealised machine (infinite memory) it
makes no difference.

From my watchtower we would evaluate tail call annotations only after it
becomes absolutely clear that whatever TCO we implement is not enough to
cover majority of the usecases by itself.

On Oct 3, 2016 10:10, "Demi Marie Obenour" [email protected] wrote:

Tail calls do have an observable effect: they can change a program that
crashes with a stack overflow into one that does not.

On Oct 3, 2016 01:19, "ticki" [email protected] wrote:

I'd prefer using attributes, since tail calls ideally shouldn't affect
the
behavior.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1760 (comment),
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AGGWB4w-jQLt-
VtxDxylfFVbe6zjjfQ1ks5qwJB8gaJpZM4KMMef>
.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1760 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AApc0krNPoHIjXM0sSXQhvX3Dkpq4q0hks5qwKpLgaJpZM4KMMef
.

[burdges]

@ticki Attributes sound more like the solution for making non-tail calls explicit: If you want explicit talk calls, then you could write become as suggested here. An attribute version like #[must_tail_call] works, but not sure people want attributes at call sites.

An alternative approach would be to require tail calls in many situations and give warnings in situations that look like tail calls should be required but they seem impossible. An attribute like #[relax_tail_cal_requirement(..)] could be less tied to the call site.

I think explicit tail calls via become or whatever sound good as a start. We should not become overly attached to become though since we might prefer this sort of warning for tail call failure in future.

[eddyb]

I'd prefer using attributes, since tail calls ideally shouldn't affect the behavior.

Tail calls (become) need to drop everything in scope before the call, which isn't the case today.

[steveklabnik]

Closing as a duplicate of #271, let's try to keep each feature to one thread 😄