Compile-time static stack size

[nwmcsween]

This is a possible idea to implement #3608 by static analysis of stack usage to create a statically sized stack depth at compile-time.

[bblum]

This could be done on its own, but I think it could also be an elegant sub-feature of a polymorphic effect system.

[emberian]

Still relevant

[pnkfelix]

visiting for triage, email from 2013-08-26

switching from I-enhancement to I-wishlist.

(My personal thinking is that: 1. we are unlikely to get a polymorphic effect system in, at least not anytime soon, certainly not for rust 1.0, and 2. maybe one could implement this as a standalone tool for a subset of the language; but I have not really fleshed out much of my thinking about the latter.)

[nwmcsween]

On Friday, August 30, 2013 3:13:15 AM, Felix S Klock II wrote:

visiting for triage, email from 2013-08-26

switching from |I-enhancement| to |I-wishlist|.

(My personal thinking is that: 1. we are unlikely to get a polymorphic
effect system in, at least not anytime soon, certainly not for rust
1.0, and 2. maybe one could implement this as a standalone tool for a
subset of the language; but I have not really fleshed out much of my
thinking about the latter.)

—
Reply to this email directly or view it on GitHub
#4389 (comment).

The reason this is an issue is rust currently relies on gcc / glibc
"split stacks" which 1. Is specific to glibc / gcc and 2. Is signal
unsafe.

[thestinger]

There's no dependency on glibc or gcc for split stacks, since it's implemented by LLVM and it just calls our __morestack implementation. Signals can be dealt with by using sigaltstack.

[emberian]

Related: http://cmr.github.io/blog/2013/10/21/on-stack-safety/ and http://cmr.github.io/blog/2013/10/28/more-on-stack-safety/

[thestinger]

While this is an interesting prospect, it's not going to permit using recursion, function pointers, trait objects or closures. I'm not really sure the remaining language could still be considered as Rust. A specific proposal would need to be made through the RFC process, as this feature request isn't actionable without consensus on a design.