Bringing in std::task generates an absurd amount of LLVM bytecode

[alexcrichton]

fn main() {          
    use std::task;   
    task::try(|| {});
}                    

When compiling with optimizations, this program takes nearly three seconds to finish. All of the time is spent in LLVM, and we're generating nearly thirty thousand lines of LLVM bytecode for this tiny function.

It would be great to understand why this is generating so much bytecode, and see if there's a way to reduce it.

[thestinger]

I've looked at the IR and nothing stands out to me as unique to this test case. This is the price of overusing macros, closures, unwinding, not hoisting common code out of generics and making use of reflection (std::repr).

Beyond throwing out the standard library and starting over (rust-core) I don't think this is actionable. The standard library just hasn't been written with attention to code bloat or performance. For example, implementing #8981 would cut down on this a lot but it's not an easy task to migrate the existing code base and deal with code generation hard-wired into the compiler.

[thestinger]

Limiting the use of repr to debug builds would be one example of tackling these systemic issues but I don't think I would be able to get support for making changes like that.