We generate way too many syscalls for `fn main() {}`

[bstrie]

Here's the strace output of a minimal C program, int main() {}:

https://gist.github.com/bstrie/5871278

...and the strace output of the Rust program fn main() {}:

https://gist.github.com/bstrie/5871221

[Aatch]

@bstrie that's not really fair, as our fn main() {} starts up the scheduler. Try comparing against #[start] fn start(_argc: int, argv: **u8, crate_map: *u8) -> int { 0 }.

EDIT: Ok, I did it myself, and that line produces this: https://gist.github.com/Aatch/5873394 which is pretty bad.

[thestinger]

@Aatch: it's still not right that we're making two calls to clone - why do we need 3 threads?

[Aatch]

I also did an strace for

#[no_std];
mod zero;
#[start]
fn start(_argc:int, _argv: **u8, _map: *u8) -> int { 0 }

and got this: https://gist.github.com/Aatch/5873394#file-start_no_std-txt

[thestinger]

@Aatch: well without the scheduler it's mostly just showing the symptoms of our rpath abuse

[brson]

The new scheduler will have 1 less thread per runtime.

[thestinger]

Closing because these seem to be caused by expected issues. I'll probably open an issue about using CLOEXEC though.