Compiler doesn't actually write output if rerun quickly

[alexcrichton]

I have a suspicion that we're not calling ar correctly if an rlib previously existed. This script, for example, prints 0 both times:

echo 'extern crate foo; fn main() { println!("{}", foo::foo()); }' > bar.rs
echo 'pub fn foo() -> int { 0 }' > foo.rs
rustc foo.rs --crate-type lib
rustc bar.rs -L .
./bar

echo 'pub fn foo() -> int { 1 }' > foo.rs
rustc foo.rs --crate-type lib
rustc bar.rs -L .
./bar

If I insert sleep 1 in the middle, this prints 0/1 always, and when run repeatedly it will intermittently print 0/1. I'm currently running this on linux.

[nodakai]

I think the u option given to ar doesn't examine timestamps with sub-second accuracy and thus sometimes ignores the new object file.

echo 'extern crate foo; fn main() { println!("{}", foo::foo()); }' > bar.rs
echo 'pub fn foo() -> int { 0 }' > foo.rs
strace -f -s 200 -e process -e open rustc foo.rs --crate-type lib
stat libfoo.rlib
rustc bar.rs -L .
./bar

echo 'pub fn foo() -> int { 1 }' > foo.rs
strace -f -s 200 -e process -e open rustc foo.rs --crate-type lib
stat libfoo.rlib
rustc bar.rs -L .
./bar

sleep 1
strace -f -s 200 -e process -e open rustc foo.rs --crate-type lib
stat libfoo.rlib
rustc bar.rs -L .
./bar