Segmentation fault on Sierra #164
Comments
|
Does the same happen with How did you install Can you run What if you run |
No. 0.2.1 works right.
Manually. I downloaded the x86_64 darwign package, and copied
Yes.
Same. Segfault. |
|
I can reproduce this on Yosemite: |
|
This looks like the specific point of failure: This points back into the implementation of |
|
I can reproduce this when compiling from source, but only on Mac, not Linux. |
|
A debug build does not segfault. |
|
sigh It looks like this is happening because of thread locals. I don't believe I'm misusing them, but if I remove the thread local, things work again. |
|
To clarify, I changed this code: pub fn matched_stripped(&self, path: &Path, is_dir: bool) -> Match {
thread_local! {
static MATCHES: RefCell<Vec<usize>> = {
RefCell::new(vec![])
}
};
MATCHES.with(|matches| {
let mut matches = matches.borrow_mut();
let candidate = Candidate::new(path);
self.set.matches_candidate_into(&candidate, &mut *matches);
for &i in matches.iter().rev() {
let pat = &self.patterns[i];
if !pat.only_dir || is_dir {
return if pat.whitelist {
Match::Whitelist(pat)
} else {
Match::Ignored(pat)
};
}
}
Match::None
})
}to this: pub fn matched_stripped(&self, path: &Path, is_dir: bool) -> Match {
let mut matches = vec![];
let candidate = Candidate::new(path);
self.set.matches_candidate_into(&candidate, &mut matches);
for &i in matches.iter().rev() {
let pat = &self.patterns[i];
if !pat.only_dir || is_dir {
return if pat.whitelist {
Match::Whitelist(pat)
} else {
Match::Ignored(pat)
};
}
}
Match::None
} |
|
Compiling with Rust stable works. Compiling with Rust beta or nightly yields the sigsegv above. |
|
Not sure if (still) relevant, but I found this. |
|
@elmart Hmm that looks related to OS X 10.6 or older. |
|
I can't seem to come up with a minimally reproducible example. Trying this works fine on Rust beta/nightly on Mac: use std::cell::RefCell;
fn foo(n: usize) -> bool {
thread_local! {
static MATCHES: RefCell<Vec<usize>> = {
RefCell::new(vec![])
}
};
MATCHES.with(|matches| {
let mut matches = matches.borrow_mut();
matches.clear();
for i in 0..n {
matches.push(i);
}
for &i in matches.iter().rev() {
if i > 0 && i % 7 == 0 {
return true;
}
}
false
})
}
fn main() {
println!("{}", foo(::std::env::args().count()));
println!("{}", foo(::std::env::args().count() * 2));
} |
|
cc @alexcrichton Do you have any ideas about this? TL;DR - I'm seeing a segfault on macos (not Linux) from using thread locals on current Rust beta/nightly, but not on current Rust stable. Efforts to repro a small example have failed. :-( |
|
Oh dear, sounds quite suspicious! The only funkiness I know of with OSX and thread locals has to do with destructors maybe, but there's no interesting destructors here. @BurntSushi what's the current reproduction? (large or small). This is quite worrisome! |
|
@alexcrichton This should do it: (I'm about to commit a work around that stops using thread locals, so this issue is hopefully going to get closed, but it's not because I've figured out the underlying problem. :-)) |
|
I've fixed this on master and will put out a (The fix was to switch to the |
|
The plot thickens! I can't seem to reproduce... (no segfaults) |
|
ripgrep |
|
You're welcome. Thx for fixing it so quickly. |
I just installed release 0.2.2 on Sierra, and have a segfault when running
rg anything.Core dump attached.
rg_2016-10-11-142243_Eliseos-MacBook-Pro.txt
The text was updated successfully, but these errors were encountered: