Fix static mut issues with routines
#677
Conversation
crates/harp/src/routines.rs
Outdated
| // Now unwrap the definitions from our thread-safe type | ||
| let unwrapped: Vec<R_CallMethodDef> = R_ROUTINES | ||
| .lock() | ||
| .unwrap() | ||
| .iter() | ||
| .map(|def| def.inner) | ||
| .collect(); |
There was a problem hiding this comment.
I'm fairly surprised you dont get a warning about "dropping temporary value" since no one is holding the lock in a local variable here. I seem to get that all the time when doing stuff like this.
| unsafe impl Send for CallMethodDef {} | ||
| unsafe impl Sync for CallMethodDef {} | ||
|
|
||
| static R_ROUTINES: Mutex<Vec<CallMethodDef>> = Mutex::new(vec![]); |
There was a problem hiding this comment.
Nice I also tried this when fiddling with this one last week but got stuck on the R_CallMethodDef not being send+sync problem
| unsafe impl Send for CallMethodDef {} | ||
| unsafe impl Sync for CallMethodDef {} |
There was a problem hiding this comment.
I think my main comment is that #[harp::register] is pretty freaking magical. It generates some code that runs at application startup time in an OS specific way.
I guess we can assume that code is always run on a single thread? I have no idea how startup code gets run on a per OS basis. I just want to make sure this doesn't screw us up.
There was a problem hiding this comment.
We're just saying it is fine to send these raw pointers across threads, which it is as long as you don't do anything with them. We're not introducing any new area of unsafety AFAICT.
lionel-
force-pushed
the
task/gd-static-mut
branch
from
6d2a559 to
49d2cfa
Compare
It creates references to `static mut`
lionel-
force-pushed
the
task/routines-static-mut
branch
from
fc25a0d to
ea72243
Compare
Co-authored-by: Davis Vaughan <[email protected]>
Closes #661.
Ideally
R_ROUTINESwould be thread-local but that's currently inconvenient in unit tests.