This repo contains minimum viable programs for double locking bugs affecting RwLock,
prompted by sigp/lighthouse#2245.
The program bad.rs naively uses two calls to RwLock::read on the same thread,
which can deadlock if a write is performed concurrently, per this execution:
$ cargo run --release --bin bad
start
t1 trying first read
t2 trying write
t1 trying second read
i.e. the deadlock occurs when thread 1 already holding the lock attempts to re-lock it, at the same time that thread 2 is trying to obtain a write lock.
It's also possible for the "bad" program to run without deadlocking, e.g.
start
t1 trying first read
t1 trying second read
t1 obtained two read locks
t2 trying write
t2 obtained write lock
Running it in a loop will quickly surface the deadlock (try ./loop_bad.sh).
In contrast, the program good.rs uses RwLock::read_recursive for the 2nd read lock and
will never deadlock (try ./loop_good.sh).
The trade-off is that read_recursive is prone to starving writers.