[libcxx] r237745 - Fix race condition in thread test.
Eric Fiselier
eric at efcs.ca
Tue May 19 16:41:04 PDT 2015
Author: ericwf
Date: Tue May 19 18:41:04 2015
New Revision: 237745
URL: http://llvm.org/viewvc/llvm-project?rev=237745&view=rev
Log:
Fix race condition in thread test.
Modified:
libcxx/trunk/test/std/thread/thread.threads/thread.thread.class/thread.thread.member/detach.pass.cpp
Modified: libcxx/trunk/test/std/thread/thread.threads/thread.thread.class/thread.thread.member/detach.pass.cpp
URL: http://llvm.org/viewvc/llvm-project/libcxx/trunk/test/std/thread/thread.threads/thread.thread.class/thread.thread.member/detach.pass.cpp?rev=237745&r1=237744&r2=237745&view=diff
==============================================================================
--- libcxx/trunk/test/std/thread/thread.threads/thread.thread.class/thread.thread.member/detach.pass.cpp (original)
+++ libcxx/trunk/test/std/thread/thread.threads/thread.thread.class/thread.thread.member/detach.pass.cpp Tue May 19 18:41:04 2015
@@ -16,26 +16,41 @@
// void detach();
#include <thread>
-#include <new>
-#include <cstdlib>
+#include <atomic>
#include <cassert>
+std::atomic_bool done = ATOMIC_VAR_INIT(false);
+
class G
{
int alive_;
+ bool done_;
public:
static int n_alive;
static bool op_run;
- G() : alive_(1) {++n_alive;}
- G(const G& g) : alive_(g.alive_) {++n_alive;}
- ~G() {alive_ = 0; --n_alive;}
+ G() : alive_(1), done_(false)
+ {
+ ++n_alive;
+ }
+
+ G(const G& g) : alive_(g.alive_), done_(false)
+ {
+ ++n_alive;
+ }
+ ~G()
+ {
+ alive_ = 0;
+ --n_alive;
+ if (done_) done = true;
+ }
void operator()()
{
assert(alive_ == 1);
assert(n_alive >= 1);
op_run = true;
+ done_ = true;
}
};
@@ -45,12 +60,14 @@ bool G::op_run = false;
int main()
{
{
- std::thread t0((G()));
+ G g;
+ std::thread t0(g);
assert(t0.joinable());
t0.detach();
assert(!t0.joinable());
- std::this_thread::sleep_for(std::chrono::milliseconds(250));
+ while (!done) {}
assert(G::op_run);
- assert(G::n_alive == 0);
+ assert(G::n_alive == 1);
}
+ assert(G::n_alive == 0);
}
More information about the cfe-commits
mailing list