[ruby-core:30040] Re: [Bug #1341] pthread_cond_timedwait failing in 1.9.1-p0 thread tests

Hi,

2010/5/6 Graham Agnew <[e-mail]>:
> On HP-UX, that patch stopped the rb_bug_errno happening, although the test/ruby/test_io.rb and test/ruby/test_thread.rb scripts both blocked indefinitely.
>
> The code in your patch doesn't look right to me. ?Shouldn't the code re-fetch the time using gettimeofday each time through the loop, and then add the MEGA_SEC to that? ?As it is, it's wrong because it adds a MEGA_SEC to tvn, so after one MEGA_SEC it will enter a hard loop.


Thank you for your testing!  How about the following patch?

To tell the truth, I'm writing a patch without test because I
don't have HP-UX.  If this is wrong again, It is really helpful
for you to correct the patch by yourself.


diff --git a/thread_pthread.c b/thread_pthread.c
index e6295db..3c13f72 100644
--- a/thread_pthread.c
+++ b/thread_pthread.c
@@ -631,8 +631,29 @@ native_sleep(rb_thread_t *th, struct timeval *tv)
 		int r;
 		thread_debug("native_sleep: pthread_cond_timedwait start (%ld, %ld)\n",
 			     (unsigned long)ts.tv_sec, ts.tv_nsec);
+	      again:
 		r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
 					   &th->interrupt_lock, &ts);
+		if (r == EINVAL) {
+		    /* workaround for Solaris: wait by MEGA_SEC's.
+		     * on Solaris, pthread_cond_timedwait fails with EINVAL
+		     * if time is too far from now.  [Bug #1341]
+		     * - http://docs.sun.com/app/docs/doc/806-0630/6j9vkb8ct?a=view
+		     * - http://bugs.opensolaris.org/view_bug.do?bug_id=4038480
+		     */
+#define MEGA_SEC 1000000
+		    struct timespec lts;
+		    r = ETIMEDOUT;
+		    while (r == ETIMEDOUT) {
+			gettimeofday(&tvn, NULL);
+			lts.tv_sec = tvn.tv_sec + MEGA_SEC;
+			lts.tv_nsec = tvn.tv_usec * 1000;
+			if (lts.tv_sec >= ts.tv_sec) goto again;
+			r = pthread_cond_timedwait(&th->native_thread_data.sleep_cond,
+						   &th->interrupt_lock, &lts);
+			if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);
+		    }
+		}
 		if (r && r != ETIMEDOUT) rb_bug_errno("pthread_cond_timedwait", r);

 		thread_debug("native_sleep: pthread_cond_timedwait end (%d)\n", r);

-- 
Yusuke Endoh <[e-mail]>