[patch] testsuite: watchthreads-reorder: Linux kernel compat.

[patch] testsuite: watchthreads-reorder: Linux kernel compat.

[Jan Kratochvil]

Hi,

since Linux kernel linux-2.6.33 this testcase started to FAIL due to:
	http://lkml.org/lkml/2009/12/17/141
	464763cf1c6df632dccc8f2f4c7e50163154a2c0

I believe GDB should remain compatible with both Linux kernel variants.
Testing the Linux kernel behavior should not be a goal of the GDB testsuite,
such test has been checked into
	cvs -d :pserver:anoncvs:anoncvs@sources.redhat.com:/cvs/systemtap co ptrace-tests  
as:
	tests/state-tracing-stop.c 

OK to check-in?


Thanks,
Jan


2010-05-30  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Accept the new Linux kernel "t (tracing stop)" string.
	* gdb.threads/watchthreads-reorder.c (thread1_func, thread2_func):
	Update comment.
	(proc_string) <T (tracing stop)>: New.
	(main): Update the state_wait expect string.

--- a/gdb/testsuite/gdb.threads/watchthreads-reorder.c
+++ b/gdb/testsuite/gdb.threads/watchthreads-reorder.c
@@ -99,7 +99,7 @@ thread1_func (void *unused)
 
   rwatch_store = thread1_rwatch;
 
-  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
+  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
   timed_mutex_lock (&terminate_mutex);
   i = pthread_mutex_unlock (&terminate_mutex);
   assert (i == 0);
@@ -125,7 +125,7 @@ thread2_func (void *unused)
 
   rwatch_store = thread2_rwatch;
 
-  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
+  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
   timed_mutex_lock (&terminate_mutex);
   i = pthread_mutex_unlock (&terminate_mutex);
   assert (i == 0);
@@ -165,6 +165,12 @@ proc_string (const char *filename, const char *line)
 	  exit (EXIT_FAILURE);
 	}
 
+      /* torvalds/linux-2.6.git 464763cf1c6df632dccc8f2f4c7e50163154a2c0
+	 has changed "T (tracing stop)" to "t (tracing stop)".  Make the GDB
+	 testcase backward compatible with older Linux kernels.  */
+      if (strcmp (&buf[line_len], "T (tracing stop)") == 0)
+	buf[line_len] = 't';
+
       return &buf[line_len];
     }
   if (errno != 0)
@@ -336,9 +342,9 @@ main (int argc, char **argv)
     {
       /* s390x-unknown-linux-gnu will fail with "R (running)".  */
 
-      state_wait (thread1_tid, "T (tracing stop)");
+      state_wait (thread1_tid, "t (tracing stop)");
 
-      state_wait (thread2_tid, "T (tracing stop)");
+      state_wait (thread2_tid, "t (tracing stop)");
     }
 
   cleanup ();

Re: [patch] testsuite: watchthreads-reorder: Linux kernel compat.

[Pedro Alves]

On Sunday 30 May 2010 22:13:26, Jan Kratochvil wrote:

> OK to check-in?

I tried to understand why you did this within proc_string, but I don't
think I got it, since proc_string is used to extract other lines, not
just the "State:" line.  Why not do this within state_wait?  Okay with
that change.  Okay as is too if there's a good reason to do it as is.

> 
> 
> Thanks,
> Jan
> 
> 
> 2010-05-30  Jan Kratochvil  <jan.kratochvil@redhat.com>
> 
> 	Accept the new Linux kernel "t (tracing stop)" string.
> 	* gdb.threads/watchthreads-reorder.c (thread1_func, thread2_func):
> 	Update comment.
> 	(proc_string) <T (tracing stop)>: New.
> 	(main): Update the state_wait expect string.
> 
> --- a/gdb/testsuite/gdb.threads/watchthreads-reorder.c
> +++ b/gdb/testsuite/gdb.threads/watchthreads-reorder.c
> @@ -99,7 +99,7 @@ thread1_func (void *unused)
>  
>    rwatch_store = thread1_rwatch;
>  
> -  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
> +  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
>    timed_mutex_lock (&terminate_mutex);
>    i = pthread_mutex_unlock (&terminate_mutex);
>    assert (i == 0);
> @@ -125,7 +125,7 @@ thread2_func (void *unused)
>  
>    rwatch_store = thread2_rwatch;
>  
> -  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
> +  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
>    timed_mutex_lock (&terminate_mutex);
>    i = pthread_mutex_unlock (&terminate_mutex);
>    assert (i == 0);
> @@ -165,6 +165,12 @@ proc_string (const char *filename, const char *line)
>  	  exit (EXIT_FAILURE);
>  	}
>  
> +      /* torvalds/linux-2.6.git 464763cf1c6df632dccc8f2f4c7e50163154a2c0
> +	 has changed "T (tracing stop)" to "t (tracing stop)".  Make the GDB
> +	 testcase backward compatible with older Linux kernels.  */
> +      if (strcmp (&buf[line_len], "T (tracing stop)") == 0)
> +	buf[line_len] = 't';
> +
>        return &buf[line_len];
>      }
>    if (errno != 0)
> @@ -336,9 +342,9 @@ main (int argc, char **argv)
>      {
>        /* s390x-unknown-linux-gnu will fail with "R (running)".  */
>  
> -      state_wait (thread1_tid, "T (tracing stop)");
> +      state_wait (thread1_tid, "t (tracing stop)");
>  
> -      state_wait (thread2_tid, "T (tracing stop)");
> +      state_wait (thread2_tid, "t (tracing stop)");
>      }
>  
>    cleanup ();
> 

-- 
Pedro Alves

Re: [patch] testsuite: watchthreads-reorder: Linux kernel compat.

[Jan Kratochvil]

On Mon, 31 May 2010 04:10:07 +0200, Pedro Alves wrote:
> I tried to understand why you did this within proc_string, but I don't
> think I got it, since proc_string is used to extract other lines, not
> just the "State:" line.  Why not do this within state_wait?  Okay with
> that change.

I fully agree, thanks for catching it.

Checked-in with that change.


Regards,
Jan


http://sourceware.org/ml/gdb-cvs/2010-05/msg00255.html

--- src/gdb/testsuite/ChangeLog	2010/05/28 23:47:40	1.2293
+++ src/gdb/testsuite/ChangeLog	2010/05/31 03:31:16	1.2294
@@ -1,3 +1,11 @@
+2010-05-31  Jan Kratochvil  <jan.kratochvil@redhat.com>
+
+	Accept the new Linux kernel "t (tracing stop)" string.
+	* gdb.threads/watchthreads-reorder.c (thread1_func, thread2_func):
+	Update comment.
+	(state_wait) <T (tracing stop)>: New.
+	(main): Update the state_wait expect string.
+
 2010-05-28  Pedro Alves  <pedro@codesourcery.com>
 
 	* limits.c, limits.exp: Delete files.
--- src/gdb/testsuite/gdb.threads/watchthreads-reorder.c	2010/01/20 21:09:30	1.3
+++ src/gdb/testsuite/gdb.threads/watchthreads-reorder.c	2010/05/31 03:31:17	1.4
@@ -99,7 +99,7 @@
 
   rwatch_store = thread1_rwatch;
 
-  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
+  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
   timed_mutex_lock (&terminate_mutex);
   i = pthread_mutex_unlock (&terminate_mutex);
   assert (i == 0);
@@ -125,7 +125,7 @@
 
   rwatch_store = thread2_rwatch;
 
-  /* Be sure the "T (tracing stop)" test can proceed for both threads.  */
+  /* Be sure the "t (tracing stop)" test can proceed for both threads.  */
   timed_mutex_lock (&terminate_mutex);
   i = pthread_mutex_unlock (&terminate_mutex);
   assert (i == 0);
@@ -211,6 +211,13 @@
   do
     {
       state = proc_string (filename, "State:\t");
+
+      /* torvalds/linux-2.6.git 464763cf1c6df632dccc8f2f4c7e50163154a2c0
+	 has changed "T (tracing stop)" to "t (tracing stop)".  Make the GDB
+	 testcase backward compatible with older Linux kernels.  */
+      if (strcmp (state, "T (tracing stop)") == 0)
+	state = "t (tracing stop)";
+
       if (strcmp (state, wanted) == 0)
 	{
 	  free (filename);
@@ -336,9 +343,9 @@
     {
       /* s390x-unknown-linux-gnu will fail with "R (running)".  */
 
-      state_wait (thread1_tid, "T (tracing stop)");
+      state_wait (thread1_tid, "t (tracing stop)");
 
-      state_wait (thread2_tid, "T (tracing stop)");
+      state_wait (thread2_tid, "t (tracing stop)");
     }
 
   cleanup ();