Re: [PATCH v2 4/4] testsuite/smokey/psostests: Fix compile error seen with gcc 16

On Mon, 2026-05-18 at 13:49 +0200, Jan Kiszka wrote:
> On 18.05.26 12:24, Florian Bezdeka wrote:
> > On Mon, 2026-05-18 at 11:21 +0200, Jan Kiszka wrote:
> > > On 18.05.26 11:02, Florian Bezdeka wrote:
> > > > Fixes the following compile errors:
> > > > 
> > > > rn-1.c: In function ‘alloc_task’:
> > > > rn-1.c:16:18: error: variable ‘n’ set but not used [-Werror=unused-but-set-variable=]
> > > >    16 |         int ret, n;
> > > >       |                  ^
> > > > rn-1.c:15:22: error: variable ‘alloc_size’ set but not used [-Werror=unused-but-set-variable=]
> > > >    15 |         u_long size, alloc_size = 0;
> > > >       |
> > > > 
> > > > task-2.c: In function ‘backgroundTask’:
> > > > task-2.c:19:42: error: variable ‘count’ set but not used [-Werror=unused-but-set-variable=]
> > > >    19 |         unsigned int safety = 100000000, count = 0;
> > > >       |                                          ^~~~~
> > > > cc1: all warnings being treated as errors
> > > > 
> > > > The counting busy loop has been replaced by an interval based busy
> > > > loop.
> > > > 
> > > > Signed-off-by: Florian Bezdeka <[email protected]>
> > > > ---
> > > >  testsuite/smokey/psostests/rn-1.c   | 7 +++----
> > > >  testsuite/smokey/psostests/task-2.c | 5 ++---
> > > >  2 files changed, 5 insertions(+), 7 deletions(-)
> > > > 
> > > > diff --git a/testsuite/smokey/psostests/rn-1.c b/testsuite/smokey/psostests/rn-1.c
> > > > index 5e82d1506ca029ae2bf812ccbfbf9c3a9ab9cc7e..d2e847375212c9746de3e550a5d10a8be4694de5 100644
> > > > --- a/testsuite/smokey/psostests/rn-1.c
> > > > +++ b/testsuite/smokey/psostests/rn-1.c
> > > > @@ -12,15 +12,15 @@ static u_long tid, rnid;
> > > >  
> > > >  static void alloc_task(u_long a1, u_long a2, u_long a3, u_long a4)
> > > >  {
> > > > -	u_long size, alloc_size = 0;
> > > > -	int ret, n;
> > > > +	u_long size;
> > > >  	void *buf;
> > > > +	int ret;
> > > >  
> > > >  	traceobj_enter(&trobj);
> > > >  
> > > >  	srandom(0x11223344);
> > > >  
> > > > -	for (n = 0;; n++) {
> > > > +	for (;;) {
> > > >  		size = (random() % (sizeof(rn_mem) / 8)) + 4;
> > > >  		ret = rn_getseg(rnid, size, RN_NOWAIT, 0, &buf);
> > > >  		if (ret) {
> > > > @@ -28,7 +28,6 @@ static void alloc_task(u_long a1, u_long a2, u_long a3, u_long a4)
> > > >  			break;
> > > >  		}
> > > >  		memset(buf, 0xaa, size);
> > > > -		alloc_size += size;
> > > >  	}
> > > >  
> > > >  	traceobj_exit(&trobj);
> > > > diff --git a/testsuite/smokey/psostests/task-2.c b/testsuite/smokey/psostests/task-2.c
> > > > index d24c13465367c7c2a6975dd4cbb01eb75f8fd1ad..be1b6ca18c5354ba092d55b3655bae9a294f2ce7 100644
> > > > --- a/testsuite/smokey/psostests/task-2.c
> > > > +++ b/testsuite/smokey/psostests/task-2.c
> > > > @@ -2,6 +2,7 @@
> > > >  #include <stdio.h>
> > > >  #include <stdlib.h>
> > > >  #include <copperplate/traceobj.h>
> > > > +#include <copperplate/threadobj.h>
> > > >  #include <psos/psos.h>
> > > >  
> > > >  static struct traceobj trobj;
> > > > @@ -16,7 +17,6 @@ static u_long sem_id;
> > > >  
> > > >  static void backgroundTask(u_long a1, u_long a2, u_long a3, u_long a4)
> > > >  {
> > > > -	unsigned int safety = 100000000, count = 0;
> > > >  	int ret;
> > > >  
> > > >  	traceobj_enter(&trobj);
> > > > @@ -28,8 +28,7 @@ static void backgroundTask(u_long a1, u_long a2, u_long a3, u_long a4)
> > > >  
> > > >  	traceobj_mark(&trobj, 2);
> > > >  
> > > > -	while (--safety > 0)
> > > > -		count++;
> > > > +	threadobj_spin(1000);
> > > 
> > > That's just 1 µs, isn't it? That sounds rather short, not only for VM
> > > setups.
> > > 
> > 
> > Arg, this is crazy. I measured the execution time of the previous
> > counting loop on my nested virtualized test system and tried to get down
> > as much as possible until CI would complain. Turned out: This test is
> > actually never run in CI...
> > 
> 
> Haha!
> 
> > There is a comment in psostests.c that this test (as it was) is broken,
> > so disabled.
> > 
> > To get that running on my nested VM setup here I have to increase to
> > something around 2ms.
> > 
> > Let me spawn another couple of CI test runs, maybe we can re-enable that
> > test - or delete it entirely?
> 
> Someone who wrote it may help us to understand its original purpose:
> adding Philippe.
> 
> testsuite/smokey/vxworkstests/task-2.c looks similar on first glance,
> but it is not disabled.
> 

My understanding is that those tests are actually testing the same
thing, just for different skins.

In my case both busy loops (the counting ones) are executed that fast
that I was always entering the "should not be reached anymore" sections.
In case of the vxworkstest I'm ending up in pause(). That's why I simply
removed the busy waiting in this case.

Let's wait for Philippe, I can't tell from the available information
(git log, ...) what exactly should be tested.

Florian