Re: git: ac52fa1f1b97 - main - sendfile tests: Fix an error check in tcp_socketpair()

On Mon, Jun 24, 2024 at 01:04:59PM -0700, Gleb Smirnoff wrote:
>   Mark,
> 
> On Fri, Jun 21, 2024 at 06:21:57PM +0000, Mark Johnston wrote:
> M> commit ac52fa1f1b9768d281b9deed6d1158f43a8564ca
> M> Author:     Mark Johnston <markj@FreeBSD.org>
> M> AuthorDate: 2024-06-21 18:18:22 +0000
> M> Commit:     Mark Johnston <markj@FreeBSD.org>
> M> CommitDate: 2024-06-21 18:18:22 +0000
> M> 
> M>     sendfile tests: Fix an error check in tcp_socketpair()
> M>     
> M>     Fixes:  c368d3f20fc0 ("tests/sendfile: factor out tcp_socketpair()")
> M>     Differential Revision:  https://reviews.freebsd.org/D45598
> M> ---
> M>  tests/sys/kern/sendfile_helper.c | 2 +-
> M>  1 file changed, 1 insertion(+), 1 deletion(-)
> M> 
> M> diff --git a/tests/sys/kern/sendfile_helper.c b/tests/sys/kern/sendfile_helper.c
> M> index 703b04fdea6c..6365531e312c 100644
> M> --- a/tests/sys/kern/sendfile_helper.c
> M> +++ b/tests/sys/kern/sendfile_helper.c
> M> @@ -84,7 +84,7 @@ tcp_socketpair(int *sv)
> M>  	if (fcntl(sv[0], F_SETFL, flags) == -1)
> M>  		err(1, "fcntl +O_NONBLOCK");
> M>  
> M> -	if (connect(sv[0], (void *)&sin, sizeof(sin)) != -1 ||
> M> +	if (connect(sv[0], (void *)&sin, sizeof(sin)) == -1 &&
> M>  	    errno != EINPROGRESS)
> M>  		err(1, "connect cs");
> 
> Sorry for not reviewing in a timely manner. Did you actually observe connect(2)
> on a non-blocking TCP socket to immediately return success?

Yes, I was seeing occasional failures from my test runner where the
helper exited with "connect cs: <nonsensical error message>".  It's easy
to reproduce this by running the tests on a system with some background
CPU load.