On 31 March 2016 at 16:11, Christophe Lyon <christophe.l...@linaro.org> wrote: > On 31 March 2016 at 12:20, Ben Elliston <b...@air.net.au> wrote: >> On Thu, Mar 31, 2016 at 12:11:59PM +0200, Christophe Lyon wrote: >> >>> But since we currently have "|&cat", we do redirect stderr, so the >>> "and that standard error is not redirected" claude above is not >>> fulfilled. Do you know what happens in this case? >> >> No, but I would suggest that it is cleaner and more portable to use >> Tcl exec's own way of doing this: >> >> 2>@1 >> Standard error from all commands in the pipeline is redirected to >> the command result. This operator is only valid at the end of the >> command pipeline. >> >> This is used like an option, before the actual command. >> >> Give that a go. It might work better. >> > > I tried it, and saw no change in a small manual run. > > Anyway, as Maxim explained better, this is going to merge the > 2 streams on the builder, not on the board which actually > runs the test. So I think embedding the "2>&1" as part of > command executed remotely is better. > > I need to check how I can run a full validation with this > change in our environment. >
For the record, I identified what caused several random results in our setup. That's because we use 'ssh -tt' to access the remote tester. The use of '-tt' was an attempt to workaround potential stdout/stderr problems by forcing a remote testing environment as close as possible to a native one. Using 'ssh -t' seems to work better: the random results have disappeared. I will run a few more tests to determine whether adding "2>&1" to the remote command as discussed above is needed. Thanks, Christophe _______________________________________________ DejaGnu mailing list DejaGnu@gnu.org https://lists.gnu.org/mailman/listinfo/dejagnu
