Re: PR/48875 CVS commit: src/tests/bin/sh

[Robert Elz <kre%munnari.OZ.AU@localhost>]

The following reply was made to PR bin/48875; it has been noted by GNATS.

From: Robert Elz <kre%munnari.OZ.AU@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: PR/48875 CVS commit: src/tests/bin/sh
Date: Wed, 08 Jun 2016 02:29:15 +0700

     Date:        Tue,  7 Jun 2016 16:35:00 +0000 (UTC)
     From:        David Holland <dholland-bugs%netbsd.org@localhost>
     Message-ID:  <20160607163500.C74B37AAAE%mollari.NetBSD.org@localhost>
 
   |  Minor note on this in passing: our sleep(1) honors "sleep 0.1" and the
   |  tests don't really have to be totally portable,
 
 Yes, I know, and while I'd prefer the tests to be as portable as possible,
 not all of them are, but ...
 
   | so you might be able to improve the overall sleep time.
 
 It might be possible to make some minor improvement, but the real issue
 is that the sequencing of the sleeps finishing matters to the tests
 working correctly, on an unloaded system we might expect that
 	sleep 0.2 & sleep 0.1 &
 would have the 2nd sleep finish first, but if the system is just a bit
 loaded and it takes 0.1 of a second for the second one to even start,
 we have a race, and the test will sporadically report failures.
 
 Rather than having that happen, I think it better to have longer sleeps,
 with bigger differences in the intervals, even if that does mean an additional
 20 seconds (of idleness) during the test run, so that even on truly busy
 systems it is very unlikely that there will end up being false failures
 (or real failures that aren't a result of a failure in what is really
 being tested, however you want to characterise it.)
 
 kre