Re: bin/53285: Build times tripled with make/job.c 1.193

["Simon J. Gerraty" <sjg%juniper.net@localhost>]

The following reply was made to PR bin/53285; it has been noted by GNATS.

From: "Simon J. Gerraty" <sjg%juniper.net@localhost>
To: <gnats-bugs%NetBSD.org@localhost>
Cc: <gnats-admin%NetBSD.org@localhost>, <netbsd-bugs%NetBSD.org@localhost>,
        Andreas Gustafsson
	<gson%gson.org@localhost>, <sjg%juniper.net@localhost>
Subject: Re: bin/53285: Build times tripled with make/job.c 1.193
Date: Sun, 13 May 2018 10:18:06 -0700

 Robert Elz <kre%munnari.OZ.AU@localhost> wrote:
 >    | Please revert the commit in case.
 >  =
 
 >  Before giving up completely, perhaps it would be worth trying changing =
 the
 >  "sleep(1)" into a much smaller usleep() or nanosleep() - which would ha=
 ve
 >  the same effect, of avoiding busy waiting, while not adding nearly as m=
 uch
 >  delay (if it happens that the sleep needs to be a second occasionally, =
 then
 >  that might mean 100 * usleep(10000) rather than 1 sleep(1), but it seem=
 s that
 >  most of the time, the wait does not need to be nearly that long).
 
 Removing the sleep entirely, would not be unreasonable since
 the busy wait issue has existed for several years.
 Ie. when there are no tokens and no jobs running is when you'd
 want wantToken=3D1 but that's not when it was being set.
 
 Will try usleep(10000)....