Re: g95

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

    Date:        Wed, 1 Aug 2018 11:38:58 +0100
    From:        Patrick Welche <prlw1%cam.ac.uk@localhost>
    Message-ID:  <20180801103858.GB3010@quartz>

  | The reason I was thinking of MAKE_JOBS is if the awk script was creating
  | options.h, but hadn't finished at the time the compiler included options.h.
  | How else could the #ifndef be there, but not the #endif => race condition?

This is entirely feasible (knowing nothing about the case in question though.)

But regular NetBSD builds occasionally fail for exactly that reason.  If you 
look at:
	http://releng.netbsd.org/cgi-bin/builds.cgi
the "daily" buildbots status page, you'll see that the build of HEAD from
	2018-07-30 10:10 UTC
built 66 ports correctly, but the zauras build failed.   Now this could always
be because of some arch specific build bug, but it wasn't this time, it
was ...

	dependall ===> compat/arm/oabi/../../../lib/../external/bsd/file/lib
	In file included from /home/source/ab/HEAD/src/external/bsd/file/dist/src/buffer.c:39:0:
	./magic.h:27:0: error: unterminated #ifndef
	 #ifndef _MAGIC_H
 
	nbmkdep: compile failed.

which sounds exactly like what is being discussed here.   67 builds of that
from the same sources (several of which are earm arch).  66 succeeded.

1 found a missing #endif in magic.h  (most likely more than that was missing,
but as the build stopped there we will never know.)

magic.h is a header generated on the fly by a script.

That is almost certainly exactly what you just described (different context).

kre

ps: if you want to go view this on the web page, do it soon (within a day or 
so) or the data from that build will be gone.    Also note that the more recent
"66 failed, 1 succeeded" is an entirely different kind of problem (actual issue
should be fixed for the next build) and nothing racy at all.