> While testing the -dx code, I found that output was being lost without
> this change (though that was with bmake, but on *BSD they should be
> the same).  I don't recall which test makefile I was using, and it
> wasn't 100% reproducable, so I figured it better to put it in our make
> as well.
> 
> You'll note that the Job_CatchOutput() is made just prior to removing
> it from the jobs list so this is the last chance to gather output for
> it.

Hm.  THen, JobCLose() (or JobDoOutput()) needs to be called in
JobFinish() instead even if a subprocess exits normaly, doesn't it?

enami.