Re: pkg/38221 (bootstrap/bmake, devel/bmake are boken under solaris, sparc)

[Lloyd Parkes <Lloyd.Parkes%ird.govt.nz@localhost>]

The following reply was made to PR pkg/38221; it has been noted by GNATS.

From: Lloyd Parkes <Lloyd.Parkes%ird.govt.nz@localhost>
To: gnats-bugs%netbsd.org@localhost, pkgsrc-bugs%netbsd.org@localhost
Cc: 
Subject: Re: pkg/38221 (bootstrap/bmake, devel/bmake are boken under solaris, 
sparc)
Date: Mon, 21 Jul 2008 12:55:21 +1200

 Hi all,
 I've been looking at this a bit more, in particular I ran truss on bmake 
failing 
 and working. Any -j option causes the bmake test to fail. Running bmake 
without 
 -j work just fine. Here are truss outputs from the failing test, first with 
two 
 -j options that don't work and the last that did work.
 
 http://homepages.paradise.net.nz/lparkes/dotwait-make-j1.bz2
 http://homepages.paradise.net.nz/lparkes/dotwait-make-j4.bz2
 http://homepages.paradise.net.nz/lparkes/dotwait-make-noj.bz2
 
 In the failed -j1 output we can clearly see poll (pollsys) indicating EOF on 
fd 
 8 by returning "rev=POLLIN" which is promptly ignored. fd 8 is clearly the fd 
 where the results from the child appear.
 
 I haven't finished digging (and I won't get a chance to do more today), but 
line 
 2180 of job.c seems to say that the table of descriptors being polled should 
 have two entries before and children we are reading data from. The truss logs 
 clearly show only one. I have no idea where the other one has gone.
 
 Cheers
 -- 
 Lloyd Parkes
 Senior Systems Programmer
 Open Systems
 Ph: +64 4 890 2437