[CUWiN-Dev] builds using current NetBSD

[Bill Comisky]

On Wed, 25 May 2005, David Young wrote:

> On Wed, May 25, 2005 at 02:43:06AM -0500, Bill Comisky wrote:
>> On Tue, 24 May 2005, David Young wrote:
>>
>>> On Tue, May 24, 2005 at 05:55:12PM -0500, Bill Comisky wrote:
>>>>
>>>> I've tried a variety of NetBSD sources, but here's a chronicle of my
>>>> latest attempts.  The source used is described below:
>>>>
>>>> netBSD current dated 05-21-2005, with the following 2 modifications:
>>>> --------------------------------------------------------------------
>>>> - Dave's host-tools-patch-22-Apr-2005 applied.
>>>> - src/sbin/veriexecctl/veriexecctl.c updated to v1.11
>>>>  (to avoid build error)
>>>>
>>>> [note to Dave: this version of current includes v1.197 of
>>>> src/sys/uvm/uvm_map.c]
>>>>
>>>> CUWiN source r3244 with attached patch applied
>>>> -----------------------------------------------
>>>> Patch includes minor changes to allow crossbuilding, fix location of
>>>> postinstall, add stuff to the trimlist, and a couple tweaks to get the
>>>> "Official" build and svnrev to output properly for my linux build.  A
>>>> couple of the things I added to the trimlist seem like they should've been
>>>> taken care of through the trimming of base-locale-ca and base-locale-sk
>>>> set lists (I'm over my head here, maybe makeplist isn't returning all the
>>>> files it should?).
>>>>
>>>> Results using same source on two build platforms:
>>>> =================================================
>>>>
>>>> Linux build:
>>>> - Build completes, but hslsd still segfaults when started from /etc/rc.d
>>>>  (at boot time or from command line).  Just running /usr/sbin/hslsd
>>>>  without options does not cause a segfault.
>>>
>>> Sorry if I asked before, but what's in /etc/rc.conf.d/hslsd when the
>>> segfault occurs?
>>
>> I put that file (+core files, build output, etc, see below) here:
>> http://bcomisky.com/cuwin/
>>
>>>
>>>> - No $BUILDDIR/Z directory is produced.. another MAKEOBJDIR definition
>>>>  problem?  I'll try to find something that works under linux and build
>>>>  again later, so I can send unstripped binaries with the core files.
>>>>
>>>> NetBSD 1.6.2 build:
>>>> - Build fails with:
>>>>
>>>> [snip]
>>>> all ===> iswlan
>>>> nbmake: don't know how to make
>>>> /usr/local/cuwin/build/D/usr/include/errno.h. Stop
>>>
>>> How do you start the build?
>>
>> I've got a wrapper script, but it calls mkstaboot in this case like:
>>
>> ./mkstaboot -C /usr/local/cuwin/svnrepo/src/boot-image/cuw_config \
>>  -P /usr/local/cuwin/svnrepo/src/boot-image/pwds \
>>  -b /usr/local/cuwin/netbsd/build -s /usr/local/cuwin/netbsd/current/src \
>>  -U /usr/local/cuwin/images/cuwin-0.5.6+.3244.tar.gz
>>
>> oh crap, I just noticed the errno.h path is wrong in the error message
>> above.  The build dir should be: /usr/local/cuwin/netbsd/build as per the
>> mkstaboot command above.. just checked, I had deleted the build directory
>> and set the path to a new location, but there were some .depend files
>> lingering in the cuwin src subdirectories with the old build path. [some
>> time passes] ok, after deleting those .depend files the build on my netbsd
>> box finishes, but hslsd segfaults using the generated image from that
>> build as well.  The /etc/rc.conf.d/hslsd files were identical.
>>
>>> Is there anything in /usr/local/cuwin/build/D/ ?  What about
>>> /usr/local/cuwin/build/ ?  Are your NetBSD sources cluttered with .o
>>> and .d files, by chance?
>>
>> The only difference I can see is that on the Linux build machine there is
>> no $BUILDDIR/Z directory.  The $BUILDDIR/Z directory on the netbsd build
>> machine looks like it just has CUWiN /src object files in it.  The D I O T
>> directories are there and the $BUILDDIR/D directory is full.  And neither
>> build machines have .o/.d files cluttering the netbsd sources.
>>
>> Since I do have a $BUILDDIR/Z machine on the netbsd box, I tar'd it up and
>> put it with the rest of files you may want to see, including:
>>
>> /etc/rc.conf.d/hslsd
>> /var/core/hslsd.*
>> build output files (x2 for before and after I cleaned up the .depend files)
>
> Something stands out in the build outputs files: many steps are skipped.
> I suggest restarting the build earlier than 'modules'.  With -u -o,
> it will not take all day if you restart at step 'tools'.
>
> (from your build-output.2)
>
> mkstaboot: skipped step 'tools'
> mkstaboot: begin step 'toolenv'
> mkstaboot: finished step 'toolenv'
> mkstaboot: will write an upgrade tar to /usr/local/cuwin/images/cuwin-0.5.6+.3244.tar.gz
> mkstaboot: skipped step 'tools'
> mkstaboot: begin step 'toolenv'
> mkstaboot: finished step 'toolenv'
> mkstaboot: begin step 'mediaenv'
> mkstaboot: finished step 'mediaenv'
> mkstaboot: begin step 'makewrapperenv'
> mkstaboot: finished step 'makewrapperenv'
> mkstaboot: skipped step 'distrib'
> mkstaboot: skipped step 'metalog'
> mkstaboot: skipped step 'patch'
> mkstaboot: skipped step 'makewrapper'
> mkstaboot: skipped step 'mv-root-home'
> mkstaboot: skipped step 'flash-kernel'
> mkstaboot: skipped step 'extras'
> mkstaboot: skipped step 'users'
> mkstaboot: begin step 'modules'
>

Right.. I figured it'd be ok since I made it all the way to the iswlan 
code the first time and I only deleted some .depend files in the CUWiN 
source before restarting from modules.

Just to be safe I did a clean build (removed $BUILDDIR) on both netbsd and 
linux platforms last night with the same code.  On both hslsd is still 
segfaulting.  I put a new bzip'd set of build output from the netbsd build 
here:

http://bcomisky.com/cuwin/build-output.bz2

bill

--
Bill Comisky
bcomisky at pobox.com