Re: 5.1 install-from-source

[Alan Barrett <apb%cequrux.com@localhost>]

On Sat, 12 Nov 2011, Mouse wrote:
 [using build.sh ... -x ... install=/]

> Every time, the install step complains
>
> 2011-11-12 23:24:26.01 rc check:
> 2011-11-12 23:24:26.80  ///etc/rc.d/xdm does not exist
> 2011-11-12 23:24:26.80  ///etc/rc.d/xfs does not exist
>
> and tells me
>
> 2011-11-12 23:24:30.72 postinstall checks failed: rc
> 2011-11-12 23:24:30.77 To fix, run:
> 2011-11-12 23:24:30.77     /usr/src/usr.sbin/postinstall/postinstall -s 
> '/usr/src' -d // fix rc

Using postinstall with "-s srcdir" is unlikely to do the right 
thing in general, due to differences between the build environment 
and the environment in which postinstall is run, but that should 
not be such a problem when it's run under build.sh.  I always 
recommend using "-s /path/to/etc.tgz -s /path/to/xetc.tgz".

It also seems wrong to run postinstall from the src tree, but
build.sh is using ${HOST_SH} so it will work.

> but this is doubly wrong: first, the postinstall script under 
> /usr/src has no execute bits and thus that command won't do 
> anything but complain about their lack,

If build.sh were to run postinstall from the objdir or destdir, 
then it would have execute bits.

> and second, even after prepending "sh" to it, it doesn't help; 
> it runs, but /etc/rc.d/xdm and /etc/rc.d/xfs still don't exist, 
> and the next run produces the same complaints.

When postinstall is run with "-s srcdir", it attempts to detect 
whether X11 should be included using "get_makevar MKX11 X11ROOTDIR 
X11SRCDIR".  This gives different results depending 
on whether or not it's run from build.sh, because build.sh sets 
MKX11=yes.

> Oddly, manually running the command printed just above the 
> postinstall check run,
>
> AWK=/home/mouse/i386/OBJDIR/tooldir.NetBSD-5.1-i386/bin/nbawk 
> MAKE=/home/mouse/i386/OBJDIR/tooldir.NetBSD-5.1-i386/bin/nbmake /bin/sh 
> /usr/src/usr.sbin/postinstall/postinstall -s /usr/src -d // check; if [ $? -gt 
> 1 ]; then exit 1; fi
>
> does not produce any complaints during the "rc check" portion. 
> I don't know why; the files don't exist, and the next "sh 
> build.sh ... install" run complains.

When you run it by hand, it thinks MKX11 is unset, so it doesn't 
check for those files.  When you run it from build.sh, it thinks 
MKX11=yes, so it does check for those files.  This doesn't 
explain why the files don't exist or why postinstall fix doesn't 
install them, but it does explain why the behaviour changes under 
build.sh.

> Is this known and won't be fixed in 5.x, known and fixed in 
> the 5 branch and will appear in the next 5.x release if any, 
> something that needs a PR, something that should be asked 
> elsewhere (where?),  something else, what?

Please submit a PR, and I'll try to improve the behaviour in 
-current.  Meanwhile, I suggest that you manually install those 
files for 5.x.

--apb (Alan Barrett)