Re: Messages out of sync? [textproc/icu on NetBSD 6.1.2 (GENERIC) i386]

["Michael A. Patton" <MAP%MAP-NE.com@localhost>]

Hi, Greg.

Some background that just might clear up how we got here...and BTW, I
consider this to be all cleared up for my part, and beat to death with
fanatic agreement...

I used to just do all the package stuff in a separate window running
"su" (with bright colors to ensure I knew things I typed there were
dangerous).  But with the JIT su feature, I've switched over to running
it in my own really large window (the better to see the copious output
sometimes).  So, this time as I was running it, I realized that shortly
after bringing that window up to type the root password, I needed to do
a WM operation that would freeze the screen for the duration, so I could
use that freeze of the display to see what RR was deciding to do next,
before it scrolled away.  But that also meant the root part of the one
that just finished was (usually) on the screen.  So, I decided to apply
some paranoia and actually observe what it was saying about the actions
with privileges.

The first one I observed in this way was the icu snippet I sent at the
start of this thread.  It surprised me to see it talking about working
with old versions of the package and it worried me a bit that this was
happening while it had privs.  In retrospect, before actually sending
the message, I probably should have waited for a few more to go by, but
it was after my (Doctor recommended) bedtime (as it is now), so I
didn't.  But, as the discussion has disclosed, it's doing expected
(although superfluous, since it deletes the result) actions, just not
explaining them well enough for a partially informed reader.

Greg Troxel <gdt%ir.bbn.com@localhost> writes:

> "Michael A. Patton" <MAP%MAP-NE.com@localhost> writes:
>
> First, my standard comment: when pkg_rolling-replace does something that
> surprises you, run "make replace" by itself, because all pkg_rr does is
> choose the set of packages to replace, and in what order (actually it
> calls make clean first, and make package afterwards also).  Then address
> the underlying issue, because more people will be interested in that.

That's probably what I would have done as step 2.  And then looked in
the Makefile collection to learn what it was doing for myself.  But, as I
already mentioned, the pkg_rolling-replace is still running after
several days, so I didn't want to upset things even more.  Since I had
that surprising, to me on initial reading, log snippet that was going to
scroll away, I thought it worth remarking on.  As I said, mostly "for
the record"...

>     => Creating binary package /nobackup/pkgsrc/textproc/icu/work/.packages/icu-54.1nb1.tgz
>
> So this has built the binary package preparatory to replace.  [...]
>
>     => Becoming ``root'' to make su-replace (/usr/bin/su)
>     Password:
>     Terminal type is xterm.
>     Creating binary package: icu-53.1
>     Creating package /nobackup/pkgsrc/textproc/icu/work/icu-53.1
>
> [...]
>
>     ===> Updating using binary package of icu-54.1nb1
>     /usr/pkg/sbin/pkg_add -K /var/db/pkg -U -D /nobackup/pkgsrc/textproc/icu/work/.packages/icu-54.1nb1.tgz

and, after dropping root, there was another "Creating package" message I
didn't include (which was for 54.1nb1).  I think part of the confusion
on my part stemmed from the plethora of similar messages, some of which
were talking about different versions, and none of which came with any
further information to figure out why it had to say twice that it was
making each version, and which were actually for what purpose.

Does it actually make the new binary package twice?  Once before
becoming root and after root has done a pkg_add of that one and privs
get dropped, it makes a second one.  The second one is the one that ends
up in the packages dir.  That means the one in the packages dir isn't
actually the one installed (although I assume they get made with
(essentially) the same command so _should_ be the same).  And also,
while it says "Creating package" twice while root, is that just extra
verbosity?  I assume it doesn't do it twice...

Eventually, I might still go wandering through the Makefile collection
to understand better (and maybe suggest some improvements), but I don't
think it's needed at this point, the discussion has made me more
informed about what's happening...

But, while I was typing this message, the RR seems to have lost in a
more major way.  So, I'll look into that tomorrow and maybe start a new
thread...

	-MAP