Georg Schwarz | 3 Oct 10:58 2006
Picon

proposed fix for problem when installing in unprivileged mode: opinions needed

For packages that do not provide their own install routine but  
instead rely on bmake's, installation fails because bmake tries to  
install things as root (cf. definition of BINGRP, BINOWN, etc. in  
share/mk/bsd.own.mk). One example of such a package is net/host. The  
reason is that pkgsrc's mk/unprivileged.mk does set the necessary  
variables, but pkgsrc fails to pass them to make at install time, so  
the defaults are used.
I'd like to suggest the following patch:

--- pkgsrc/mk/bsd.pkg.mk.orig   2006-10-03 10:38:16.000000000 +0200
+++ pkgsrc/mk/bsd.pkg.mk        2006-10-03 10:48:30.000000000 +0200
 <at>  <at>  -305,6 +305,10  <at>  <at> 
                 BSD_INSTALL_GAME=$ 
{INSTALL_GAME:Q}                      \
                 BSD_INSTALL_GAME_DATA=$ 
{INSTALL_GAME_DATA:Q}            \
                 BSD_INSTALL_GAME_DIR=${INSTALL_GAME_DIR:Q}
+.for _VAR in BINGRP BINOWN GAMEGRP GAMEOWN MANGRP MANOWN  
SHAREGRP      \
+               SHAREOWN BINMODE NONBINMODE
+INSTALL_MACROS+=       ${_VAR}=${${_VAR}:Q}
+.endfor
MAKE_ENV+=     ${INSTALL_MACROS}
SCRIPTS_ENV+=  ${INSTALL_MACROS}

What do you think of it?

Another thought is whether the variables (as set in pkgsrc's mk/ 
unprivileged.mk) are the correct and sufficient ones when comparing  
with what is defined in share/mk/bsd.own.mk.
(Continue reading)

Hubert Feyrer | 3 Oct 11:08 2006
Picon

Re: proposed fix for problem when installing in unprivileged mode: opinions needed

On Tue, 3 Oct 2006, Georg Schwarz wrote:
> +.for _VAR in BINGRP BINOWN GAMEGRP GAMEOWN MANGRP MANOWN SHAREGRP      \
> +               SHAREOWN BINMODE NONBINMODE
> +INSTALL_MACROS+=       ${_VAR}=${${_VAR}:Q}
> +.endfor
> MAKE_ENV+=     ${INSTALL_MACROS}
> SCRIPTS_ENV+=  ${INSTALL_MACROS}

The change makes sense to me, but I've never ran into this, and would wait 
for other people's opinion.

The only thing I'd suggest is get rid of the extra variable and do 
something like

 	for _VAR in ...
 	MAKE_ENV+= 	${_VAR}=${${_VAR}:Q}
 	SCRIPTS_ENV+= 	${_VAR}=${${_VAR}:Q}
 	.endfor

  - Hubert

Hans Rosenfeld | 3 Oct 11:28 2006

bin-install broken?

When I try to build a package from source using the bin-install target,
bin-install locks against itself.

panic# make bin-install BINPKG_SITES=
===> Binary install for digest-20060826
=> Installing digest-20060826 from /usr/pkgsrc/packages/All
pkg_add: no pkg found for 'digest-20060826.tgz', sorry.
pkg_add: 1 package addition failed
=> No binary package found for digest-20060826 -- installing from source
...
=> Unwrapping files-to-be-installed.
===> Installing for digest-20060826
=> Lock is held by pid 339
=> Lock is held by pid 339

This is what pstree has to say about the processes involved:

-make,5830) bin-install BINPKG_SITES
   `-sh,17217) -c ...
      `-make,339) _MAKE /usr/bin/make OPSYS NetBSD OS_VERSION 4.0_BETA ...
          `-sh,25704) -c ...
              `-make,22821) BINPKG_SITES  HOST_OSTYPE NetBSD-4.0_BETA-i386 ...
                  `-sh,5880) -c ...
                      `-make,5561) BINPKG_SITES  HOST_OSTYPE ...
                          `-sh,14520) -c ...
                              `-make,22564) BINPKG_SITES DEPENDS_TARGET ...
                                  `-sh,17168) -c ...
                                      `-sleep,22077) 5

--

-- 
(Continue reading)

Roland Illig | 3 Oct 13:08 2006
Picon

Re: bin-install broken?

Hans Rosenfeld wrote:
> When I try to build a package from source using the bin-install target,
> bin-install locks against itself.

Thank you for reporting this. I have fixed it in 
mk/install/bin-install.mk, revision 1.5.

Roland

Greg Troxel | 3 Oct 19:13 2006
Picon

CHANGES files and cvs


'make changes-entry' currently appends a line to
pkgsrc/doc/CHANGES-2006.   I have two problems with this:

1) I use cvs with CVSREAD=t in my environment, so all my checked-out
files are 444 until I do 'cvs edit' from the shell, or more likely M-x
vc-toggle-read-only.  This helps prevent accidental changes, as well
as records a kind of watch on the file ('cvs editors', and the mail
one gets when someone commits to a file I'm editing).  So CHANGES-2006
is always 444, and I have to go cvs edit it first.

2) I tend to 'cvs update' in pkgsrc every week or so, and sometimes
update a package in that state, doing cvs up in the pkg dir but not
the whole tree first.  Even if I cvs update, sometimes by the time
I've tested for an hour someone else will have committed to
CHANGES-yyyy.

So, as I see it, the steps to take after committing to a package are

1) (cd ../../doc && cvs update)
(IMHO, needed for everyone)

2) (cd ../../doc && cvs edit CHANGES-2006)
(needed for those who set CVSREAD)

The following patch makes changes-entry do what I want.  I'm guessing
some will object, so I'd like to understand the objections so I can
modify it to be ok - perhaps turning on this behavior only if a
variable is set (or turning it off if a variable is set, depending on
the prevailing view) is rw (probably a good idea).
(Continue reading)

Roland Illig | 3 Oct 20:09 2006
Picon

Re: CHANGES files and cvs

Greg Troxel wrote:
> The following patch makes changes-entry do what I want.  I'm guessing
> some will object, so I'd like to understand the objections so I can
> modify it to be ok - perhaps turning on this behavior only if a
> variable is set (or turning it off if a variable is set, depending on
> the prevailing view) is rw (probably a good idea).

I think if you name your target "commit-changes-entry", there will be 
less objections. I definitely would like to have such a target.

Roland

Georg Schwarz | 3 Oct 20:58 2006
Picon

anyone wants to try net/xtraceroute 0.9.1?

I've just upgraded net/xtraceroute to 0.9.1. Before I put it into the  
CVS repository I'd like to ask if others first would like to give it  
a try on various platforms.

--

-- 
Georg Schwarz    http://home.pages.de/~schwarz/
georg.schwarz <at> freenet.de  +49 178 8545053

Rhialto | 3 Oct 22:50 2006
Picon

multimedia/dvdrip depends on archivers/rar-bin

but obviously archivers/rar-bin only works on i386.
So dvdrip won't even try to build.

This is in the new pkgsrc-2006Q3 branch.

-Olaf.
--

-- 
___ Olaf 'Rhialto' Seibert      -- You author it, and I'll reader it.
\X/ rhialto/at/xs4all.nl        -- Cetero censeo "authored" delendum esse.

Dan McMahill | 4 Oct 12:58 2006
Picon

adding RCS Id's to bulk build output

I wonder if it would be useful in the bulk build code to grab all of the 
$NetBSD$ ID's from the Makefile, distinfo file, patches/*, etc and 
include that in the broken log.  That might help when reviewing logs 
from slower machines that may be many weeks behind in terms of their 
pkgsrc tree.

Anyone see a reason not do to this?

-Dan

Joerg Sonnenberger | 4 Oct 13:18 2006
Picon

Re: CHANGES files and cvs

On Tue, Oct 03, 2006 at 01:13:18PM -0400, Greg Troxel wrote:
> 'make changes-entry' currently appends a line to
> pkgsrc/doc/CHANGES-2006.   I have two problems with this:

[snip]

If you want to modify the current logic, I think it would be a better
idea to adjust the logic to add entries to something like
CHANGES-2006.$user and commit target to run cvs up first, append that
file, commit it and remove it again. This would reduce the number of 
conflicts extremely. The old behaviour could be kept as integrated
target doing both.

Joerg


Gmane