update build fails on mtree invocation

[Rhialto]

This doesn't happen with a fresh build, but a re-build in update mode
fails like this (with -j 4, in case it matters):

obj ===> regress/lib/libdes
obj ===> regress/lib/libc/siginfo/sigfpe
obj ===> regress/lib/libc/siginfo/sigsegv
clean_METALOG ===> .
clean_METALOG ===> distrib/sets
nbmtree: unknown group `ftp'
nbmtree: failed at line 1301 of the specification
--- /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised ---
*** [/vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised] Error code 1
1 error
nbmake: stopped in /tank/vol-fourquid-1/olafs/netbsd-git/cvs/src/distrib/sets
--- clean_METALOG ---
*** [clean_METALOG] Error code 2

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

[Rhialto]

On Wed 04 Nov 2009 at 21:52:56 +0100, Rhialto wrote:
> This doesn't happen with a fresh build, but a re-build in update mode
> fails like this (with -j 4, in case it matters):

Oops, I spoke too soon. It happens on a regular clean build as well.
Crosscompiling from FreeBSD, which apparently matters:

    install  /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/etc/fonts/conf.d/80-delicious.conf
    install  /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/etc/fonts/conf.d/90-synthetic.conf
makesetfiles ===> /tank/vol-fourquid-1/olafs/netbsd-git/cvs/src/distrib/sets
nbmtree: unknown group `ftp'
nbmtree: failed at line 1301 of the specification
--- /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised ---
*** [/vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised] Error code 1
1 error
nbmake: stopped in /tank/vol-fourquid-1/olafs/netbsd-git/cvs/src/distrib/sets
--- distribution ---
*** [distribution] Error code 2
1 error

The file .../destdir.amd64/METALOG does indeed contain this on line 1301:

./var/chroot/named/etc/namedb/cache type=dir uid=14 gname=ftp mode=0775

and it is the first time that gname=ftp is mentioned.

What strikes me as odd though is that ftp is used there at all; and
indeed in an earlier METALOG file from a native build I find

./var/chroot/named/etc/namedb/cache type=dir uname=named gname=named mode=0775

Maybe something in the build system uses the hosts's uid and gid
numbers? The host's /etc/group file contains

ftp:*:14:

where on my own netbsd system I have the same gid for named:

named:*:14:

Maybe the METALOG should always use numerical values for these, if they
are meant to be specific numbers. (I wonder where it got the mapping
named -> 14 from... and why the sanitizing step cannot use it as well...
or tar when it builds the tar files...)

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

[hayashi]

Hi,

This problem is
http://cvsweb.netbsd.org/bsdweb.cgi/src/tools/compat/compat_defs.h.diff?r1=1.72&r2=1.73&only_with_tag=MAIN&f=h
Diff for /src/tools/compat/compat_defs.h between version 1.72 and 1.73

Some hosts (ex. FreeBSD) have group_from_gid() (HAVE_GROUP_FROM_GID), but dont have
pwcache_groupdb() (HAVE_PWCACHE_GROUDB).
NetBSD (compatible) library's pwcache_groupdb() changes user database access
routine of group_from_gid(), but this group_from_gid() is NetBSD library.
FreeBSD's group_from_gid() can't change database access routine.
then, it is necessary to treat HAVE_GROUP_FROM_GID, HAVE_GID_FROM_GROUP and
HAVE_PWCACHE_GROUDB treat as a group.
by same reason, UID definitions will be treated as a group.


Rhialto wrote:

> On Wed 04 Nov 2009 at 21:52:56 +0100, Rhialto wrote:
>> This doesn't happen with a fresh build, but a re-build in update mode
>> fails like this (with -j 4, in case it matters):
>
> Oops, I spoke too soon. It happens on a regular clean build as well.
> Crosscompiling from FreeBSD, which apparently matters:
>
>     install  /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/etc/fonts/conf.d/80-delicious.conf
>     install  /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/etc/fonts/conf.d/90-synthetic.conf
> makesetfiles ===> /tank/vol-fourquid-1/olafs/netbsd-git/cvs/src/distrib/sets
> nbmtree: unknown group `ftp'
> nbmtree: failed at line 1301 of the specification
> --- /vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised ---
> *** [/vol-fourquid-1/olafs/netbsd-git//work/destdir.amd64/METALOG.sanitised] Error code 1
> 1 error
> nbmake: stopped in /tank/vol-fourquid-1/olafs/netbsd-git/cvs/src/distrib/sets
> --- distribution ---
> *** [distribution] Error code 2
> 1 error
>
> The file .../destdir.amd64/METALOG does indeed contain this on line 1301:
>
> ./var/chroot/named/etc/namedb/cache type=dir uid=14 gname=ftp mode=0775
>
> and it is the first time that gname=ftp is mentioned.
>
> What strikes me as odd though is that ftp is used there at all; and
> indeed in an earlier METALOG file from a native build I find
>
> ./var/chroot/named/etc/namedb/cache type=dir uname=named gname=named mode=0775
>
> Maybe something in the build system uses the hosts's uid and gid
> numbers? The host's /etc/group file contains
>
> ftp:*:14:
>
> where on my own netbsd system I have the same gid for named:
>
> named:*:14:
>
> Maybe the METALOG should always use numerical values for these, if they
> are meant to be specific numbers. (I wonder where it got the mapping
> named -> 14 from... and why the sanitizing step cannot use it as well...
> or tar when it builds the tar files...)
>
> -Olaf.

[Rhialto]

On Thu 05 Nov 2009 at 23:28:00 +0900, Akihiko HAYASHI wrote:

> Hi,
>
> This problem is
> http://cvsweb.netbsd.org/bsdweb.cgi/src/tools/compat/compat_defs.h.diff?r1=1.72&r2=1.73&only_with_tag=MAIN&f=h
> Diff for /src/tools/compat/compat_defs.h between version 1.72 and 1.73
>
> Some hosts (ex. FreeBSD) have group_from_gid() (HAVE_GROUP_FROM_GID), but dont have
> pwcache_groupdb() (HAVE_PWCACHE_GROUDB).
> NetBSD (compatible) library's pwcache_groupdb() changes user database access
> routine of group_from_gid(), but this group_from_gid() is NetBSD library.
> FreeBSD's group_from_gid() can't change database access routine.
> then, it is necessary to treat HAVE_GROUP_FROM_GID, HAVE_GID_FROM_GROUP and
> HAVE_PWCACHE_GROUDB treat as a group.
> by same reason, UID definitions will be treated as a group.

Thank you! I have reverted the patch, and started an update build.
Now I can see that the line in METALOG now again contains

./var/chroot/named/etc/namedb/cache type=dir uname=named gname=named mode=0775

The build is still running (it seems to re-build much more than seems
needed) but I have good hopes that it finishes now.

Can you or someone else commit this?

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