Fun with pungi

[Ignacio Vazquez-Abrams]

On a lark, I decided to use pungi against fedora-livecd-desktop.ks in
order to produce smaller Rawhide install media. After a few false starts
I managed to get working media, and I installed to an external USB HD.
The install booted, but firstboot soiled itself due to a missing
libdb-4.5.so. I booted the rescue media in order to install it, but I
discovered that it wasn't even on the media. Long story short, the
following packages were not included on the media or even in the package
download cache:

compat-db45
hunspell
hunspell-en

I'm sure that pungi is inevitably to blame, but I'm wondering if we
shouldn't add those packages to fedora-livecd-desktop.ks and/or
fedora-live-base.ks in the meantime.

-- 
Ignacio Vazquez-Abrams <ivazqueznet(a)gmail.com&gt;

PLEASE don't CC me; I'm already subscribed

[Jeroen van Meeuwen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ignacio Vazquez-Abrams wrote:
| On a lark, I decided to use pungi against fedora-livecd-desktop.ks in
| order to produce smaller Rawhide install media. After a few false starts
| I managed to get working media, and I installed to an external USB HD.
| The install booted, but firstboot soiled itself due to a missing
| libdb-4.5.so. I booted the rescue media in order to install it, but I
| discovered that it wasn't even on the media. Long story short, the
| following packages were not included on the media or even in the package
| download cache:
|
| compat-db45
| hunspell
| hunspell-en
|
| I'm sure that pungi is inevitably to blame, but I'm wondering if we
| shouldn't add those packages to fedora-livecd-desktop.ks and/or
| fedora-live-base.ks in the meantime.
|

Could this be considered a packaging issue maybe?

- -Jeroen
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iEYEARECAAYFAkkohC4ACgkQKN6f2pNCvwjAZwCggtAJxtgxBRCujtf3c3px7POO
> 5CIAoLyD77ZLfzxnBAWIeToPGZcaGGNa
> =3iT2
> -----END PGP SIGNATURE-----

[Jesse Keating]

On Fri, 2008-11-21 at 21:21 -0500, Ignacio Vazquez-Abrams wrote:
>  
>  compat-db45
>  hunspell
>  hunspell-en
>  
>  I'm sure that pungi is inevitably to blame, but I'm wondering if we
>  shouldn't add those packages to fedora-livecd-desktop.ks and/or
>  fedora-live-base.ks in the meantime. 
Does your compose log mention any missing deps?  A fresh install with
the media you created, does yum report any problems with the rpmdb?
This could simply be a case of missing deps recorded in the packages
themselves, and if the deps are missing, pungi can't know to grab 'em.

However given that installing from live manages to work, it really
sounds like there is a discrepancy between what pungi gathers + anaconda
installs vs what livecd-creator installs.  This could be bad, although
it is know that pungi does things slightly different at the gather stage
(gathers all possible resolvers to a given dep rather than the "best"
resolver.  Deciding "best" is delayed until actual anaconda
installation)

-- 
Jesse Keating
Fedora -- Freedom² is a feature!
identi.ca: http://identi.ca/jkeating

[Ignacio Vazquez-Abrams]

On Sun, 2008-11-23 at 11:32 -0800, Jesse Keating wrote:
>  On Fri, 2008-11-21 at 21:21 -0500, Ignacio Vazquez-Abrams wrote:
>  > 
>  > compat-db45
>  > hunspell
>  > hunspell-en
>  > 
>  > I'm sure that pungi is inevitably to blame, but I'm wondering if we
>  > shouldn't add those packages to fedora-livecd-desktop.ks and/or
>  > fedora-live-base.ks in the meantime.
>  
>  Does your compose log mention any missing deps? 
Quite a few. Here are the relevant ones:

$ grep "Unresolvable dependency .*\(hunspell\|db\)" i386.log | sort -u
Pungi.WARNING: Unresolvable dependency compat-db45 in rpm.i386
Pungi.WARNING: Unresolvable dependency hunspell-en in libsexy.i386
Pungi.WARNING: Unresolvable dependency libdb-4.5.so in rpm.i386
Pungi.WARNING: Unresolvable dependency libdb-4.5.so in rpm-libs.i386
Pungi.WARNING: Unresolvable dependency libhunspell-1.2.so.0 in
enchant.i386

>  A fresh install with the media you created, does yum report any
>  problems with the rpmdb? 
Once I put libdb-4.5.so in place so I could access the rpmdb,
package-cleanups --problems reported those 3 packages missing.

>  This could simply be a case of missing deps recorded in the packages
>  themselves, and if the deps are missing, pungi can't know to grab 'em. 
The packages seem to be excluded a bit further up:

http://ivazquez.fedorapeople.org/files/pungi-i386.log

-- 
Ignacio Vazquez-Abrams <ivazqueznet(a)gmail.com&gt;

PLEASE don't CC me; I'm already subscribed

[Jesse Keating]

On Sun, 2008-11-23 at 22:45 -0500, Ignacio Vazquez-Abrams wrote:
>  
>  The packages seem to be excluded a bit further up:
>  
>  http://ivazquez.fedorapeople.org/files/pungi-i386.log
>   
Well now that's interesting.  I wonder if pungi and livecd-creator
handle excludes differently.  Definitely worth checking out, thanks for
bringing this up.

-- 
Jesse Keating
Fedora -- Freedom² is a feature!
identi.ca: http://identi.ca/jkeating

[Jeroen van Meeuwen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Jesse Keating wrote:
| On Sun, 2008-11-23 at 22:45 -0500, Ignacio Vazquez-Abrams wrote:
|> The packages seem to be excluded a bit further up:
|>
|> http://ivazquez.fedorapeople.org/files/pungi-i386.log
|>
|
| Well now that's interesting.  I wonder if pungi and livecd-creator
| handle excludes differently.  Definitely worth checking out, thanks for
| bringing this up.
|

Yeah, pungi and livecd-tools do treat excludes differently; I've brought
up a similar issue before:
https://www.redhat.com/archives/fedora-livecd-list/2008-July/msg00106.html

pungi extends the yum exclude list, eliminating the packages from the
package sack (src/pypungi/__init__.py:339), and those will thus not be
available during package selection, nor dependency resolving, nor
whatever other stage except buildinstall.

livecd-tools only removes the packages in the exclude list after the
complete groupLst and packageLst have been selected. Ergo, after
dependency resolving, "excluded" packages may have been pulled back in.

Kind regards,

Jeroen van Meeuwen
- -kanarip
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iEYEARECAAYFAkkrxAIACgkQKN6f2pNCvwiTSACfdLXJsHJ+rep/yM3gcfPFncJ1
> DGQAniLw/BrrPoKYwFhGfFj3M6+LMj9I
> =lL1Y
> -----END PGP SIGNATURE-----

[Bryan Kearney]

Jeroen van Meeuwen wrote:
>  -----BEGIN PGP SIGNED MESSAGE-----
>  Hash: SHA1
>  
>  Jesse Keating wrote:
>  | On Sun, 2008-11-23 at 22:45 -0500, Ignacio Vazquez-Abrams wrote:
>  |> The packages seem to be excluded a bit further up:
>  |>
>  |> http://ivazquez.fedorapeople.org/files/pungi-i386.log
>  |>
>  |
>  | Well now that's interesting.  I wonder if pungi and livecd-creator
>  | handle excludes differently.  Definitely worth checking out, thanks for
>  | bringing this up.
>  |
>  
>  Yeah, pungi and livecd-tools do treat excludes differently; I've brought
>  up a similar issue before:
>  https://www.redhat.com/archives/fedora-livecd-list/2008-July/msg00106.html
>  
>  pungi extends the yum exclude list, eliminating the packages from the
>  package sack (src/pypungi/__init__.py:339), and those will thus not be
>  available during package selection, nor dependency resolving, nor
>  whatever other stage except buildinstall.
>  
>  livecd-tools only removes the packages in the exclude list after the
>  complete groupLst and packageLst have been selected. Ergo, after
>  dependency resolving, "excluded" packages may have been pulled back in. 
Is one of these approaches "correct"?

--bk

[Jesse Keating]

On Tue, 2008-11-25 at 10:23 +0100, Jeroen van Meeuwen wrote:
>  pungi extends the yum exclude list, eliminating the packages from
> the
>  package sack (src/pypungi/__init__.py:339), and those will thus not be
>  available during package selection, nor dependency resolving, nor
>  whatever other stage except buildinstall.
>  
>  livecd-tools only removes the packages in the exclude list after the
>  complete groupLst and packageLst have been selected. Ergo, after
>  dependency resolving, "excluded" packages may have been pulled back in. 
Thinking about it some...

Ok, I think it all depends on where the exclude is encountered.  If an
exclude is encountered at a repo declaration line, that should be passed
into the yum object like pungi is doing.  But if the exclude (or -foo)
is encountered in %packages, it should be treated differently, it should
be treated the same way they would be treated should the kickstart file
be used with anaconda to perform an install.  That is the way that
livecd-tools is doing it, so that things with - signs could still get
pulled in if they are needed to resolve deps of other things selected.

Obviously I have a bug in pungi with this, I'll be looking into it.

-- 
Jesse Keating
Fedora -- Freedom² is a feature!
identi.ca: http://identi.ca/jkeating