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