Deciphering one of dpkg’s weirdest errors: unable to open ‘/path/to/foo.dpkg-new’

We already studied one weird error of dpkg, let’s do another one:

Unpacking replacement libexo-common ...
dpkg: error processing /var/cache/apt/archives/libexo-common_0.6.1-1_all.deb (--unpack):
 unable to open '/usr/share/doc/exo/html/C/images/exo-preferred-applications-internet.png.dpkg-new': No such file or directory

Rather difficult to understand on the first look right? Let’s see in detail what’s usually happening when you get this error.

The first hint comes from the file extension “.dpkg-new”. This extension is used by dpkg to unpack the updated files near the old files. When everything has been unpacked, they are renamed over the old files.

The failure happens precisely when dpkg tries to rename the file (in fact when it tries to fsync() it before the rename)… but why does it fail?

Usually because there are unexpected symlinks (or bind mounts) involved that resulted in two different files being installed in the same directory. For example consider package that provides /dir1/a-file and /dir2/a-file. Now imagine that on the target system, /dir1 is a real directory but /dir2 is a symbolic link that points to /dir1.

When dpkg processes /dir1/a-file.dpkg-new everything is fine, but when it tries to process /dir2/a-file.dpkg-new it will fail because that file is the same than /dir1/a-file.dpkg-new which has already been renamed.

Diagnosing further the problem requires to understand why there’s a symlink instead of a real directory. It might be two packages that were badly coordinated, or a problem in the package itself because it lacks some code that drops the symlink in the preinst (so that dpkg installs the real directory instead of keeping the symlink).

There might be variations in the way two files end up sharing the same directory, but this simple example should have clarified the nature of the underlying problem.

Subscribe to this blog by RSS, by email or on Facebook.

How to prepare patches for Debian packages

You want to start contributing to Debian and/or Ubuntu, you decided to help a package maintainer and you’re now looking for how to change a source package and how to submit your changes.

1. Retrieve the source package and install build-dependencies

The first step is to retrieve the latest version of the source package and to install the required build-dependencies. I already covered how to do this with apt-get in the article explaining how to rebuild a source package.

If you prefer, you can use dget (from the devscripts package) to directly grab the source package. You can find the URL of the .dsc files in the Package Tracking System for example.

Sometimes apt-get will warn you that the source package is maintained in a VCS repository, like this:

$ apt-get source wordpress
[...]
NOTICE: 'wordpress' packaging is maintained in the 'Git' version control system at:
git://git.debian.org/git/collab-maint/wordpress.git
[...]

In that case, you can use debcheckout to retrieve the VCS repository instead (provided that you have the corresponding VCS installed):

$ debcheckout wordpress
declared git repository at git://git.debian.org/git/collab-maint/wordpress.git
git clone git://git.debian.org/git/collab-maint/wordpress.git wordpress ...
Cloning into wordpress...

Note however that some maintainers use their VCS in a way that’s not really compatible with the explanations that I will give below.

It’s also a good idea to install the package “packaging-dev”. It’s a meta-package depending on the most common tools that are used for Debian packaging work.

2. Do the changes

Execute dch --nmu to record the fact that you’re working on an update prepared by someone who is not the maintainer (NMU means Non Maintainer Upload). This also ensures that if we build the package, we won’t overwrite the original source package that we downloaded, thus making it possible to generate a “diff” between both versions.

2.1. Modify Debian packaging files

Now fire your text editor and do the required changes in the “debian” sub-directory. You will probably run dch -a multiple times to document each subsequent change.

2.2. Modify upstream files

If you have to modify upstream files, the proper way to do it depends on the source package format (“1.0” vs “3.0 (quilt)” vs “3.0 (native)”, see the debian/source/format file) and on the presence or not of a patch system (the what-patch can help you identify it). In this explanation, I’ll assume that the package is using the recommended format: “3.0 (quilt)”. (It also works for “1.0” if quilt is used and if you configured ~/.quiltrc as recommended by /usr/share/doc/quilt/README.source).

First you should ensure that all patches have been applied with quilt push -a. If there’s no patch yet, you want to create the debian/patches directory (with mkdir debian/patches). Note that you should better invoke quilt from the root of the source package (and the examples below assume this).

2.2.1. Import a patch

If the upstream changes are already in a patch file (say /tmp/patch that you downloaded from the upstream VCS repository) you can import that patch like this:

$ quilt import -P fix-foobar.patch /tmp/patch
Importing patch /tmp/patch (stored as fix-foobar.patch)
$ quilt push
Applying patch fix-foobar.patch
[...]
Now at patch fix-foobar.patch

The -P option allows to select the name of the patch file created in debian/patches/. As you see, the new patch file is recorded in debian/patches/series but not applied by default, we’re thus doing it with quilt push.

2.2.1. Create a new patch

If the upstream changes that you want to make are not in a patch yet, you should tell quilt that you’re going to create one:

$ quilt new fix-foobar.patch
Patch fix-foobar.patch is now on top

Then you must record every file that you’re going to modify with a quilt add invocation. quilt then makes a backup of those files so that it can generate the patch later on. If you’re going to modify the files with your text editor you can just do quilt edit file-to-modify, it’s the same than quilt add file-to-modify followed by sensible-editor file-to-modify.

$ quilt edit foobar.c
File foobar.c added to patch fix-foobar.patch

The last step is tell quilt to generate the patch:

$ quilt refresh
Refreshed patch fix-foobar.patch

3. Test your changes

You should build your modified package with “debuild -us -uc”. You can easily install the resulting package with “debi”. Verify that everything works as expected. If not, continue your modifications until you’re satisfied with the result.

4. Generate a patch and mail it

If you followed the instructions, you should have two .dsc files in the parent directory, like this:

$ cd ..
$ ls wordpress_*.dsc
../wordpress_3.0.5+dfsg-1.1.dsc
../wordpress_3.0.5+dfsg-1.dsc

Generating the patch to send to the maintainer is then just a matter of running debdiff:

$ debdiff wordpress_3.0.5+dfsg-1.dsc wordpress_3.0.5+dfsg-1.1.dsc >/tmp/wp-debdiff

You can send the /tmp/wp-debdiff file to the wordpress maintainer. Usually you send it via the bugreport that your update is fixing and you add the “patch” tag to the report.

This can be automated with the nmudiff utility. By default it assumes that you’re using mutt but it can also directly feed the resulting mail to sendmail. The default text that nmudiff proposes assumes that you’re actually performing an NMU and that the result has been uploaded. If that’s not the case, you should edit the text and make it clear that you’re just sending a patch.

If you have been working in a VCS repository, instead of using debdiff you can simply use the diff feature integrated in your VCS (git diff, svn diff, etc.). But note that with a distributed VCS (like git/bzr/mercurial, unlike svn) you should probably have committed all individual changes in separate changesets. And instead of sending a single patch, you’re probably going to send a series of patches (though it might be easier to just upload your branch in a public repository and give the corresponding URL to the maintainer).

Found it useful? Be sure to not miss other packaging tips (or lessons), click here to subscribe to my free newsletter and get new articles by email.

Deciphering one of dpkg’s weirdest errors: short read on buffer copy

As a Debian/Ubuntu user, you’re likely to be exposed at some point to an error reported by dpkg. In a series of articles, I’ll explain some of the errors that you might encounter.

Some error messages can be confusing at times. Most of the error strings do not appear very often and developers thus tend to use very terse description of the underlying problem. In other cases the architecture of the software makes it difficult to pin-point the real problem because the part that displays the error is several layers above the one that generated the initial error.

This is for example the case with this error of dpkg:

Unpacking replacement xulrunner-1.9.2 ...
dpkg-deb (subprocess): data: internal gzip read error: '<fd:0>: too many length or distance symbols'
dpkg-deb: error: subprocess <decompress> returned error exit status 2
dpkg: error processing /var/cache/apt/archives/xulrunner-1.9.2_1.9.2.17+build3+nobinonly-0ubuntu1_amd64.deb (--unpack):
 short read on buffer copy for backend dpkg-deb during `./usr/lib/xulrunner-1.9.2.17/components/libdbusservice.so'

First, the decompression layer discovers something unexpected in the data read in the .deb file and dpkg-deb outputs the error message coming from zlib (“too many length or distance symbols”). This causes the premature end of dpkg-deb --fsys-tarfile that dpkg had executed to extract the .data.tar archive from the deb file. In turn, dpkg informs us that dpkg-deb did not send all the data that were announced (and hence the “short read” in the error message) and that were meant to be part of the file ‘/usr/lib/xulrunner-1.9.2.17/components/libdbusservice.so’.

That’s all nice but it doesn’t help you much in general. What you must understand from the above is that the .deb file is corrupted (sometimes just truncated). In theory it should not happen since APT verifies the checksums of files when they are downloaded. But computers are not infallible and even if the downloaded data was good, it can have been corrupted when stored on disk (for example cheap SSD disks are known to not last very well).

Try removing the file (usually with apt-get clean since it’s stored in APT’s cache) and let APT download it again. Chances are that it will work on the second try. Otherwise consider doing a memory and HDD check as something is probably broken in your computer.

Join my free newsletter and learn more tips for users. Or click here to support my work on dpkg with Flattr, consider subscribing for a few months.

apt-get, aptitude, … pick the right Debian package manager for you

This is a frequently asked question: “What package manager shall I use?”. And my answer is “the one that suits your needs”. In my case, I even use different package managers depending on what I’m trying to do.

APT vs dpkg, which one is the package manager?

In the Debian world, we’re usually thinking of APT-based software when we’re referring to a “package manager”. But in truth, the real package manager is dpkg. It’s the low-level tool that takes a .deb file and extracts its content on the disk, or that takes the name of a package to remove the associated files, etc.

APT is better known because it’s the part of the packaging infrastructure that matters to the user. APT makes collection of software available to the user and does the dirty work of downloading all the required packages and installing them by calling dpkg in the correct order to respect the dependencies.

But APT is not a simple program, it’s a library and several different APT frontends have been developed on top of that library. The most widely known is apt-get since it’s the oldest one, and it’s provided by APT itself.

Graphical APT front-ends

update-manager is a simple frontend useful to install security updates and other trivial daily upgrades (if you’re using testing or sid). It’s the one that you get when you click in the desktop notification that tells you that updates are available. In cases, where the upgrade is too complicated for update-manager, it will suggest to run synaptic which is full featured package manager. You can browse the list on installed/available packages in numerous ways, you can mark packages for installation/upgrade/removal/purge and then run in one go all the recorded actions.

software-center aims to be an easy to use application installer, it will hide most of the packaging details and will only present installed/available applications (as defined by a .desktop file). It’s very user friendly and has been developed by Ubuntu.

Of the graphical front-ends, I use mainly synaptic and only when I’m reviewing what I have installed to trim the system down.

Console-based GUI APT front-ends

In this category, I’ll cite only aptitude. Run without parameter, it will start a powerful console-based GUI. Much like synaptic, you can have multiple views of the installed/available packages and mark packages for installation/upgrade/removal/purge before executing everything at once.

Command-line based package managers and APT front-ends

This is where the well known apt-get fits, but there are several other alternatives: aptitude, cupt, wajig. Wajig and cupt are special cases as they don’t use libapt: the former wraps several tools including apt-get, and the latter is a (partial) APT reimplementation (versions 1.x were in Perl, 2.x are now is C++).

You’re welcome to try them out and find out which one you prefer, but I have never felt the need to use something else than apt-get and aptitude.

apt-get or aptitude?

First I want to make it clear that you can use both and mix them without problems. It used to be annoying when apt-get did not track which packages were automatically installed while aptitude did, but now that both packages share this list, there’s no reason to avoid switching back and forth.

I would recommend apt-get for the big upgrades (i.e. dist-upgrade from one stable to the next) because it will always find quickly a relatively good solution while aptitude can find several convoluted solutions (or none) and it’s difficult to decide which one should be used.

On the opposite for regular upgrades in unstable (or testing), I would recommend “aptitude safe-upgrade“. It does a better job than apt-get at keeping on hold packages which are temporarily broken due to some not yet finished changes while still installing new packages when required. With aptitude it’s also possible to tweak dynamically the suggested operations while apt-get doesn’t allow this. And aptitude’s command line is probably more consistent: with apt-get you have to switch between apt-get and apt-cache depending on the operation that you want to do, aptitude on the other hand does everything by itself.

Take some time to read their respective documentation and to try them.

Click here to subscribe to my free newsletter and get my monthly analysis on what’s going on in Debian and Ubuntu. Or just follow along via the RSS feed, Identi.ca, Twitter or Facebook.