Bindep is a tool for checking the presence of binary packages needed to use an application / library. It started life as a way to make it easier to set up a development environment for OpenStack projects. While OpenStack depends heavily on pip for installation of Python dependencies, some dependencies are not Python based, and particularly for testing, some dependencies have to be installed before pip can be used - such as virtualenv and pip itself.


Create a file called bindep.txt and in that list any requirements your application / library has. In your README or INSTALL or other documentation you can tell users to run bindep to report on missing dependencies. Users without bindep installed can consult the bindep.txt file by hand if they choose, or install bindep first and then use it.

If no bindep.txt file exists, bindep will look at the old location other-requirements.txt.

The output from bindep is fairly verbose normally, but passing an option of -b/–brief outputs just the missing packages one per line, suitable for feeding to your package management tool of choice.

If you need to maintain multiple requirements list files you can pass a specific filename with the -f/–file command line option. If you want to read the list from standard input in a pipeline instead, use a filename of “-“.

When bindep runs, its exit code is 0 if no described packages are missing, but 1 if there are packages which it believes need to be installed.


Profiles can be used to describe different scenarios. For instance, you might have a profile for using PostgreSQL which requires the PostgreSQL client library, a profile for MySQL needing that client library, and a profile for testing which requires both libraries as well as the servers. To select a profile just pass it when running bindep - e.g.:

$ bindep test

When running bindep a single profile can be chosen by the user, with no explicit selection resulting in the selected profile being default. bindep will automatically activate additional profiles representing the platform bindep is running under, making it easy to handle platform specific quirks.

The available profiles are inferred by inspecting the requirements file and collating the used profile names. Users can get a report on the available profiles:

$ bindep --profiles

Writing Requirements Files

The requirements file bindep.txt lists the dependencies for projects. Where non-ascii characters are needed, they should be UTF8 encoded.

The file is line orientated - each line is a Debian binary package name, an optional profile selector and optional version constraints. (Note - if you are writing an alternative parser, see the Debian policy manual for the parsing rules for packagenames). Debian package names are used as a single source of truth - bindep can be taught the mapping onto specific packaging systems. Alternatively, profiles may be used to encode platform specific requirements.

Profiles are used to decide which lines in the requirements file should be considered when checking dependencies. Profile selectors are a list of space separated strings contained in []. A selector prefixed with ! is a negative selector. For a line in the requirements file to be active:

  • it must not have a negative selector that matches the active profile.

  • it must either have no positive selectors, or a positive selector that matches the active profile.

For instance, the profile selector [!qpid] will match every profile except qpid and would be suitable for disabling installation of rabbitmq when qpid is in use. [default] would match only if the user has not selected a profile (or selected default). [default postgresql test] would match those three profiles but not mysql. [platform:rhel] will match only when running in a RHEL linux environment.

Note that platform selectors are treated as kind of filter: If a line contains a platform selector, then the package only gets installed if at least one of the platform selectors matches in addition to the match on the other selectors. As an example, [platform:rpm test] would only install a package on a RPM platform if the test selector is used.

Profiles can also be grouped together using (). In a group, all profiles must match for the group to match. Given the example [test (ceph glance !lvm)], to select the package you must either specify test OR (ceph AND glance AND NOT lvm). Platform selectors will not work inside of the group.

Version constraints are a comma separated list of constraints where each constraint is (== | < | <= | >= | > | !=) VERSION, and the constraints are ANDed together (the same as pip requirements version constraints, except that they require a space between the dep and version constraints).

Comments are allowed: everything from the first # to the end of the line is ignored.


A simple example with using a test profile is:

# A runtime dependency
# A build time dependency
libffi-devel [test]

bindep would select the libffi6 package in all cases and if the test profile gets choosen with bindep test, then both packages would be selected.

If a repository needs for deployment the libxml2 development libraries for support of Debian, Gentoo, and RPM based distros, the bindep.txt file can contain:

libxml2-dev [platform:dpkg]
libxml2-devel [platform:rpm]
libxml2-utils [platform:dpkg]
dev-libs/libxml2 [platform:gentoo]

This would select libxml2-dev and libxml2-utils packages on Debian based distributions like Debian and Ubuntu since those entries have the platform:dpkg profile, libxml2-devel on RPM based distributions like CentOS, Fedora, openSUSE, Red Hat, or SUSE Linux since those entries have the platform:rpm profile, and dev-libs/libxml2 on Gentoo since the entry has the platform:gentoo profile.

Additionally, you can use platform:redhat or platform:suse to only match RedHat-like or SUSE-like distributions respectively as shown in the following example:

openssh-server [platform:redhat]
openssh [platform:suse]

If you need to distinguish between operating systems where python2 or python3 is the official interpreter, you can use base-py2 and base-py3 labels. Keep in mind that only one would be exposed for a specific operating system even if the system could support installation of multiple python versions:

python3-lxml [(platform:redhat platform:base-py3)]
python-lxml [(platform:redhat platform:base-py2)]

The example above will install lxml python modules on official python used by platform. Keep it mind that base-py[23] support is currently implemented only on Debian, Ubuntu, RedHat flavours and MacOS.

To select Python3 development packages, you can use:

python3-all-dev [platform:dpkg !platform:ubuntu-precise]
python3-devel [platform:fedora]
python34-devel [platform:centos]

This selects python3-all-dev on all Debian based distributions with the exception of Ubuntu Precise, python3-devel on Fedora and python34-devel on CentOS.

To select the curl package:

curl [!platform:gentoo]
net-misc/curl [platform:gentoo]

This selects the curl package on all distributions with the exception of Gentoo, and selects net-misc/curl on Gentoo only.

To select a package based on a group of profiles:

ceph-common [ceph]
python-rbd [(ceph glance)]

This selects the ceph-common package when the profile ceph is specified. However, it will only select the python-rbd package when both ceph and glance profiles are active.

To specify package versions:

python >=3.7,<=3.8
grep >=3.3

# OR with a platform profile
python [platform:brew] ==3.7.3

For a more comprehensive example check bindep.txt file that is used to test bindep itself on multiple platforms.