What does it take to make packaging easier?

We often get contacted by people who wrote some scientific software and would like to see it being available from the NeuroDebian repository. Most of these developers don’t know what it actually means to package software, how time-consuming the process is, and who needs to do what to get things done.

There is already a lot of information about distribution packaging on the web, including “best-practices” documentation for upstream-developers, such as Debian’s upstream guide, and it would be pointless to repeat all these bits here. Instead we will focus on a few core aspects that we found to be the most ignored and misunderstood concepts that heavily influence the speed and complexity of the packaging process.

So what exactly is a package? The ultra-short summary: it is first and foremost an executable description of how to build software from source and install the resulting binaries, while ensuring compatibility and interoperability with other parts of a particular system. Hence, although Debian-based operating systems install packages containing binaries, the main focus of the packaging process is on the source package. The end-product of the packaging process is a source package that enables anybody to produce binary packages in a uniform and deterministic fashion, with an exhaustive description of its dependencies.

To understand how that is done any interested person needs to read a fair bit of documentation, for example a packaging tutorial, and the Debian policy – the ultimate guidelines for proper Debian packages. But albeit being an interesting read for anyone, it is actually the package maintainer‘s job to get all this done right. However, there are a few key aspects about packaging that any upstream developer should know, as they will make the difference between getting a package done in a few hours, or only after weeks, months, or years of tedious work. These are:

  • A reliable versioning scheme
  • Exhaustive documentation of involved licenses
  • Clear separation of 3rd-party code
  • An open ear

Have a deterministic version!

Software needs versions, packages need versions. A version is used to determine whether an update is available, whether a particular package can be used to satisfy a versioned dependency of another package, to associate bug reports with source code, and, of course, to tell users what version of a software they are running. Surprisingly, there is a lot of scientific software out there that doesn’t have a version, or pretends to have one, but the actual source code changes without a corresponding change in the version. This is wrong. If you want to have your software packaged, have a reliable versioning scheme:

  • Increment the version whenever you make a release of your project, and make a new release whenever you want to push out some changes to an existing release (even for tiny changes).
  • Put the version into the filename of your source or binary download.
  • And/or use a version control system (VCS) that helps you track changes and associates them with unique identifiers that could serve as a version (if nothing better is available).

If you don’t have a deterministic version, the package maintainer needs to come up with one. Such custom version could be based on the time-stamp of a download, or the last modification time of any file in the source distribution. But whatever the maintainer will come up with, it will take time to implement, and it will be different from what you do. This will make packaging more complex, and it will confuse users.

Be conscious about all licenses!

Licenses are important. They determine whether a 3rd-party, such as a Debian, is legally allowed to redistribute your software. They determine whether a package maintainer is allowed to modify software for improved system integration or bug fixing. They also have an impact on people’s motivation to contribute to a project – many people out there would rather not invest their precious time in software with restrictive licenses.

Be aware that the collection of all licenses in your source code form the legal terms of your software. We often see “open-source” software that is “free” (to use), but depends on, or includes source code that “may not be redistributed”. This is most likely wrong. Moreover, we encountered quite a few projects that didn’t pick any license at all. If you intend other people to use your tools this is wrong too, as no license typically means no permissions at all – not even to download or to use.

The package maintainer needs to sort all these things out, needs to make sure that redistribution doesn’t impose a legal threat to anyone (including repository mirror operators), and needs to make sure that people expecting free and open-source software only get free and open-source software.

If you want to facilitate the packaging process: Make sure that you are aware of all licenses covering the source code and other materials (such as documentation and images) in your software. Make sure to document them properly. Make sure that the licenses covering the source code are compatible to each other (e.g. you cannot release under the GPL if the license of some other part of your code says “must not be used on Thursdays”). The easiest way to avoid unnecessary complications is to use a standard license (such as BSD, or GPL) that are known and have been evaluated by legal experts regarding their implications and their compatibility with each other. For any license, check if your legal terms comply with the open-source definition or the Debian free software guidelines (that are the basis of this definition).

Allow for modularity!

Usually, software authors already have means to build their code for Debian or Ubuntu systems before contacting a potential packager. However, quite often these procedures need to be adjusted (and sometimes even abandoned) for a distribution package. A prominent reason is the inclusion of 3rd-party source code. Virtually all software is built on top of some other software – may it be a GUI toolkit or a numerical library. Occasionally, it makes sense to include the more exotic dependencies into a source distribution of some software, mainly to make building and installing easier for people on less fortunate platforms.

However, for a distribution package such setup is typically not acceptable. An operating system like Debian is a modular system were duplication needs to be avoided. There should only be a single copy of a particular library or tool in the distribution. All packages that require a piece of 3rd-party software need to declare a dependency on the corresponding package and should not ship their own copy. This has the advantage that bugs in a software can be fixed in a single place and all dependent software automatically benefits from this fix, without further human intervention. Only a modular design like this makes it possible to successfully maintain an integrated system of tens of thousands of software packages with a reasonable amount of manpower.

So if your source distribution contains 3rd-party software, clearly identify its origin and version (the license as well, of course). This informs the packager which dependencies need to be declared, or potentially which other software still needs to be packaged separately, in order to serve as a dependency.

As a consequence of the dependency system, you should avoid modifying 3rd-party software. If you, for example, take the source of a library and modify it a bit to make it easier for you to do a certain task, there is suddenly the need for having two almost identical libraries in the operating system. The package for your software cannot use the system package for this library, because it is modified. Instead of keeping a modified version forever: if you need to fix bugs in 3rd-party software, make sure to forward the fixes to the original developers. If you need to adapt its behavior, try to do it in a way that is modular, keeping the adaptor separate from the 3rd-party code. This will make it much easier for you to track future developments of this code, as well as help the packager integrate your software into the operating system.

Also, you will significantly facilitate the packaging process if your build-system allows to optionally build and link against system libraries instead of the convenience copies that may be included in your source code. Keep in mind that anything that is required for packaging software for Debian needs to be added or modified by the package maintainer. All modifications can potentially change the behavior of your software and may confuse users and/or result in unnecessary support requests that need to be dealt with. Be assured that it is in the very interest of the package maintainer to keep the differences minimal. If you keep modularity aspects in mind while developing, you can massively facilitate a packaging effort.

Be prepared for feedback!

The package maintainer might send you a few patches during the initial packaging that either fix bugs on the Debian platform or that were added to gain compliance with the Debian policy. Be prepared to evaluate these patches and merge them into your code base or discuss necessary modifications. The package maintainer needs to keep track of all modifications done to your software and needs to refresh them for every new release that is made. If you make it easy for the maintainer to do this work, for example quickly merging modifications, exposing a version control system to track modifications, or at least offering a reliable communication channel that informs the maintainer about the fate of the patches, you will help to streamline long-term package maintenance and contribute to a reliable package. All this will help disseminating your software in an extremely convenient form to a very large audience.

On a final note: If you keep these things in mind you won’t only make a packager’s life easier. You will also have removed most hurdles for a smooth packaging in general. Now you could actually think about doing the packaging yourself. Take a look at a packaging tutorial to get a sense of what it would involve. If you decide to venture down this road, you are very welcome to contact us@NeuroDebian – we would be glad to guide you through an efficient packaging process and upload fully packaged software for psychological and neuroscience research into the main Debian archive and the NeuroDebian repository.

Comments

blog comments powered by Disqus