Managing dependencies in C and C++ projects is a common challenge and lots of different ways have been invented to deal with these issues, each with varying results. Conan aims to solve this challenge once and for all by providing a universal approach to managing dependencies with all the benefits of a package manager.
Managing dependencies in C and C++ projects has always presented a challenge. As such I have seen a variety of inventive ways to share C and C++ packages. A very non-exhaustive list of such approaches would cover the main strategies, such as:
Conan is a package manager designed for C and C++, meaning it supports multiple platforms and build tools, stores source code, as well as pre-built libraries/binaries in a remote repository. Conan, much like any other package manager, provides the simple concepts of creating, publishing and resolving packages, so there are no surprises there. However, most package managers do not need to deal with the C and C++ landscape - which means supporting multiple compilers, linkers and a vast array of build tools.
Conan has quite a simple yet powerful model comprising settings and options to deal with the different build environments (such as compilers and linkers). Settings typically define project- wide and environment related settings, such as the OS and compilers, whereas options define package-specific options that can have defaults (such as whether a package should be statically or dynamically linked).
If you build a package with a different setting or option, Conan will treat this as a different binary. This, on a very high level, is how Conan enables distributing pre-built binaries across multiple platforms and environments.
Where Conan also packs a punch is when it comes to integrating with multiple build tools. I like to think of it as Conan generates the "glue" for your build system to understand which packages Conan has resolved and where to find the relevant files (e.g. where to find header and library files).
With CMake, for example, Conan will generate a .cmake file that defines CMake targets for each of the required packages that can be used in your CMakeLists.txt file.
One interesting thing to note is that if your desired combination of settings and options has not yet been pre-compiled into a Conan package, Conan also stores the source code for creating the package. You can then build Conan packages as part of "resolving" them. And this does not just include direct dependencies, but also indirect ("transitive") dependencies. That's right: Conan of course supports transitive dependencies too!
For any framework to become successfully adopted and embraced in today's market there needs to be a thriving community, so let's talk a bit about the one surrounding Conan. For starters, all the source code is available on GitHub for you to explore.
The documentation is very nice and there is an active Slack channel in the Cpplang Slack workspace. JFrog are actively supporting the project with conferences Conan Days - and free online training is available in partnership with Verifa.
Last but not least, Conan Center is a publicly available Conan repository with a vast number of open source projects that have already been "Conanized". In other words, it's the process of taking an ordinary piece of source code and creating a Conan package out of it.
The Conan Center is populated by the conan-center-index repository on GitHub which contains all the Conan recipes for creating the packages.
Still not convinced Conan is worth taking a closer look at? Then let's talk about some of the pains that I have experienced in the past without it, and how Conan solves those problems.
Understanding what has gone into a final product in C and C++ can be really, really difficult, and often requires a lot of undesired process and documentation.
But what if there was a convenient file that just specified the dependencies (or requirements) and whether each dependency was used only for the build/test phase, or if they actually got shipped with the product?
Hmmm, it sounds like a package manager would solve this problem! Bonus points for tools like oss-review-toolkit that support Conan (thanks to a contribution from Verifa to help build the clearing process into Continuous Integration.
It is quite common for teams to be building a set of libraries or an entire platform, which application teams then develop their logic on top of. Often the application teams need to rebuild the libraries or platform because what they get delivered are not pre-built libraries, but the source code with instructions for how to include it in their projects.
What if the application team could simply specify the libraries or platform they want to use with a semantic version range, and then only had to compile their code and not the whole product? A versatile package manager could solve this by providing pre-built binaries.
This is perhaps a slightly more obscure use case for Conan, but I wanted to mention it. Lots of teams create a README or some instructions for setting up their development environment which includes things like the compiler and linter. What if you could specify these build requirements in a file, with versions, so that developers could set up their environments automatically?
As Conan is built on Python it can make use of Python's virtualenv to configure an environment (e.g. setting PATH) to point to binaries that could be included in Conan packages - your linter, for example.
It would not be fair to talk so highly about Conan without also considering some of the challenges I have observed, so let's cover that.
Conan is not the most simple tool; it has to provide a lot of features and flexibility and nothing in the C and C++ world is usually plug and play. That said, the Slack thread is available and the documentation for Conan is pretty good. Plus you can always reach out to us at Verifa about organising some training.
The second point to mention is that if you are developing lots of small components and need to make continuous changes to several of these components at the same time before build and test, then you need to consider the overhead of making each of these components a Conan package. It can be counterproductive and add overhead.
So the granularity of Conan packages should be considered, which is more of a design or architectural decision to make than necessarily being a drawback of Conan. Excitingly, this is a known topic and work is underway to help, such as Conan's workspace feature.
There are other package managers in the C and C++ space but as we do not have as much experience with them (added to the fact they are constantly changing), it is not easy to make any kind of fair comparison.
We have decided to settle for Conan as our choice of package manager because of the great support we have received from both the community and from industry tools like Artifactory to store pre-built binaries. This provides a great option for internal Conan repositories for large software projects.
In conclusion, if you are looking to solve some of the common challenges mentioned above I would strongly consider looking at Conan and running a proof-of-concept.
At Verifa we have helped several companies adopt Conan and in running proof-of-concepts on customer codebases to get an understanding of its benefits and how it affects your workflow. If you are interested in learning more, we suggest reading our What Is Conan? post. Alternatively, you can get in touch with us and we can provide all the necessary services and training to your team.
We look forward to hearing from you!
In Verifa’s first podcast we discuss Google’s new Autopilot for Kubernetes, and what it means to have a managed Kubernetes service.
From recruitment to product launches, it proved to be a most eventful 12 months
How to know your code is always in a releasable state
By submitting your email you agree to our terms and conditions.
We will never share your email with anybody else.