Cross-Cutting Concerns in Library Design

A mental framework for library design

For those with plenty of experience managing complexity in large complex codebases, this post will likely be nothing new. However many open-source libraries, frameworks, and tools make mistakes in how they handle cross-cutting concerns and end up being difficult to use as a result. I’m no stranger to this, and have several times found myself unsatisfied with the design of a library that I’ve created only to realise that it’s due to mishandling of cross-cutting concerns.

This post is a not a set of rules, but rather a framework for thinking about the design of libraries and tools. It’s also not intended to be the only framework used to think about the design, there are lots of ways of slicing the design problem that each provide value in a different way.

The post focuses on libraries as this issue tends to matter more at the point of integration between systems, but much could apply to frameworks and tools, the line is often blurred between these anyway.

Careful consideration of which cross-cutting concerns the code has an opinion on, which it defers to the user, and which don’t apply, will lead to code that is more usable and that is a better citizen in the ecosystems it’s a part of.

What are cross-cutting concerns and why do they matter so much when building libraries? The term “cross-cutting concern” originates from Aspect Oriented Programming where it has a more specific meaning, but here it’s used to mean shared concerns that affect multiple areas of the code – that “cut across” the core functionality with supporting, secondary functionality.

Logging, configuration, connection pooling, authorisation – there are many that crop up time and time again. The reason they matter for library design is that if these concerns don’t line up with the contexts in which the libraries are being used, it creates an impedance mismatch that makes integration harder or impractical.

Taking logging as an example, there are many different ways of using logs.

  • Some teams don’t use logs, their usefulness depends on what you’re building.
  • Other teams might only use logs in development and be happy with any output that helps them debug.
  • Some may want all their logs to be written to disk and managed with logrotate, necessitating certain file handle use.
  • Others may be legally required to store their logs centrally in specified formats for a minimum period of time for auditing purposes.

Authors of open source packages are usually trying to solve a problem they have. An author in the first group may not include any logging, preventing others from using it. An author in the second group may write their own file handling making it difficult for those in the later groups to control their logging. An author in the last group may write a package that requires so much logging configuration that the first two groups would find the package unapproachable.

None of these issues have anything to do with the core functionality of the package, they just take an opinion on a cross-cutting concern that is accidentally incompatible with the requirements of some users.

Worked example

To further illustrate the point, consider a Twitter API client library. It provides a language-native interface to the Twitter API in the language of your choice, turning raw HTTP requests and responses into functions, classes, methods, or another language appropriate interface.

Each cross-cutting concern needs to be handled in one of three ways…

Irrelevant concerns

Handling irrelevant concerns is by far the easiest, they don’t need handling. The important thing is to be aware of the existence of the concerns and to knowingly ignore them.

Example: code discovery

There is no concept of finding units of code for a Twitter API library. A plugin system doesn’t make sense, but double checking whether it makes sense and actively deciding to ignore this concern is important.

Example: logging

A gotcha to be avoided here is that in some cases ignoring is as good as not-supporting, which is itself an opinion on the concern that will limit who can use some code. While it may be reasonable to some for a Twitter API library to not have any logging in it, pushing responsibility to the callsite, there may be use-cases that require logging on network calls or at some other point, and not having any logging excludes these use-cases.

Opinionated concerns

Next easiest is probably the concerns that the library is going to have an opinion about. Again these are fairly easy because by deciding to take ownership of the decisions, the author is able to achieve these however they choose.

Some of these are uncontroversial, but for many the tricky bit is not the implementation, but making the right decision and backing it up.

Example: service discovery

This example is likely uncontroversial. A Twitter API client could allow for defining API endpoints for arbitrary services that happen to be implementing the API contract, but a library that hard-codes this to is unlikely to cause issues for most. In a way this is a core part of the library, not a cross-cutting concern, and therefore it’s reasonable for a library to have an opinion on it rather than making it configurable.

Example: concurrency

This example however could be much more complex, depending on the language. Most of the Python ecosystem is still using synchronous code, while newer codebases for things like web services that are often I/O bound are starting to use asynchronous I/O to improve throughput. Supporting both is often difficult so many libraries decide to either be synchronous or asynchronous. Another example is the use of Promises or callbacks in the Node JS ecosystem.

While neither of these are insurmountable, it’s possible to use a library designed for one in the environment of another, the “glue code” to make that work is more code to maintain and can often be challenging to write.

Being opinionated on a cross-cutting concern usually makes sense for details that don’t matter, and a small number of major details, where it may be possible for an alternative open-source package to fill the space left on the other side of the decision. This typically does not make sense for large numbers of decisions in a piece of software, unless it’s a large framework. For the Twitter API library example it would be reasonable for it to be asyncio based, or Promise based, or the equivalent for other ecosystems, and to leave it up to alternative libraries to fill the other use-cases.

Example: secrets management

An example of a concern that should probably not be opinionated for the case of a Twitter API client library would be credentials management.

One possible design would be to read the API key from a file on disk in a specific place. This would be easy to use, but raises a number of questions: How does the file get there? What are the permissions on the file? Where do the credentials pass through to get there? Who else has access because it’s on disk? Each of these could prevent a user from using this library either due to technical constraints or security policy constraints.

Taking the API key as an argument to functions in the library is likely a much better decision as that shifts responsibility to the user, allowing them to use the file strategy if they like, or environment variables, or an existing config or secrets management system.

Unopinionated concerns

Lastly there are those concerns that are unopinionated. These are often the hardest, because to remain neutral on them means creating the extensibility necessary to hand off responsibility to the user, and because it’s so easy to miss something and unintentionally take an opinion.

What engineers identify here will often depend on what they’ve had issues with in the past – if they have never worked on a codebase with translation support they may not consider it a high priority or may forget about it as a concern entirely.

Unopinionated concerns are hard to handle because of the myriad of ways to hand off responsibility to the user. This could be as simple as adding an argument to a function so that the user can pass in some data, or as complex as a plugin system so that users can implement plugins that interface with systems unknown to the library author. Developing an instinct for the best solutions to these problems typically means having a wide experience of the particular ecosystem the code exists in.

Example: logging

Python comes with its own built-in logging system. Because of this, the best choice for software written in Python that wants to be unopinionated about logging is to use the built-in logging system. This ensures that the user has control over the logging in a well-defined and documented way and that it plays the part of a good citizen in the ecosystem.

Opinions on Python’s logging system are mixed, and so it would be easy for an author to believe they can do it better, but the fact that it is standardised between most libraries and frameworks means that there’s an ecosystem of components that replace the core logging functionality and which can be used without needing support from the library author. This is a great example of the benefits of playing nicely with the ecosystem.

In the Twitter API library example, it would be important to choose a logging mechanism that is most likely to fit with the rest of the ecosystem. Allowing the user to control formatting and log redirection is important, as is naming log sources such that they can be filtered if necessary.

Example: serialisation

Serialisation and deserialisation formats are often decided by external systems so not something that can be changed. However this isn’t always the case.

To use the Twitter API library example one last time – Tweets may be returned as objects with properties, and it may be necessary to persist these objects to some form of storage, maybe an on-disk cache. It would be easy to implement a to_json method that returns a serialised string but there are many cases where this isn’t an appropriate format. A better alternative may be to provide a public interface for all the state to be read out of the Tweet, and another to re-construct that Tweet from the raw data. This would allow users to implement their own serialisation and deserialisation however they like, but in some languages this may not be very ergonomic or may require a lot of boilerplate.

Swift has a language defined protocol called Codable that allows any object that implements it to be serialised/encoded by any other object that implements Encoder, without each requiring knowledge of the other. The Swift version of the Twitter API library should probably implement Codable for Tweet and let users choose the encoder.

While these are two examples of “unopinionated” concerns, they are still in fact opinionated in that they force the use of Python logging or Swift’s Codable, each decisions that will limit usability in some way.

The best choices here come from a deep experience in an ecosystem – understanding how libraries and tools interact and how they are used in order to find the best way of relinquishing responsibility to the user for each cross-cutting concern. There’s no precise definition of what’s opinionated and what’s unopinionated, it’s up to the standards of the ecosystems – the languages, frameworks, operating systems, communities, and organisations.

This isn’t the only way to think about design, in fact there’s nothing here about how to design the core functionality of a library or tool, but hopefully this is a useful mental framework or thought experiment that can be used to check the suitability of design ideas.

Cross-cutting concerns to consider

These are just a few that came to mind while writing this post. I’ll be referring back to this list when I write my own libraries and tools.

  • Logging
  • Metrics
  • Tracing
  • Authorisation
  • Dates, Times, Timezones
  • Localisation, Internationalisation, Translation
  • Accessibility
  • UI Styling
  • Database access
  • Credentials – where they are stored and security requirements
  • Configuration – location, format, support for hot-reloading
  • Execution control – threads, green-threads, promises, futures
  • Code discovery – plugins, test discovery
  • Scheduling – cron or time-based scheduled tasks
  • Service discovery
  • Connection management – TCP, HTTP, databases, connection pooling
  • File storage – filesystem, cloud storage
  • Serialisation
  • Randomness – controllable sources, seeding of pseudo-random sources