Quarks

A collection of source-only NuGet packages representing tiny bits of functionality that don't belong together in a monolithic "utility" library. These are typically delivered as self-contained source files added to your projects as internal classes that can be easily kept up-to-date with NuGet.

The name "Quarks" was chosen for these packages' root namespace to signify that they are very small, fundamental building blocks in your application, just like in quantum physics:

A quark is an elementary particle and a fundamental constituent of matter.

The inspiration for creating this collection of packages was born out of my own frustrations related to bundling miscellaneous helpers into a monolithic library, seeing it done before in NETFx, and from reading a blog post titled Packaging Source Code With NuGet. I decided to create this project instead of contribute to NETFx because I was frustrated with how cumbersome it is to contribute new packages to NETFx, as well as the fact that NETFx packages are not marked as development dependencies.

Why

Who doesn't have an ever growing and ever less cohesive miscellaneous collection of helpers, extension methods and utility classes in the usual "Common.dll"? Well, the problem is that there's really no good place for all that baggage: do we split them by actual behavioral area and create "proper" projects for them?

In most cases, that's totally overkill and you end up in short time with the same pile of assorted files as you try to avoid setting up an entire new project to contain just a couple cohesive classes.

But it turns out that in the vast majority of cases, those helpers are just meant for internal consumption by the actual important parts of your code. In many cases, they are just little improvements and supplements over the base class libraries, such as adding missing overloads via extension methods, adding factory methods for otherwise convoluted object initialization, etc. It's almost inevitable that as the .NET framework and its languages evolve, existing APIs will start to look dated and lacking (i.e. lack of generics from v1 APIs, or lack of async-friendly Task-based APIs, etc.).

But with the advent of NuGet there's a new way to maintain, evolve and share those useful little helpers: just make them content files in a NuGet package!

And thus Quarks was born: a repository of the source and accompanying unit tests for all those helpers, neatly organized by target namespace being extended, deployed exclusively using NuGet, and licensed entirely under MIT for everyone to use and contribute.

Quarks package conventions and practices

Quarks NuGet packages follow a number of useful conventions which make them less intrusive and more consistent/predictable when installed in target projects:

• If the package is a class definition which derives from an existing class, the package ID should be prefixed with "Quarks", followed by the namespace of the class being derived from, followed by the name of the class. For example, "Quarks.System.Web.FakeHttpResponse" derives from System.Web.HttpResponseBase.

• If the package is a class definition which doesn't derive from an existing class, the package ID should be prefixed with "Quarks", followed by the namespaced name of your class. For example, "Quarks.AppSettings" or "Quarks.Page" in the "root" namespace, or "Quarks.Machine.Fakes.ConfigForASystemTimeOf" in the "Quarks.Machine.Fakes" namespace.

• If the package is an extension method, the package ID should be prefixed with "Quarks", followed by the target type you are extending (with an "Extensions" suffix added), followed by the name of the extension method. For example, "Quarks.StringExtensions.Contains" extends System.String with a method named Contains. If the type you are extending is in the root System namespace, you can omit the "System" part in the package ID. For the class that contains your extension method, it should be named in the singular and defined as partial, so for "Quarks.StringExtensions.Contains", the class is defined as static partial class StringExtension. If the extension method is extending an interface rather than a class, the same rule applies except that you drop the I prefix from your extension method class, so for "Quarks.IEnumerableExtensions.Contains", the class is defined as static partial class EnumerableExtension. This allows other extension method packages for the same type to share the same class name for consistency and ease of discovery.

• Pick a name that is very specific to the helper/extension method you're creating. For example, instead of "Quarks.StringExtensions" containing two or more extension methods, create a "Quarks.StringExtensions.SomeMethod" package for each extension method individually. This rule minimizes the chances of a single package becoming too big and evolving into its own "Common.cs" hell.

• Each Quarks package is contained in a single .cs file, with the corresponding .nuspec file having the same name and nested beneath it using the <DependentUpon> syntax in the .csproj file. A useful Visual Studio extension called File Nesting makes this easy to do. The .nupsec file should specify that the package is a development dependency in the metadata, and the target of the file should be a folder that matches the namespace of the package.

• Quarks packages should only specify dependencies on other development dependency packages, be they other Quarks packages or other source-only packages (or even build-time tools etc.). It's okay to have code in a Quarks package that depends on another NuGet package that is not a development dependency, but in that case it should just omit that dependency in the .nuspec file. The reason for this is that the consumer of the Quarks package would necessarily already depend on a version of that dependent package which is being extended, thus allowing ultimate flexibility regarding versioning.

• Types and methods in Quarks packages should be marked as internal in order to not pollute the target code's public API.

When to create a Quarks package

There are plenty of situations where a Quarks package does not make sense. Here's a few things to consider:

• TENTATIVELY consider creating a Quarks package for "utility" libraries that feature heavy usage of static and/or extension methods. Examples of these types of utility libraries include unit test assertion libraries and the popular DataAnnotationsExtensions package. However, the idea behind Quarks is best suited to individual classes or methods rather than full featured libraries.

• TENTATIVELY consider creating a Quarks package for small single purpose libraries. SimpleJson is already doing this (though not in the Quarks namespace and not following these conventions) but you can imagine any code appropriate for a blog post or Gist would fit the definition well. However, the idea behind Quarks is best suited to individual classes or methods rather than full featured libraries.

• DO consider creating a Quarks package for common configuration and setup code or any code which will require tweaking by the user.

• DO NOT consider creating a Quarks package as a means to simply make step-debugging easier. Instead leverage a symbols package.

Modifying the code deployed by a Quarks package

The way that source-only packages work with NuGet, permits a user to modify the code that was delivered when the package was installed. These modifications will persist when a NuGet package restore happens, but when updating the package these changes will be overwritten (after accepting a prompt to do so). In that case, a user can then merge their changes into the new code (if they are still relevant).

NuGet 3.1 deprecated support for delivering elements in the /content folder of packages

Starting with NuGet 3.1 when using project.json, support for delivering elements in the /content folder of packages has been deprecated. Unfortunately this means that currently, Quarks packages won't work out-of-the-box for projects based on the new project.json format. However according to this blog post, "Content files can still be carried inside the packages, and will be ignored currently, however the end user can still copy them into the right spot.", so this method can be used as a temporary measure for projects based on the new project.json format.

There's an issue currently being discussed that proposes to bring back support for content files. I'll be watching this issue closely to see if and when it gets implemented. If however it doesn't get implemented, I'll be looking into migrating this suite of packages to Bower.