The most flexible file system abstraction, ever. Built in partnership with the UWP Community.
This package enables early adoption of CommunityToolkit.Storage, a package in the proposal stage in Toolkit Labs.
With the help of the open source community, we're continuing to iterate on, improve and test the concept using real working code, with additional freedom to make breaking changes if needed.
Published releases are available on NuGet. To install, run the following command in the Package Manager Console.
PM> Install-Package OwlCore.Storage
Or using dotnet
> dotnet add package OwlCore.Storage
If you're updating from a previous version, please see the release notes for breaking changes.
- In memory
- System.IO
- HttpFile
- StreamFile
- Zip archives (via System.IO.Compression)
- Windows.Storage
- IPFS / IPNS / MFS
- OneDrive
- Archives (7z, tar, etc) (via SharpCompress, third-party)
- NTFS (via NtfsReader, third-party)
- Audio Devices via Silk.NET OpenAL
Below is a recent copy of the proposal.
Or, visit the proposal directly
Listing a few off the top of my head: (click to expand)
On-disk APIs
- System.IO
- StorageFile
- Win32
Cloud storage
- OneDrive
- Google Drive
- Dropbox
- Box
- Mega
Network protocol
- FTP
- SMB
- DLNA
- IPFS
- WebDav
Blob storage
- Azure Blob
- Amazon S3
- Firebase
- Google Cloud
- Oracle Cloud
Plus countless custom HTTP based file systems. These are usually created in-house for a specific project, but they still count.
- These are all file systems, and there isn't a single good abstraction between them.
- Lack of a standard means a lot of tools and packages that don't work with each other.
- If you write code that touches one file system, it can't be swapped out for another without changing all your code.
- You can't easily mock a file system for unit tests, especially when code is hardcoded to a specific API.
This started out as an experiment called AbstractStorage. It was a 1:1 recreation of the most used parts of the StorageFile APIs, but in .NET Standard 2.0, and was created in Strix Music v2 before being moved into OwlCore, which enabled others to start using it and giving feedback.
Over time, we gathered feedback from those in the UWP Community who were using it: Fluent Store, Strix Music, Quarrel, ZuneModdingHelper, etc.
We learned a lot from the feedback we got, but the AbstractStorage experiment is now over. The next stage is in this proposal.
We'll use Labs to build a significantly improved version of AbstractStorage under the name CommunityToolkit.Storage
, and work to bring our experiment to the .NET Community Toolkit.
While this seems easy on the surface, we can't just turn the Windows StorageFile API into a set of ns2.0 interfaces and call it a day. We learned from our first version that this approach has issues once we put them to purpose.
We can do better.
Note
This proposal has been updated. Based on feedback, some major amendments have been made. For all the details, see: Amendment 1: Getting folder contents Amendment 2: Storage properties
Note
Due to popular request, we've made these interfaces (and a System.IO implementation) available for early adoption under OwlCore.Storage.
Toolkit Labs doesn't have an NS2.0 project template yet, and the Toolkit team is generally focused on other things. It could be a while before a published experiment is ready.
Without sacrificing potential functionality or performance, we should strive to:
- Maximize compatibility - with as many file systems as reasonably possible
- Minimize effort - for implementors, consumers and maintainers.
- Create the minimal possible interface for files/folders (the "starting point")
- Additional capabilities beyond basics should be done using derived interfaces (see examples below)
- As separate interfaces, new APIs break nothing. As a derived interface, implementations work with existing code.
- Breaking changes will affect as few implementations as possible
- Determining lack of support by checking
if (instance is IMyThing thing)
is faster and more natural than catching thrownNotSupportException
s.
- Implementors only need to implement what they can do, no more or less.
- Consumers only ask for what's needed, no more or less. Simply ask (or check) for that interface.
- We should have an inbox fallback implementation for certain things (like copy/move) that can work across different file systems.
- This allows us to align the majority of cross-storage implementations without requiring a base class
- Removes the need for everyone to do their own implementation
- Allows for a more performant version if present.
- We should consider file/folder implementations that differ from traditional on-disk systems (e.g. blob storage, mocked files)
Not all file systems are created equal, but they do share some common traits. To maximize compatibility, this should be our starting point.
Note These interfaces are named so developers use them by default when building with and for this API, unless they need more functionality.
public interface IStorable
{
/// <summary>
/// A unique and consistent identifier for this file or folder.
/// </summary>
/// <remarks>
/// Not all files are addressable by path.
/// Uri paths (like OneDrive) can change entirely when re-authenticating.
/// Therefore, we need a dedicated resource identifier.
/// </remarks>
public string Id { get; }
/// <summary>
/// The name of the file or folder, with the extension (if any).
/// </summary>
public string Name { get; }
}
public interface IFile : IStorable
{
/// <summary>
/// Opens a new stream to the file.
/// </summary>
public Task<Stream> OpenStreamAsync(FileAccessMode accessMode = FileAccessMode.Read, CancellationToken cancellationToken = default);
}
public interface IFolder : IStorable
{
/// <summary>
/// Retrieves the folders in this directory.
/// </summary>
public IAsyncEnumerable<IStorable> GetItemsAsync(StorableKind kind = StorableKind.All, CancellationToken cancellationToken = default);
}
[Flags]
public enum StorableKind
{
None = 0, // Required because it's a flag
File = 1, // Only get files
Folder = 2, // Only get folders
All = File | Folder, // Get all items.
}
Note See conversation below on why we went with a single method + enum for
IFolder
Even with just this, you can do a LOT. Keep in mind, these work with any local disk API, with any Cloud provider, with FTP, SCP, SMB, IPFS, HTTP and pretty much anything else. Plus -- you can mock them for unit tests.
Some examples of things I've made with just this:
- Store and read app settings, user documents, literally anything, in the place of the user's choice.
- Depth-first and breadth-first search tools
- Basic in-app file explorer / folder picker
- Audio file metadata scanner
- Duplicate file finder (across different sources)
- Mock file/folders used to unit test everything above.
Now that we have this, we can build on top of it.
There are 3 ways to obtain a file.
- Get it from a folder
- Get it standalone from other code.
- Create it yourself. Either 100% in memory, or by wrapping another file system.
A file, in its most basic form, must be a unique resource with a data stream and a name.
This description perfectly matches both in-memory files and some custom file types like blobs - none of which have a "containing" folder.
Examples of this:
(click to expand):
- Most blobs in blob storage have the characteristics of a File but aren't location-addressable. No path, no parent folder.
- Unit tests rely on abstractions to mock working implementations for the things you're NOT testing. This abstraction is created entirely in-memory to avoid usage of the actual file system. No path, no parent folder.
- For some libraries like taglib-sharp just need the concept of a file, and (already, without a standard) expect you to implement their
IFileAbstraction
. You can do this 100% in memory if you need to. No path, no parent folder.
Even when there's no folder structure built around it, a file can exist standalone. In some edge cases, a file could even exist in multiple folders at once!
For full compatibility, we need to address this.
Note These interfaces were renamed and finalized in #19.
Any APIs meant to be used in the context of an identifiable folder structure should be abstracted to a new interface:
/// <summary>
/// Represents a file or folder that resides within a folder structure.
/// </summary>
public interface IAddressableStorable : IStorable
{
/// <summary>
/// A path or uri where the folder resides.
/// </summary>
public string Path { get; }
/// <summary>
/// Gets the containing folder for this item, if any.
/// </summary>
public Task<IAddressableFolder?> GetParentAsync(CancellationToken cancellationToken = default);
}
/// <summary>
/// Represents a file that resides within a folder structure.
/// </summary>
public interface IAddressableFile : IFile, IAddressableStorable
{
}
/// <summary>
/// Represents a folder that resides within a folder structure.
/// </summary>
public interface IAddressableFolder : IFolder, IAddressableStorable
{
}
Note
IAddressableFile
andIAddressableFolder
are included out of the box to avoid requiring a generic check when implementing/consuming.
Note A standard for basic file properties is being discussed in the UWP Community on Discord
"Storage properties" are the additional information about a resource provided by most file systems. It's everything from the name to the "Last modified" date to the Thumbnail, and the majority of them can change / be changed.
In AbstractStorage, we simply copied the c# properties and recreated StorageFileContentProperties
, and even though we only added GetMusicPropertiesAsync
and made the property nullable, it caused a lot of confusion for everyone else who implemented it.
We'll continue using our strategy of "separate interfaces" for this. However, there are a LOT of storage properties we can add, so as long as we separate them into different interfaces, can safely leave "which" for later and instead figure out the "how".
There are 3 requirements
- The information must be accessible and always up-to-date.
- We must provide the option to notify the user of changes
- We must provide the option to change the value (separate interface, see mutability vs modifiability below)
This gives us 2 options:
- Events + properties +
SetThingAsync
methods- Property must be nullable if it can't be assigned in the constructor.
- Events +
GetThingAsync
Methods +SetThingAsync
methods
After some brainstorming, we have a basic skeleton that serves as a guide for any property set.
// The recommended pattern for file properties.
public interface IStorageProperty<T> : IDisposable
{
public T Value { get; }
public event EventHandler<T> ValueUpdated;
}
// Example usage with music properties.
public class MusicData
{
// todo
}
public interface IMusicProperties
{
// Must be behind an async method
public Task<IStorageProperty<MusicData>> GetMusicPropertiesAsync();
}
// If the implementor is capable, they can implement modifying props as well.
public interface IModifiableMusicProperties
{
public Task UpdateMusicProperties(MusicData newValue);
}
Doing it this way means
- There's a standard for getting and tracking groups of properties. Handy for property set designers and enables extensibility scenarios.
- You can easily check if a file/folder supports the interface beforehand, with try/catch or unboxing.
- You won't get name collisions between multiple property sets
- It's behind an async task, meaning it can retrieved asynchronously and easily disposed of, allowing for setup/teardown of sockets, file handles, etc..
- There's no standard for updating properties, it's up to whoever designs the interface. Update one thing at a time, or update the entire object- whatever is most appropriate for the situation (e.g.
UpdateMusicPropertiesAsync()
vsChangeNameAsync()
).
Names are not yet final, please offer suggestions if you have them!
Since folders are responsible for holding files, folders should also be responsible for manipulating their own contents. Also, as previously established, a File can exist without the context of a folder - an important detail for Copy and Move operations.
When designing file system APIs, most implementations don't do this, but given the above, we're 100% sure that we want this method to be implemented on the folder and not the file.
We'll use that for the interface, and have an extension method that swaps them around to something like this:
// Interface method
IFile newFile = await destination.CreateCopyOfAsync(file);
// Extension method
IFile newFile = await file.CopyToAsync(destination);
Note A folder with contents that change doesn't mean you can change the contents. However, if you CAN change the contents, you know there are changes to observe. We've separated mutability and modifiability into 2 interfaces to reflect this.
/// <summary>
/// Represents a folder whose content can change.
/// </summary>
public interface IMutableFolder : IFolder
{
/// <summary>
/// Asynchronously retrieves a disposable object which can notify of changes to the folder.
/// </summary>
/// <returns>A Task. The result is a disposable object which can notify of changes to the folder.</returns>
public Task<IFolderWatcher> GetFolderWatcherAsync(CancellationToken cancellationToken = default);
}
/// <summary>
/// A disposable object which can notify of changes to the folder.
/// </summary>
public interface IFolderWatcher : INotifyCollectionChanged, IDisposable, IAsyncDisposable
{
}
/// <summary>
/// Represents a folder that can be modified.
/// </summary>
public interface IModifiableFolder : IMutableFolder
{
/// <summary>
/// Deletes the provided storable item from this folder.
/// </summary>
/// <param name="item">The item to be removed from this folder.</param>
/// <param name="cancellationToken">The cancellation token to observe.</param>
public Task DeleteAsync(IStorable item, CancellationToken cancellationToken = default);
/// <summary>
/// Creates a copy of the provided file within this folder.
/// </summary>
/// <param name="fileToCopy">The file to be copied into this folder.</param>
/// <param name="overwrite"><code>true</code> if the destination file can be overwritten; otherwise, <c>false</c>.</param>
/// <param name="cancellationToken">The cancellation token to observe.</param>
public Task<IStorable> CreateCopyOfAsync(IFile fileToCopy, bool overwrite = default, CancellationToken cancellationToken = default);
/// <summary>
/// Moves a storable item out of the provided folder, and into this folder. Returns the new item that resides in this folder.
/// </summary>
/// <param name="itemToMove">The storable item being moved into this folder.</param>
/// <param name="source">The folder that <paramref name="itemToMove"/> is being moved from.</param>
/// <param name="overwrite"><code>true</code> if the destination file can be overwritten; otherwise, <c>false</c>.</param>
/// <param name="cancellationToken">The cancellation token to observe.</param>
public Task<IStorable> MoveFromAsync(IStorable itemToMove, IModifiableFolder source, bool overwrite = default, CancellationToken cancellationToken = default);
/// <summary>
/// Creates a new folder with the desired name inside this folder.
/// </summary>
/// <param name="name">The name of the new folder.</param>
/// <param name="overwrite"><code>true</code> if the destination file can be overwritten; otherwise, <c>false</c>.</param>
/// <param name="cancellationToken">The cancellation token to observe.</param>
public Task<IFolder> CreateFolderAsync(string name, bool overwrite = default, CancellationToken cancellationToken = default);
/// <summary>
/// Creates a new file with the desired name inside this folder.
/// </summary>
/// <param name="name">The name of the new file.</param>
/// <param name="overwrite"><code>true</code> if the destination file can be overwritten; otherwise, <c>false</c>.</param>
/// <param name="cancellationToken">The cancellation token to observe.</param>
public Task<IFile> CreateFileAsync(string name, bool overwrite = default, CancellationToken cancellationToken = default);
}
Our original requirements aimed to:
- Maximize compatibility - with as many file systems as reasonably possible
- Minimize effort - for implementors, consumers and maintainers.
The above proposal allows for:
- Minimal effort from implementors.
- Minimal breaking changes when they occur.
- No breaking changes to add new features.
- Maximum compatibility using polyfill extension methods.
- Maximum performance when the underlying file system can be more efficient than our polyfills.
Requirements have definitely been met! Now on to the next part.
Notice that we left out implementation details for properties entirely. This is a large, complex space, and now that we know the requirements and options, we can start to gather decent suggestions for a given property or set of properties.
We might skip this entirely for the first PR and just do the basics, since adding a new interface is not a breaking change.
(Make sure you've read the full proposal first)
Add your feedback, ask questions, make suggestions. If we can improve this, let's do it!