Lectures

We break down the io/fs package as a “language of contracts” for files: the minimal fs.FS interface (only Open), the resource model of fs.File (why Close and defer matter), the narrow contract fs.ReadFileFS, and also practice: reading files without a direct dependency on os.ReadFile, smart implementation selection via type assertion, and the key difference between path vs path/filepath (FS path with / versus OS path).

We break down how os.DirFS turns a directory into an implementation of the fs.FS interface, why this kind of FS uses “FS paths” internally (always with a slash /), how to read files correctly via fs.ReadFile, build names via path.Join (not filepath.Join), and how to correctly recognize “file not found” using errors.Is and fs.ErrNotExist. At the end: DirFS limitations, security nuances, and common mistakes.

Let’s break down how to test file logic in Go without a real disk: write functions that accept fs.FS, substitute fstest.MapFS in tests, verify error causes via errors.Is and fs.ErrNotExist, and also use fstest.TestFS as a quick “inspection” for a virtual file system.

We break down why a “user-provided path” is a potential vulnerability, how OS paths differ from paths inside fs.FS, what exactly fs.ValidPath checks, and why path.Clean is not a defense. We will build a safe template-reading function via os.DirFS + name validation, add a prefix-based access policy, and lock everything down with tests on fstest.MapFS, including blocking traversal attempts via ...

We break down a practical approach to API design in Go: why it’s better to make functions depend on fs.FS (or the narrower fs.ReadFileFS), and push disk access to the application boundary ( main + os.DirFS). We discuss the filename contract (FS path vs OS path: path.Join and filepath.Join), validation via fs.ValidPath, “optional” capabilities via type assertion, diskless testing with fstest.MapFS, and intentional error wrapping via %w for compatibility with errors.Is/ errors.As.