This PR adds a file-based cache implementation to Siesta. This will give most projects offline access & fast launch for almost free.

A word on why this is so compelling for Siesta: traditional HTTP caches (such as URLSession’s cache) exist to prevent requests for data that is still fresh (enough). This makes them good for reducing network traffic, but bad for offline access: it’s a strict choice between getting the cached data or the fresh network data. Worse yet, the API gets to make that choice for you based on (usually ill-behaved) HTTP headers!

Siesta’s architecture, however, makes it possible for a Resource to simultaneously have data and be in the midst of requesting data. This makes it possible for your app to launch with the last data it last had — no matter how old, you decide whether you want to show it — and show that stored data even as it is requesting fresh data, even if that request fails.

This all happens fairly transparently for Siesta apps. Called loadIfNeeded and you’ll get a newData(cache) event when the cached data arrives, then a newData(network) event if the load request succeeds. All you have to do is build your UI so that loading and network errors don’t completely obscure the UI, and bingo! offline access.

There is still documentation work and cleanup to do here, but the code on this branch should be ready to use in your project! I encourage Siesta users to try it and report back. Please give it a whirl! 🙏

To try it out:

service.configure {
    $0.pipeline[.rawData].cacheUsing {
        try FileCache<Data>(
            poolName: “some locally unique name for your API",
            dataIsolation: .perUser(identifiedBy: someUserSpecificValue))
    }
}

…or if you are using public data that can be shared across users of the app:

            dataIsolation: .sharedByAllUsers

This PR also fixes numerous subtle problems with the existing cache infrastructure. It’s long been possible roll your own persistent caching for Siesta, but it always involved a bit of per-project finagling. Improvements here should make custom caches both easier to write and easier to work with — most notably, caches now only apply to GET requests and do not apply to cross-resource requests with load(using:). These two gotchas were a common source of trouble with custom cache implementations.

Fixes #289. Fixes #291.