Error handling, cleaning up, and unwinding

[marcosnils]

TL;DR:

It'd be awesome if the Dagger API offers some DX friendly way of:

• a way to build up an unroll stack
• a way to attach error handlers to a step or named pipeline

From @verdverm in Discord (https://discord.com/channels/707636530424053791/1124479852901310495/1124479852901310495)

The problem is that I don't know when the pipeline failed and if I need to clean up. If I fail in 4, I need to cleanup, if I fail in 1, I do not need to.

There used to be an idea for dagger down that would have the reverse of some steps. Maybe it would be better to have something like WithDefer, which creates an unwind stack? Perhaps this could have conditions for running like pass, fail, always?

Here's how some could prototype something like this given the current API:

func main() {
    ctx := context.Background()
    client, err := dagger.Connect(ctx, dagger.WithLogOutput(os.Stderr))
    if err != nil {
        panic(err)
    }
    defer client.Close()

    rb := client.Pipeline("rollback").Container().
        From("alpine")

    _, err = client.Pipeline("test").
        Container().
        From("alpine").
        With(func(c *dagger.Container) *dagger.Container {
            c, werr := c.WithExec([]string{"apk", "add", "curll"}).Sync(ctx)
            if werr != nil {
                rb.WithExec([]string{"whatever"})
            }
            return c
        }).Sync(ctx)

    if err != nil {
        // pipeline failed, run rb pipeline
        rb.Sync(ctx)
    }
}

cc @shykes