Shiny Module Pipelines

As a big fan of the functional approach to programming, many times I have struggled to find an elegant solution to complex Shiny apps where data manipulation from multiple sources is necessary. Today, while trying some stuff, I found a little way of organizing Shiny Modules in such a way that a pipeline of complex manipulations can be built.

This is my first attemp at this. I am not sure what the limitations or possible best practices for such a way of writing Shiny Apps could be, but here is what I have come up with. I would love to hear back from the community with opinions or possibly contributions.

The approach

When dealing with modules, purpose is really important. For this to work, a module must return only a reactive that is update only by inputs and reactives inside the module's server logic. It is best for the module to do one thing and thing only so that its output can be piped into the next module.

# This is in the server.R
global_data <- initial_data() %>>%
  apply_filter(id = "phase1") %>>%
  apply_filter(id = "phase2") %>>%
  apply_filter(id = "phase3")

In the example above, we have a reactive called initial_data. This could be a reactive that listens to user input and fetches some data from a database or what ever, lets pretend this is a dataframe of lazy loaded tibble. We pass this reactive to module server function named apply_filter. Let's asume this module returns a subset of the data passed on to it. We repeat this 2 more times and assign this new reactive to global_data.

This simplifies the process of interacting between modules. Furthermore, you can control the flow of data really easily by adding the bindEvent function between the pipes.

# This is in the server.R
global_data <- initial_data() %>>%
  apply_filter(id = "phase1") %>>%
  apply_filter(id = "phase2") %>>%
  bindEvent(input$apply_filters) %>>%
  apply_filter(id = "phase3") %>>%
  bindEvent(input$apply_final_filter)

By adding bindEvent in between the piping, we can control what the user does. If we want the user to first apply the phase1 and phase2 filters and then the phase3 filter, we can easily control that in a very concise manner.

I included the code to a very basic app showing off this approach of writing Shiny code. I would love to hear back from the community, thank you. This currently only works with the pipe provided by the pipeR package due to how things are evualuated with the magrittr and base pipes.