Stateful ProvidersPublished See discussion on Twitter
My team at Wayfair has been hard at work rebuilding our design system documentation site over the past few months.
It has been a ground up rewrite, breaking out of our previous monolithic codebase into a separate design system focused monorepo (living right next to our components and other packages), building on top of a Node.js backend served from a docker container running in Kubernetes (previously we used a mostly SPA architecture built on top of Wayfair's PHP webstack).
On top of this major backend re-architecture, we have also approached
architecting the frontend of the site from a fresh viewpoint. The previous site
was mostly built when render props were all the rage, and only lightly used the
This new site is built from the ground up using hooks (last I checked the only
class component we have in the codebase is our ErrorBoundary component), as well
as a deeper integration with
In addition to using existing React patterns we also decided to build the site on top of the experimental builds of React. We have heavily leaned into using Suspense for data fetching, async event management, as well as code splitting.
One of the early patterns we identified in our site was around our context usage, we knew that defaulting to a single context provider parent for the entire site could lead to unnecessary re-renders when changing single fields of a larger state object.
To mitigate this risk we started a pattern around using single, stateful contexts across the site, where we would previously have a single large stateful context provider at the root of the application, we now have several contexts, each representing a different slice of state.
We render these context providers only as high as they need to be within the application, e.g. if some slice of state is only needed within the header for the site then we only wrap the header in the provider.
We call these contexts stateful because the value that they provide aligns with
the return value of
useReducer, the context provides an array of
value and a setter or dispatcher to update the value.
These context modules expose an interface that looks roughly like this, exporting the Provider and a hook to read the value:
As our codebase continues to evolve, and we begin to polish the site up for its internal launch we may refine this pattern even more. Some of the interesting discussions we have had around this pattern include:
- Do we default to
useStatefor these contexts? Or should we
- Should we expose an explicit
useXContexthook or should we export the raw context object and have consuming sites call
- How does this interplay with using Suspense and external data caches?
As with any code pattern, this architecture might not be right for your application. However it is valuable to be considerate of how you manage state within your application. For us, this pattern helps us to avoid using Redux which we saw as requiring unnecessary ceremony for maintaining and updating state within our application.
If you have other ideas on this pattern, or alternatives I would love to hear them! Reach out on Twitter to share your thoughts, @immatthamlin.