Building a Code Generator (Part 1)

As Inflearn improved its legacy system, the problem of type synchronization between the API and the client came to the fore. Previously, developers manually declared TypeScript types by looking at Swagger documentation, but this had limitations: as a project progressed, the accuracy of the types would degrade, leading to Runtime errors.

In particular, mismatches in documentation between backend and frontend developers, and TypeErrors caused by insufficient handling of null or undefined, were major factors that reduced development productivity. To solve this, the goal was to minimize human intervention and implement automation.

As a solution, adopting GraphQL was considered, but it was not adopted for the following reasons:

• Exposure of DB modeling and dependency on the ORM
• Uncertainty over whether monitoring tools would be supported
• The management burden of introducing a BFF (Backend For Frontend)
• Differences in technical proficiency within the team and the learning curve
• Concerns about conversion costs due to rapid business changes

Beyond this, the team also explored various alternatives such as Client side GraphQL and gRPC, going through a process to find the optimal automation tool.