Our Journey to Type‑Checking 4 Million Lines of Python at Dropbox

Dropbox, a major Python user, faced growing difficulty understanding its massive, dynamically‑typed codebase, prompting the company to gradually adopt mypy for static type checking to improve productivity and code clarity.

Static typing becomes crucial as projects scale; without type annotations developers struggle to answer basic questions about function return types, parameter purposes, and attribute types, which hampers collaboration and maintenance.

Type checkers like mypy provide verified documentation, catch subtle bugs, simplify refactoring, accelerate feedback loops, and enable IDE features such as autocomplete and error highlighting, thereby boosting developer efficiency.

Dropbox’s migration began in 2015 with a three‑person team, quickly encountering performance bottlenecks when running mypy on CPython. They introduced incremental checking, remote caching, and a mypy daemon to reuse cached dependency information, dramatically reducing re‑check times.

To further improve speed, they developed mypyc, a compiler that translates type‑annotated Python modules into CPython C extensions, achieving roughly a 4× performance gain without full rewrites, and allowing continued use of Python tooling during development.

By 2019, Dropbox had annotated nearly 4 million lines of Python, expanded coverage reports, created stub packages for third‑party libraries, and contributed new type system features such as TypeDict, while continuously refining processes through user surveys, outreach, and static analysis tools.

The experience highlighted challenges like missing files, legacy code annotation, import cycles, and the need for strictness and coverage reporting; solutions included automated tools, strict policies, and improved editor integrations.

In conclusion, Dropbox’s journey demonstrates that systematic static type checking can transform Python into a robust language for large‑scale projects, offering performance, reliability, and developer productivity benefits that the broader community can adopt.

class Resource:
    id: bytes
    ...
    def read_metadata(self, items: Sequence[str]) -> Dict[str, MetadataItem]:
        ...