I just launched a Python package that offers both synchronous and asynchronous APIs. Besides the async/sync difference, the two APIs are mostly the same, leading to a lot of duplicate code. This duplication primarily stems from needing to change both APIs whenever there's an update to the core logic. To streamline this, I created a Python script that generates the async API directly from the sync API by utilizing specific markers in comments. I've found this approach super helpful, but I'm curious if anyone else has seen or used something similar. What do you think about my method?
5 Answers
I'm not sure why you'd need to maintain a sync version if you're already offering async functionality. Async is typically a solution for issues faced in thread-based operations, especially for I/O-bound applications.
Honestly, I would steer clear of this code generation approach. It might seem tidy now, but it can complicate things for others trying to collaborate. Consider using inheritance for sync and async classes instead; this makes the code base more intuitive and easier to maintain in the long run.
While I understand the concerns, isolating the core logic from async calls is tough, especially with all the tests involved. Being able to auto-generate the async API saves time overall.
In my project, I use an AST crawler for generating the sync version of my API. It's pretty efficient and avoids regex pitfalls. I recommend checking out the 'unasync' library as well; it could inspire your code generation approach, especially for maintaining a cleaner structure.
That sounds interesting! I'll definitely look into how AST crawling works for generating APIs.
Creating a script to generate the async API is a clever solution! However, be cautious, as this approach could falter with more complicated interfaces. Async and blocking code behave very differently, especially concerning thread safety. It might be a better strategy to write both versions by hand, allowing the async API to take full advantage of its capabilities. Ultimately, it depends on your project's complexity.
Thanks for the insights! I get that coding both manually offers more flexibility, but the time saving from automation was just too tempting for me.
My advice is to stick with either async or threads depending on your workload. Once you've established the core logic, the sync part can be an afterthought that simply wraps the core functionality without duplicating everything. This method provides performance benefits and helps maintain clarity.
I want to ensure I cater to users who may prefer synchronous calls, as not everyone has an async setup.