from Hacker News

Fixrleak: Fixing Java Resource Leaks with GenAI

by carimura on 5/13/25, 9:26 PM with 15 comments

  • by Traubenfuchs on 5/17/25, 1:39 PM

    So you tell me those 200-600k software engineers that can easily solve leetcode hard are so incompetent they missed using try-with-resources at such scale, they needed to introduce new AI tooling to fix it?

    Hey Uber, I am from the EU, I usually can‘t even solve leetcode medium but I will write you scalable, spotless Java for a third of the salary.

    Our industry and its economics are a joke.

  • by stevoski on 5/17/25, 9:32 AM

    > “Resource leaks, where resources like files, database connections, or streams aren’t properly released after use, are a persistent issue in Java applications”

    This was true maybe back in 2005. Java has had try-with-resources for a loooong time. As I see it this has been the dominant idiom for ages, for handling resources that might leak.

  • by xyst on 5/17/25, 10:55 AM

    Using AI when a static scanner like SonarQube easily picks up these types of resource leaks, especially in Java.

    Peak waste.

    What’s next?

    "Get rid of your GitHub dependabot alerts and replace it with my shitty ChatGPT wrapper”

  • by TYMorningCoffee on 5/17/25, 2:02 PM

    A lot of commenters point out that there already are many established static checkers that do this. That is not what Uber attempts here.

    Uber is not proposing a static checker. They even use sonar qube in their architecture. They propose using an LLM to resolve the leak detected by sonar qube.

  • by sigotirandolas on 5/17/25, 10:33 AM

    > This analysis ensures that FixrLeak skips functions where resources are passed as parameters, returned, or stored in fields, as these resources often outlive the function’s scope.

    > FixrLeak delivers precise, reliable fixes while leaving more complex cases for advanced analysis at the caller level.

    In other words, this will only fix trivial leaks, which are best seen as a language design issue and can be fixed by RAII, reference counting, etc.

    It won't fix the more insidious leaks like `UNBOUNDED_QUEUE.add(item)` that are more likely to pass through code review in the first place.

  • by bob778 on 5/17/25, 10:20 AM

    How much effort was spent automating this to fix 112 instances across Uber’s code base? I assume code reviews would catch any new issues so this seems like overkill for a small one-off task?

  • by rvz on 5/17/25, 11:22 AM

    Why exactly do you need LLMs for this when efficient alternatives like SonarQube or checkstyle already do this without the expensive waste LLMs create?

    This adds little to no technical advantage over existing solutions what so ever for this particular use case.

  • by hawk_ on 5/17/25, 10:27 AM

    Spotbugs or checkstyle etc... would catch these. What does AI add here?

  • by yahoozoo on 5/17/25, 11:42 AM

    stupid af