洪 民憙 (Hong Minhee)'s avatar

洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · 898 following · 1114 followers

An intersectionalist, feminist, and socialist guy living in Seoul (UTC+09:00). @tokolovesme's spouse. Who's behind @fedify, @hollo, and @botkit. Write some free software in , , , & . They/them.

서울에 사는 交叉女性主義者이자 社會主義者. 金剛兔(@tokolovesme)의 配偶者. @fedify, @hollo, @botkit 메인테이너. , , , 等으로 自由 소프트웨어 만듦.

()

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

Hello, I'm an open source software engineer in my late 30s living in , , and an avid advocate of and the .

I'm the creator of @fedify, an server framework in , @hollo, an ActivityPub-enabled microblogging software for single users, and @botkit, a simple ActivityPub bot framework.

I'm also very interested in East Asian languages (so-called ) and . Feel free to talk to me in , (), or (), or even in Literary Chinese (, )!

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to 洪 民憙 (Hong Minhee)'s post

安寧(안녕)하세요, 저는 서울에 살고 있는 30() 後半(후반) 오픈 소스 소프트웨어 엔지니어이며, 自由(자유)·오픈 소스 소프트웨어와 聯合宇宙(연합우주)(fediverse)의 熱烈(열렬)支持者(지지자)입니다.

저는 TypeScript() ActivityPub 서버 프레임워크인 @fedify 프로젝트와 싱글 유저() ActivityPub 마이크로블로그인 @hollo 프로젝트와 ActivityPub 봇 프레임워크인 @botkit 프로젝트의 製作者(제작자)이기도 합니다.

저는 ()아시아 言語(언어)(이른바 )와 유니코드에도 關心(관심)이 많습니다. 聯合宇宙(연합우주)에서는 國漢文混用體(국한문 혼용체)를 쓰고 있어요! 제게 韓國語(한국어)英語(영어), 日本語(일본어)로 말을 걸어주세요. (아니면, 漢文(한문)으로도!)

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to 洪 民憙 (Hong Minhee)'s post

こんにちは、私はソウルに住んでいる30代後半のオープンソースソフトウェアエンジニアで、自由・オープンソースソフトウェアとフェディバースの熱烈な支持者です。名前は洪 民憙ホン・ミンヒです。

私はTypeScript用のActivityPubサーバーフレームワークである「@fedify」と、ActivityPubをサポートする1人用マイクロブログである 「@hollo」と、ActivityPubのボットを作成する為のシンプルなフレームワークである「@botkit」の作者でもあります。

私は東アジア言語(いわゆるCJK)とUnicodeにも興味が多いです。日本語、英語、韓国語で話しかけてください。(または、漢文でも!)

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to charlag's post

@charlag I think you've misunderstood my point. I wasn't arguing that checked exceptions are better than Result<T, E> types or that explicit propagation is bad. I was suggesting that within Java's ecosystem, checked exceptions could be valuable if properly improved, rather than abandoned entirely. Both approaches have their merits in their respective languages and paradigms. Monadic types definitely compose better, but automatic propagation has advantages in certain contexts too.

박준규's avatar
박준규

@curry@hackers.pub · Reply to 洪 民憙 (Hong Minhee)'s post

@hongminhee 이런 표현이 있습니다.

GHC has more flags than the UN.

SorairoLake's avatar
SorairoLake

@sorairolake@misskey.io

仮名もハングルも表音文字だから日本語の文章をハングルで表記できそうと思ったらそれっぽいのがもうあった
https://ja.wikipedia.org/wiki/%E6%97%A5%E6%9C%AC%E8%AA%9E%E3%81%AE%E3%83%8F%E3%83%B3%E3%82%B0%E3%83%AB%E8%A1%A8%E8%A8%98

bgl gwyng's avatar
bgl gwyng

@bgl@hackers.pub · Reply to 티르's post

@tirr 저도 서브타이핑 기반인 TS에 상대적으로 쉽게 도입할 기능이 https://github.com/microsoft/TypeScript/issues/13219 이렇게 오랫동안 진행안되는게 불만입니다. 막상 TS 이펙트 라이브러리들은 |로 흉내내서 잘 쓰고 있더라고요. Haskell처럼 대수적 이펙트는 구현할수있지만 서브타이핑 기반은 아닌 언어에선, 서브타이핑 흉내낸다고 타입레벨 차력쇼하고 있는데 맞는 방향인지 모르겠습니다.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to @reiver ⊼ (Charles) :batman:'s post

@reiver Thanks for sharing this! I'm actually interested in the event, but I'd be more likely to attend if they had a track on the fediverse or open source social media. The current tracks do seem quite Linux-focused. If you hear anything about them adding such a track, please let me know!

티르's avatar
티르

@tirr@mitir.social · Reply to 티르's post

타입스크립트에 있었으면 좋겠음

티르's avatar
티르

@tirr@mitir.social

그 뭐라고 하더라... 대수적 이펙트?

RE:
https://hollo.social/@hongminhee/0195b690-cf8e-76b0-84bc-470c9b1c3d5b

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

Hot take: Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

geeknews_bot's avatar
geeknews_bot

@geeknews_bot@sns.lemondouble.com

자바의 체크드 예외 재고찰: 저평가된 타입 안전성 기능
------------------------------
## 주요 내용 요약

* 자바의 체크드 예외가 커뮤니티에서 널리 비판받는 기능임에도 타입 안전성 측면에서 뛰어난 장점 보유.
* Rust의
Result<T, E>나 Haskell의 Either a b와 개념적으로 유사한 타입 안전성 메커니즘 제공.
* 체크드 예외가 메서드 시그니처에 잠재적 실패 가능성을 명시적으로 표현하…
------------------------------
https://news.hada.io/topic?id=19877&utm_source=googlechat&utm_medium=bot&utm_campaign=1834

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to 洪 民憙 (Hong Minhee)'s post

Wrote a blog post about this topic.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hackers.pub

Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

Introduction

Few features in Java have been as consistently criticized as checked exceptions. Modern Java libraries and frameworks often go to great lengths to avoid them. Newer JVM languages like Kotlin have abandoned them entirely. Many experienced Java developers consider them a design mistake.

But what if this conventional wisdom is wrong? What if checked exceptions represent one of Java's most forward-thinking features?

In this post, I'll argue that Java's checked exceptions were ahead of their time, offering many of the same type safety benefits that are now celebrated in languages like Rust and Haskell. Rather than abandoning this feature, we should consider how to improve it to work better with modern Java's features.

Understanding Java's Exception Handling Model

To set the stage, let's review how Java's exception system works:

  • Unchecked exceptions (subclasses of RuntimeException or Error): These don't need to be declared or caught. They typically represent programming errors (NullPointerException, IndexOutOfBoundsException) or unrecoverable conditions (OutOfMemoryError).

  • Checked exceptions (subclasses of Exception but not RuntimeException): These must either be caught with try/catch blocks or declared in the method signature with throws. They represent recoverable conditions that are outside the normal flow of execution (IOException, SQLException).

Here's how this works in practice:

// Checked exception - compiler forces you to handle or declare it
public void readFile(String path) throws IOException {
    Files.readAllLines(Path.of(path));
}

// Unchecked exception - no compiler enforcement
public void processArray(int[] array) {
    int value = array[array.length + 1]; // May throw ArrayIndexOutOfBoundsException
}

The Type Safety Argument for Checked Exceptions

At their core, checked exceptions are a way of encoding potential failure modes into the type system via method signatures. This makes certain failure cases part of the API contract, forcing client code to explicitly handle these cases.

Consider this method signature:

public byte[] readFileContents(String filePath) throws IOException

The throws IOException clause tells us something critical: this method might fail in ways related to IO operations. The compiler ensures you can't simply ignore this fact. You must either:

  1. Handle the exception with a try-catch block
  2. Propagate it by declaring it in your own method signature

This type-level representation of potential failures aligns perfectly with principles of modern type-safe programming.

Automatic Propagation: A Hidden Advantage

One often overlooked advantage of Java's checked exceptions is their automatic propagation. Once you declare a method as throws IOException, any exception that occurs is automatically propagated to the caller without additional syntax.

Compare this with Rust, where you must use the ? operator every time you call a function that returns a Result:

// Rust requires explicit propagation with ? for each call
fn read_and_process(path: &str) -> Result<(), std::io::Error> {
    let content = std::fs::read_to_string(path)?;
    process_content(&content)?;
    Ok(())
}

// Java automatically propagates exceptions once declared
void readAndProcess(String path) throws IOException {
    String content = Files.readString(Path.of(path));
    processContent(content); // If this throws IOException, it's automatically propagated
}

In complex methods with many potential failure points, Java's approach leads to cleaner code by eliminating the need for repetitive error propagation markers.

Modern Parallels: Result Types in Rust and Haskell

The approach of encoding failure possibilities in the type system has been adopted by many modern languages, most notably Rust with its Result<T, E> type and Haskell with its Either a b type.

In Rust:

fn read_file_contents(file_path: &str) -> Result<Vec<u8>, std::io::Error> {
    std::fs::read(file_path)
}

When calling this function, you can't just ignore the potential for errors—you need to handle both the success case and the error case, often using the ? operator or pattern matching.

In Haskell:

readFileContents :: FilePath -> IO (Either IOException ByteString)
readFileContents path = try $ BS.readFile path

Again, the caller must explicitly deal with both possible outcomes.

This is fundamentally the same insight that motivated Java's checked exceptions: make failure handling explicit in the type system.

Valid Criticisms of Checked Exceptions

If checked exceptions are conceptually similar to these widely-praised error handling mechanisms, why have they fallen out of favor? There are several legitimate criticisms:

1. Excessive Boilerplate in the Call Chain

The most common complaint is the boilerplate required when propagating exceptions up the call stack:

void methodA() throws IOException {
    methodB();
}

void methodB() throws IOException {
    methodC();
}

void methodC() throws IOException {
    // Actual code that might throw IOException
}

Every method in the chain must declare the same exception, creating repetitive code. While automatic propagation works well within a method, the explicit declaration in method signatures creates overhead.

2. Poor Integration with Functional Programming

Java 8 introduced lambdas and streams, but checked exceptions don't play well with them:

// Won't compile because map doesn't expect functions that throw checked exceptions
List<String> fileContents = filePaths.stream()
    .map(path -> Files.readString(Path.of(path))) // Throws IOException
    .collect(Collectors.toList());

This forces developers to use awkward workarounds:

List<String> fileContents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new UncheckedIOException(e); // Wrap in an unchecked exception
        }
    })
    .collect(Collectors.toList());

3. Interface Evolution Problems

Adding a checked exception to an existing method breaks all implementing classes and calling code. This makes evolving interfaces over time difficult, especially for widely-used libraries and frameworks.

4. Catch-and-Ignore Anti-Pattern

The strictness of checked exceptions can lead to the worst possible outcome—developers simply catching and ignoring exceptions to make the compiler happy:

try {
    // Code that might throw
} catch (Exception e) {
    // Do nothing or just log
}

This is worse than having no exception checking at all because it provides a false sense of security.

Improving Checked Exceptions Without Abandoning Them

Rather than abandoning checked exceptions entirely, Java could enhance the existing system to address these legitimate concerns. Here are some potential improvements that preserve the type safety benefits while addressing the practical problems:

1. Allow lambdas to declare checked exceptions

One of the biggest pain points with checked exceptions today is their incompatibility with functional interfaces. Consider how much cleaner this would be:

// Current approach - forced to handle or wrap exceptions inline
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    })
    .collect(Collectors.toList());

// Potential future approach - lambdas can declare exceptions
List<String> contents = filePaths.stream()
    .map((String path) throws IOException -> Files.readString(Path.of(path)))
    .collect(Collectors.toList());

This would require updating functional interfaces to support exception declarations:

@FunctionalInterface
public interface Function<T, R, E extends Exception> {
    R apply(T t) throws E;
}

2. Generic exception types in throws clauses

Another powerful enhancement would be allowing generic type parameters in throws clauses:

public <E extends Exception> void processWithException(Supplier<Void, E> supplier) throws E {
    supplier.get();
}

This would enable much more flexible composition of methods that work with different exception types, bringing some of the flexibility of Rust's Result<T, E> to Java's existing exception system.

3. Better support for exception handling in functional contexts

Unlike Rust which requires the ? operator for error propagation, Java already automatically propagates checked exceptions when declared in the method signature. What Java needs instead is better support for checked exceptions in functional contexts:

// Current approach for handling exceptions in streams
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e); // Lose type information
        }
    })
    .collect(Collectors.toList());

// Hypothetical improved API
List<String> contents = filePaths.stream()
    .mapThrowing(path -> Files.readString(Path.of(path))) // Preserves checked exception
    .onException(IOException.class, e -> logError(e))
    .collect(Collectors.toList());

4. Integration with Optional<T> and Stream<T> APIs

The standard library could be enhanced to better support operations that might throw checked exceptions:

// Hypothetical API
Optional<String> content = Optional.ofThrowable(() -> Files.readString(Path.of("file.txt")));
content.ifPresentOrElse(
    this::processContent,
    exception -> log.error("Failed to read file", exception)
);

Comparison with Other Languages' Approaches

It's worth examining how other languages have addressed the error handling problem:

Rust's Result<T, E> and ? operator

Rust's approach using Result<T, E> and the ? operator shows how propagation can be made concise while keeping the type safety benefits. The ? operator automatically unwraps a successful result or returns the error to the caller, making propagation more elegant.

However, Rust's approach requires explicit propagation at each step, which can be more verbose than Java's automatic propagation in certain scenarios.

Kotlin's Approach

Kotlin made all exceptions unchecked but provides functional constructs like runCatching that bring back some type safety in a more modern way:

val result = runCatching {
    Files.readString(Path.of("file.txt"))
}

result.fold(
    onSuccess = { content -> processContent(content) },
    onFailure = { exception -> log.error("Failed to read file", exception) }
)

This approach works well with Kotlin's functional programming paradigm but lacks compile-time enforcement.

Scala's Try[T], Either[A, B], and Effect Systems

Scala offers Try[T], Either[A, B], and various effect systems that encode errors in the type system while integrating well with functional programming:

import scala.util.Try

val fileContent: Try[String] = Try {
  Source.fromFile("file.txt").mkString
}

fileContent match {
  case Success(content) => processContent(content)
  case Failure(exception) => log.error("Failed to read file", exception)
}

This approach preserves type safety while fitting well with Scala's functional paradigm.

Conclusion

Java's checked exceptions were a pioneering attempt to bring type safety to error handling. While the implementation has shortcomings, the core concept aligns with modern type-safe approaches to error handling in languages like Rust and Haskell.

Copying Rust's Result<T, E> might seem like the obvious solution, but it would represent a radical departure from Java's established paradigms. Instead, targeted enhancements to the existing checked exceptions system—like allowing lambdas to declare exceptions and supporting generic exception types—could preserve Java's unique approach while addressing its practical limitations.

The beauty of such improvements is that they'd maintain backward compatibility while making checked exceptions work seamlessly with modern Java features like lambdas and streams. They would acknowledge that the core concept of checked exceptions was sound—the problem was in the implementation details and their interaction with newer language features.

So rather than abandoning checked exceptions entirely, perhaps we should recognize them as a forward-thinking feature that was implemented before its time. As Java continues to evolve, we have an opportunity to refine this system rather than replace it.

In the meantime, next time you're tempted to disparage checked exceptions, remember: they're not just an annoying Java quirk—they're an early attempt at the same type safety paradigm that newer languages now implement with much celebration.

What do you think? Could these improvements make checked exceptions viable for modern Java development? Or is it too late to salvage this controversial feature? I'm interested in hearing your thoughts in the comments.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to bgl gwyng's post

@bgl Absolutely! You make a great point about subtyping—it's one of Java's underappreciated strengths for error handling. The exception hierarchy lets you catch specific exceptions or their broader subtypes as needed, giving you flexible error composition without extra syntax.

bgl gwyng's avatar
bgl gwyng

@bgl@hackers.pub · Reply to 洪 民憙 (Hong Minhee)'s post

@hongminhee I can't agree more. 'Java sucks' doesn't apply here at least. Subtyping plays really well in unioning error types.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to An Nyeong (安寧)'s post

@nyeong Yes, functionally they're similar—both enforce error handling at compile time! The key difference is that Java automatically propagates exceptions once declared, while Rust requires explicit ? for each call. Swift adding typed throws is fascinating validation of the concept.

My blog post explores these differences in depth.

The semantics matter though—Java's approach integrates better with imperative code, while Result<T, E> types fit functional paradigms. Each has its place!

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hackers.pub

Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

Introduction

Few features in Java have been as consistently criticized as checked exceptions. Modern Java libraries and frameworks often go to great lengths to avoid them. Newer JVM languages like Kotlin have abandoned them entirely. Many experienced Java developers consider them a design mistake.

But what if this conventional wisdom is wrong? What if checked exceptions represent one of Java's most forward-thinking features?

In this post, I'll argue that Java's checked exceptions were ahead of their time, offering many of the same type safety benefits that are now celebrated in languages like Rust and Haskell. Rather than abandoning this feature, we should consider how to improve it to work better with modern Java's features.

Understanding Java's Exception Handling Model

To set the stage, let's review how Java's exception system works:

  • Unchecked exceptions (subclasses of RuntimeException or Error): These don't need to be declared or caught. They typically represent programming errors (NullPointerException, IndexOutOfBoundsException) or unrecoverable conditions (OutOfMemoryError).

  • Checked exceptions (subclasses of Exception but not RuntimeException): These must either be caught with try/catch blocks or declared in the method signature with throws. They represent recoverable conditions that are outside the normal flow of execution (IOException, SQLException).

Here's how this works in practice:

// Checked exception - compiler forces you to handle or declare it
public void readFile(String path) throws IOException {
    Files.readAllLines(Path.of(path));
}

// Unchecked exception - no compiler enforcement
public void processArray(int[] array) {
    int value = array[array.length + 1]; // May throw ArrayIndexOutOfBoundsException
}

The Type Safety Argument for Checked Exceptions

At their core, checked exceptions are a way of encoding potential failure modes into the type system via method signatures. This makes certain failure cases part of the API contract, forcing client code to explicitly handle these cases.

Consider this method signature:

public byte[] readFileContents(String filePath) throws IOException

The throws IOException clause tells us something critical: this method might fail in ways related to IO operations. The compiler ensures you can't simply ignore this fact. You must either:

  1. Handle the exception with a try-catch block
  2. Propagate it by declaring it in your own method signature

This type-level representation of potential failures aligns perfectly with principles of modern type-safe programming.

Automatic Propagation: A Hidden Advantage

One often overlooked advantage of Java's checked exceptions is their automatic propagation. Once you declare a method as throws IOException, any exception that occurs is automatically propagated to the caller without additional syntax.

Compare this with Rust, where you must use the ? operator every time you call a function that returns a Result:

// Rust requires explicit propagation with ? for each call
fn read_and_process(path: &str) -> Result<(), std::io::Error> {
    let content = std::fs::read_to_string(path)?;
    process_content(&content)?;
    Ok(())
}

// Java automatically propagates exceptions once declared
void readAndProcess(String path) throws IOException {
    String content = Files.readString(Path.of(path));
    processContent(content); // If this throws IOException, it's automatically propagated
}

In complex methods with many potential failure points, Java's approach leads to cleaner code by eliminating the need for repetitive error propagation markers.

Modern Parallels: Result Types in Rust and Haskell

The approach of encoding failure possibilities in the type system has been adopted by many modern languages, most notably Rust with its Result<T, E> type and Haskell with its Either a b type.

In Rust:

fn read_file_contents(file_path: &str) -> Result<Vec<u8>, std::io::Error> {
    std::fs::read(file_path)
}

When calling this function, you can't just ignore the potential for errors—you need to handle both the success case and the error case, often using the ? operator or pattern matching.

In Haskell:

readFileContents :: FilePath -> IO (Either IOException ByteString)
readFileContents path = try $ BS.readFile path

Again, the caller must explicitly deal with both possible outcomes.

This is fundamentally the same insight that motivated Java's checked exceptions: make failure handling explicit in the type system.

Valid Criticisms of Checked Exceptions

If checked exceptions are conceptually similar to these widely-praised error handling mechanisms, why have they fallen out of favor? There are several legitimate criticisms:

1. Excessive Boilerplate in the Call Chain

The most common complaint is the boilerplate required when propagating exceptions up the call stack:

void methodA() throws IOException {
    methodB();
}

void methodB() throws IOException {
    methodC();
}

void methodC() throws IOException {
    // Actual code that might throw IOException
}

Every method in the chain must declare the same exception, creating repetitive code. While automatic propagation works well within a method, the explicit declaration in method signatures creates overhead.

2. Poor Integration with Functional Programming

Java 8 introduced lambdas and streams, but checked exceptions don't play well with them:

// Won't compile because map doesn't expect functions that throw checked exceptions
List<String> fileContents = filePaths.stream()
    .map(path -> Files.readString(Path.of(path))) // Throws IOException
    .collect(Collectors.toList());

This forces developers to use awkward workarounds:

List<String> fileContents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new UncheckedIOException(e); // Wrap in an unchecked exception
        }
    })
    .collect(Collectors.toList());

3. Interface Evolution Problems

Adding a checked exception to an existing method breaks all implementing classes and calling code. This makes evolving interfaces over time difficult, especially for widely-used libraries and frameworks.

4. Catch-and-Ignore Anti-Pattern

The strictness of checked exceptions can lead to the worst possible outcome—developers simply catching and ignoring exceptions to make the compiler happy:

try {
    // Code that might throw
} catch (Exception e) {
    // Do nothing or just log
}

This is worse than having no exception checking at all because it provides a false sense of security.

Improving Checked Exceptions Without Abandoning Them

Rather than abandoning checked exceptions entirely, Java could enhance the existing system to address these legitimate concerns. Here are some potential improvements that preserve the type safety benefits while addressing the practical problems:

1. Allow lambdas to declare checked exceptions

One of the biggest pain points with checked exceptions today is their incompatibility with functional interfaces. Consider how much cleaner this would be:

// Current approach - forced to handle or wrap exceptions inline
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    })
    .collect(Collectors.toList());

// Potential future approach - lambdas can declare exceptions
List<String> contents = filePaths.stream()
    .map((String path) throws IOException -> Files.readString(Path.of(path)))
    .collect(Collectors.toList());

This would require updating functional interfaces to support exception declarations:

@FunctionalInterface
public interface Function<T, R, E extends Exception> {
    R apply(T t) throws E;
}

2. Generic exception types in throws clauses

Another powerful enhancement would be allowing generic type parameters in throws clauses:

public <E extends Exception> void processWithException(Supplier<Void, E> supplier) throws E {
    supplier.get();
}

This would enable much more flexible composition of methods that work with different exception types, bringing some of the flexibility of Rust's Result<T, E> to Java's existing exception system.

3. Better support for exception handling in functional contexts

Unlike Rust which requires the ? operator for error propagation, Java already automatically propagates checked exceptions when declared in the method signature. What Java needs instead is better support for checked exceptions in functional contexts:

// Current approach for handling exceptions in streams
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e); // Lose type information
        }
    })
    .collect(Collectors.toList());

// Hypothetical improved API
List<String> contents = filePaths.stream()
    .mapThrowing(path -> Files.readString(Path.of(path))) // Preserves checked exception
    .onException(IOException.class, e -> logError(e))
    .collect(Collectors.toList());

4. Integration with Optional<T> and Stream<T> APIs

The standard library could be enhanced to better support operations that might throw checked exceptions:

// Hypothetical API
Optional<String> content = Optional.ofThrowable(() -> Files.readString(Path.of("file.txt")));
content.ifPresentOrElse(
    this::processContent,
    exception -> log.error("Failed to read file", exception)
);

Comparison with Other Languages' Approaches

It's worth examining how other languages have addressed the error handling problem:

Rust's Result<T, E> and ? operator

Rust's approach using Result<T, E> and the ? operator shows how propagation can be made concise while keeping the type safety benefits. The ? operator automatically unwraps a successful result or returns the error to the caller, making propagation more elegant.

However, Rust's approach requires explicit propagation at each step, which can be more verbose than Java's automatic propagation in certain scenarios.

Kotlin's Approach

Kotlin made all exceptions unchecked but provides functional constructs like runCatching that bring back some type safety in a more modern way:

val result = runCatching {
    Files.readString(Path.of("file.txt"))
}

result.fold(
    onSuccess = { content -> processContent(content) },
    onFailure = { exception -> log.error("Failed to read file", exception) }
)

This approach works well with Kotlin's functional programming paradigm but lacks compile-time enforcement.

Scala's Try[T], Either[A, B], and Effect Systems

Scala offers Try[T], Either[A, B], and various effect systems that encode errors in the type system while integrating well with functional programming:

import scala.util.Try

val fileContent: Try[String] = Try {
  Source.fromFile("file.txt").mkString
}

fileContent match {
  case Success(content) => processContent(content)
  case Failure(exception) => log.error("Failed to read file", exception)
}

This approach preserves type safety while fitting well with Scala's functional paradigm.

Conclusion

Java's checked exceptions were a pioneering attempt to bring type safety to error handling. While the implementation has shortcomings, the core concept aligns with modern type-safe approaches to error handling in languages like Rust and Haskell.

Copying Rust's Result<T, E> might seem like the obvious solution, but it would represent a radical departure from Java's established paradigms. Instead, targeted enhancements to the existing checked exceptions system—like allowing lambdas to declare exceptions and supporting generic exception types—could preserve Java's unique approach while addressing its practical limitations.

The beauty of such improvements is that they'd maintain backward compatibility while making checked exceptions work seamlessly with modern Java features like lambdas and streams. They would acknowledge that the core concept of checked exceptions was sound—the problem was in the implementation details and their interaction with newer language features.

So rather than abandoning checked exceptions entirely, perhaps we should recognize them as a forward-thinking feature that was implemented before its time. As Java continues to evolve, we have an opportunity to refine this system rather than replace it.

In the meantime, next time you're tempted to disparage checked exceptions, remember: they're not just an annoying Java quirk—they're an early attempt at the same type safety paradigm that newer languages now implement with much celebration.

What do you think? Could these improvements make checked exceptions viable for modern Java development? Or is it too late to salvage this controversial feature? I'm interested in hearing your thoughts in the comments.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to Ian Wagner's post

@ianthetechie Thanks for your thoughts! You're right about the catch-and-ignore pattern—it's a common anti-pattern in Java that undermines the benefits of checked exceptions. My blog post actually addresses this and compares Java's approach with Rust's Result<T, E> in detail.

I agree about Kotlin's choice—they sacrificed type safety for convenience. The ideal solution would be improving Java's checked exceptions to work better with modern features like lambdas, rather than abandoning them entirely.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hackers.pub

Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

Introduction

Few features in Java have been as consistently criticized as checked exceptions. Modern Java libraries and frameworks often go to great lengths to avoid them. Newer JVM languages like Kotlin have abandoned them entirely. Many experienced Java developers consider them a design mistake.

But what if this conventional wisdom is wrong? What if checked exceptions represent one of Java's most forward-thinking features?

In this post, I'll argue that Java's checked exceptions were ahead of their time, offering many of the same type safety benefits that are now celebrated in languages like Rust and Haskell. Rather than abandoning this feature, we should consider how to improve it to work better with modern Java's features.

Understanding Java's Exception Handling Model

To set the stage, let's review how Java's exception system works:

  • Unchecked exceptions (subclasses of RuntimeException or Error): These don't need to be declared or caught. They typically represent programming errors (NullPointerException, IndexOutOfBoundsException) or unrecoverable conditions (OutOfMemoryError).

  • Checked exceptions (subclasses of Exception but not RuntimeException): These must either be caught with try/catch blocks or declared in the method signature with throws. They represent recoverable conditions that are outside the normal flow of execution (IOException, SQLException).

Here's how this works in practice:

// Checked exception - compiler forces you to handle or declare it
public void readFile(String path) throws IOException {
    Files.readAllLines(Path.of(path));
}

// Unchecked exception - no compiler enforcement
public void processArray(int[] array) {
    int value = array[array.length + 1]; // May throw ArrayIndexOutOfBoundsException
}

The Type Safety Argument for Checked Exceptions

At their core, checked exceptions are a way of encoding potential failure modes into the type system via method signatures. This makes certain failure cases part of the API contract, forcing client code to explicitly handle these cases.

Consider this method signature:

public byte[] readFileContents(String filePath) throws IOException

The throws IOException clause tells us something critical: this method might fail in ways related to IO operations. The compiler ensures you can't simply ignore this fact. You must either:

  1. Handle the exception with a try-catch block
  2. Propagate it by declaring it in your own method signature

This type-level representation of potential failures aligns perfectly with principles of modern type-safe programming.

Automatic Propagation: A Hidden Advantage

One often overlooked advantage of Java's checked exceptions is their automatic propagation. Once you declare a method as throws IOException, any exception that occurs is automatically propagated to the caller without additional syntax.

Compare this with Rust, where you must use the ? operator every time you call a function that returns a Result:

// Rust requires explicit propagation with ? for each call
fn read_and_process(path: &str) -> Result<(), std::io::Error> {
    let content = std::fs::read_to_string(path)?;
    process_content(&content)?;
    Ok(())
}

// Java automatically propagates exceptions once declared
void readAndProcess(String path) throws IOException {
    String content = Files.readString(Path.of(path));
    processContent(content); // If this throws IOException, it's automatically propagated
}

In complex methods with many potential failure points, Java's approach leads to cleaner code by eliminating the need for repetitive error propagation markers.

Modern Parallels: Result Types in Rust and Haskell

The approach of encoding failure possibilities in the type system has been adopted by many modern languages, most notably Rust with its Result<T, E> type and Haskell with its Either a b type.

In Rust:

fn read_file_contents(file_path: &str) -> Result<Vec<u8>, std::io::Error> {
    std::fs::read(file_path)
}

When calling this function, you can't just ignore the potential for errors—you need to handle both the success case and the error case, often using the ? operator or pattern matching.

In Haskell:

readFileContents :: FilePath -> IO (Either IOException ByteString)
readFileContents path = try $ BS.readFile path

Again, the caller must explicitly deal with both possible outcomes.

This is fundamentally the same insight that motivated Java's checked exceptions: make failure handling explicit in the type system.

Valid Criticisms of Checked Exceptions

If checked exceptions are conceptually similar to these widely-praised error handling mechanisms, why have they fallen out of favor? There are several legitimate criticisms:

1. Excessive Boilerplate in the Call Chain

The most common complaint is the boilerplate required when propagating exceptions up the call stack:

void methodA() throws IOException {
    methodB();
}

void methodB() throws IOException {
    methodC();
}

void methodC() throws IOException {
    // Actual code that might throw IOException
}

Every method in the chain must declare the same exception, creating repetitive code. While automatic propagation works well within a method, the explicit declaration in method signatures creates overhead.

2. Poor Integration with Functional Programming

Java 8 introduced lambdas and streams, but checked exceptions don't play well with them:

// Won't compile because map doesn't expect functions that throw checked exceptions
List<String> fileContents = filePaths.stream()
    .map(path -> Files.readString(Path.of(path))) // Throws IOException
    .collect(Collectors.toList());

This forces developers to use awkward workarounds:

List<String> fileContents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new UncheckedIOException(e); // Wrap in an unchecked exception
        }
    })
    .collect(Collectors.toList());

3. Interface Evolution Problems

Adding a checked exception to an existing method breaks all implementing classes and calling code. This makes evolving interfaces over time difficult, especially for widely-used libraries and frameworks.

4. Catch-and-Ignore Anti-Pattern

The strictness of checked exceptions can lead to the worst possible outcome—developers simply catching and ignoring exceptions to make the compiler happy:

try {
    // Code that might throw
} catch (Exception e) {
    // Do nothing or just log
}

This is worse than having no exception checking at all because it provides a false sense of security.

Improving Checked Exceptions Without Abandoning Them

Rather than abandoning checked exceptions entirely, Java could enhance the existing system to address these legitimate concerns. Here are some potential improvements that preserve the type safety benefits while addressing the practical problems:

1. Allow lambdas to declare checked exceptions

One of the biggest pain points with checked exceptions today is their incompatibility with functional interfaces. Consider how much cleaner this would be:

// Current approach - forced to handle or wrap exceptions inline
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    })
    .collect(Collectors.toList());

// Potential future approach - lambdas can declare exceptions
List<String> contents = filePaths.stream()
    .map((String path) throws IOException -> Files.readString(Path.of(path)))
    .collect(Collectors.toList());

This would require updating functional interfaces to support exception declarations:

@FunctionalInterface
public interface Function<T, R, E extends Exception> {
    R apply(T t) throws E;
}

2. Generic exception types in throws clauses

Another powerful enhancement would be allowing generic type parameters in throws clauses:

public <E extends Exception> void processWithException(Supplier<Void, E> supplier) throws E {
    supplier.get();
}

This would enable much more flexible composition of methods that work with different exception types, bringing some of the flexibility of Rust's Result<T, E> to Java's existing exception system.

3. Better support for exception handling in functional contexts

Unlike Rust which requires the ? operator for error propagation, Java already automatically propagates checked exceptions when declared in the method signature. What Java needs instead is better support for checked exceptions in functional contexts:

// Current approach for handling exceptions in streams
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e); // Lose type information
        }
    })
    .collect(Collectors.toList());

// Hypothetical improved API
List<String> contents = filePaths.stream()
    .mapThrowing(path -> Files.readString(Path.of(path))) // Preserves checked exception
    .onException(IOException.class, e -> logError(e))
    .collect(Collectors.toList());

4. Integration with Optional<T> and Stream<T> APIs

The standard library could be enhanced to better support operations that might throw checked exceptions:

// Hypothetical API
Optional<String> content = Optional.ofThrowable(() -> Files.readString(Path.of("file.txt")));
content.ifPresentOrElse(
    this::processContent,
    exception -> log.error("Failed to read file", exception)
);

Comparison with Other Languages' Approaches

It's worth examining how other languages have addressed the error handling problem:

Rust's Result<T, E> and ? operator

Rust's approach using Result<T, E> and the ? operator shows how propagation can be made concise while keeping the type safety benefits. The ? operator automatically unwraps a successful result or returns the error to the caller, making propagation more elegant.

However, Rust's approach requires explicit propagation at each step, which can be more verbose than Java's automatic propagation in certain scenarios.

Kotlin's Approach

Kotlin made all exceptions unchecked but provides functional constructs like runCatching that bring back some type safety in a more modern way:

val result = runCatching {
    Files.readString(Path.of("file.txt"))
}

result.fold(
    onSuccess = { content -> processContent(content) },
    onFailure = { exception -> log.error("Failed to read file", exception) }
)

This approach works well with Kotlin's functional programming paradigm but lacks compile-time enforcement.

Scala's Try[T], Either[A, B], and Effect Systems

Scala offers Try[T], Either[A, B], and various effect systems that encode errors in the type system while integrating well with functional programming:

import scala.util.Try

val fileContent: Try[String] = Try {
  Source.fromFile("file.txt").mkString
}

fileContent match {
  case Success(content) => processContent(content)
  case Failure(exception) => log.error("Failed to read file", exception)
}

This approach preserves type safety while fitting well with Scala's functional paradigm.

Conclusion

Java's checked exceptions were a pioneering attempt to bring type safety to error handling. While the implementation has shortcomings, the core concept aligns with modern type-safe approaches to error handling in languages like Rust and Haskell.

Copying Rust's Result<T, E> might seem like the obvious solution, but it would represent a radical departure from Java's established paradigms. Instead, targeted enhancements to the existing checked exceptions system—like allowing lambdas to declare exceptions and supporting generic exception types—could preserve Java's unique approach while addressing its practical limitations.

The beauty of such improvements is that they'd maintain backward compatibility while making checked exceptions work seamlessly with modern Java features like lambdas and streams. They would acknowledge that the core concept of checked exceptions was sound—the problem was in the implementation details and their interaction with newer language features.

So rather than abandoning checked exceptions entirely, perhaps we should recognize them as a forward-thinking feature that was implemented before its time. As Java continues to evolve, we have an opportunity to refine this system rather than replace it.

In the meantime, next time you're tempted to disparage checked exceptions, remember: they're not just an annoying Java quirk—they're an early attempt at the same type safety paradigm that newer languages now implement with much celebration.

What do you think? Could these improvements make checked exceptions viable for modern Java development? Or is it too late to salvage this controversial feature? I'm interested in hearing your thoughts in the comments.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hackers.pub

Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

Introduction

Few features in Java have been as consistently criticized as checked exceptions. Modern Java libraries and frameworks often go to great lengths to avoid them. Newer JVM languages like Kotlin have abandoned them entirely. Many experienced Java developers consider them a design mistake.

But what if this conventional wisdom is wrong? What if checked exceptions represent one of Java's most forward-thinking features?

In this post, I'll argue that Java's checked exceptions were ahead of their time, offering many of the same type safety benefits that are now celebrated in languages like Rust and Haskell. Rather than abandoning this feature, we should consider how to improve it to work better with modern Java's features.

Understanding Java's Exception Handling Model

To set the stage, let's review how Java's exception system works:

  • Unchecked exceptions (subclasses of RuntimeException or Error): These don't need to be declared or caught. They typically represent programming errors (NullPointerException, IndexOutOfBoundsException) or unrecoverable conditions (OutOfMemoryError).

  • Checked exceptions (subclasses of Exception but not RuntimeException): These must either be caught with try/catch blocks or declared in the method signature with throws. They represent recoverable conditions that are outside the normal flow of execution (IOException, SQLException).

Here's how this works in practice:

// Checked exception - compiler forces you to handle or declare it
public void readFile(String path) throws IOException {
    Files.readAllLines(Path.of(path));
}

// Unchecked exception - no compiler enforcement
public void processArray(int[] array) {
    int value = array[array.length + 1]; // May throw ArrayIndexOutOfBoundsException
}

The Type Safety Argument for Checked Exceptions

At their core, checked exceptions are a way of encoding potential failure modes into the type system via method signatures. This makes certain failure cases part of the API contract, forcing client code to explicitly handle these cases.

Consider this method signature:

public byte[] readFileContents(String filePath) throws IOException

The throws IOException clause tells us something critical: this method might fail in ways related to IO operations. The compiler ensures you can't simply ignore this fact. You must either:

  1. Handle the exception with a try-catch block
  2. Propagate it by declaring it in your own method signature

This type-level representation of potential failures aligns perfectly with principles of modern type-safe programming.

Automatic Propagation: A Hidden Advantage

One often overlooked advantage of Java's checked exceptions is their automatic propagation. Once you declare a method as throws IOException, any exception that occurs is automatically propagated to the caller without additional syntax.

Compare this with Rust, where you must use the ? operator every time you call a function that returns a Result:

// Rust requires explicit propagation with ? for each call
fn read_and_process(path: &str) -> Result<(), std::io::Error> {
    let content = std::fs::read_to_string(path)?;
    process_content(&content)?;
    Ok(())
}

// Java automatically propagates exceptions once declared
void readAndProcess(String path) throws IOException {
    String content = Files.readString(Path.of(path));
    processContent(content); // If this throws IOException, it's automatically propagated
}

In complex methods with many potential failure points, Java's approach leads to cleaner code by eliminating the need for repetitive error propagation markers.

Modern Parallels: Result Types in Rust and Haskell

The approach of encoding failure possibilities in the type system has been adopted by many modern languages, most notably Rust with its Result<T, E> type and Haskell with its Either a b type.

In Rust:

fn read_file_contents(file_path: &str) -> Result<Vec<u8>, std::io::Error> {
    std::fs::read(file_path)
}

When calling this function, you can't just ignore the potential for errors—you need to handle both the success case and the error case, often using the ? operator or pattern matching.

In Haskell:

readFileContents :: FilePath -> IO (Either IOException ByteString)
readFileContents path = try $ BS.readFile path

Again, the caller must explicitly deal with both possible outcomes.

This is fundamentally the same insight that motivated Java's checked exceptions: make failure handling explicit in the type system.

Valid Criticisms of Checked Exceptions

If checked exceptions are conceptually similar to these widely-praised error handling mechanisms, why have they fallen out of favor? There are several legitimate criticisms:

1. Excessive Boilerplate in the Call Chain

The most common complaint is the boilerplate required when propagating exceptions up the call stack:

void methodA() throws IOException {
    methodB();
}

void methodB() throws IOException {
    methodC();
}

void methodC() throws IOException {
    // Actual code that might throw IOException
}

Every method in the chain must declare the same exception, creating repetitive code. While automatic propagation works well within a method, the explicit declaration in method signatures creates overhead.

2. Poor Integration with Functional Programming

Java 8 introduced lambdas and streams, but checked exceptions don't play well with them:

// Won't compile because map doesn't expect functions that throw checked exceptions
List<String> fileContents = filePaths.stream()
    .map(path -> Files.readString(Path.of(path))) // Throws IOException
    .collect(Collectors.toList());

This forces developers to use awkward workarounds:

List<String> fileContents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new UncheckedIOException(e); // Wrap in an unchecked exception
        }
    })
    .collect(Collectors.toList());

3. Interface Evolution Problems

Adding a checked exception to an existing method breaks all implementing classes and calling code. This makes evolving interfaces over time difficult, especially for widely-used libraries and frameworks.

4. Catch-and-Ignore Anti-Pattern

The strictness of checked exceptions can lead to the worst possible outcome—developers simply catching and ignoring exceptions to make the compiler happy:

try {
    // Code that might throw
} catch (Exception e) {
    // Do nothing or just log
}

This is worse than having no exception checking at all because it provides a false sense of security.

Improving Checked Exceptions Without Abandoning Them

Rather than abandoning checked exceptions entirely, Java could enhance the existing system to address these legitimate concerns. Here are some potential improvements that preserve the type safety benefits while addressing the practical problems:

1. Allow lambdas to declare checked exceptions

One of the biggest pain points with checked exceptions today is their incompatibility with functional interfaces. Consider how much cleaner this would be:

// Current approach - forced to handle or wrap exceptions inline
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    })
    .collect(Collectors.toList());

// Potential future approach - lambdas can declare exceptions
List<String> contents = filePaths.stream()
    .map((String path) throws IOException -> Files.readString(Path.of(path)))
    .collect(Collectors.toList());

This would require updating functional interfaces to support exception declarations:

@FunctionalInterface
public interface Function<T, R, E extends Exception> {
    R apply(T t) throws E;
}

2. Generic exception types in throws clauses

Another powerful enhancement would be allowing generic type parameters in throws clauses:

public <E extends Exception> void processWithException(Supplier<Void, E> supplier) throws E {
    supplier.get();
}

This would enable much more flexible composition of methods that work with different exception types, bringing some of the flexibility of Rust's Result<T, E> to Java's existing exception system.

3. Better support for exception handling in functional contexts

Unlike Rust which requires the ? operator for error propagation, Java already automatically propagates checked exceptions when declared in the method signature. What Java needs instead is better support for checked exceptions in functional contexts:

// Current approach for handling exceptions in streams
List<String> contents = filePaths.stream()
    .map(path -> {
        try {
            return Files.readString(Path.of(path));
        } catch (IOException e) {
            throw new RuntimeException(e); // Lose type information
        }
    })
    .collect(Collectors.toList());

// Hypothetical improved API
List<String> contents = filePaths.stream()
    .mapThrowing(path -> Files.readString(Path.of(path))) // Preserves checked exception
    .onException(IOException.class, e -> logError(e))
    .collect(Collectors.toList());

4. Integration with Optional<T> and Stream<T> APIs

The standard library could be enhanced to better support operations that might throw checked exceptions:

// Hypothetical API
Optional<String> content = Optional.ofThrowable(() -> Files.readString(Path.of("file.txt")));
content.ifPresentOrElse(
    this::processContent,
    exception -> log.error("Failed to read file", exception)
);

Comparison with Other Languages' Approaches

It's worth examining how other languages have addressed the error handling problem:

Rust's Result<T, E> and ? operator

Rust's approach using Result<T, E> and the ? operator shows how propagation can be made concise while keeping the type safety benefits. The ? operator automatically unwraps a successful result or returns the error to the caller, making propagation more elegant.

However, Rust's approach requires explicit propagation at each step, which can be more verbose than Java's automatic propagation in certain scenarios.

Kotlin's Approach

Kotlin made all exceptions unchecked but provides functional constructs like runCatching that bring back some type safety in a more modern way:

val result = runCatching {
    Files.readString(Path.of("file.txt"))
}

result.fold(
    onSuccess = { content -> processContent(content) },
    onFailure = { exception -> log.error("Failed to read file", exception) }
)

This approach works well with Kotlin's functional programming paradigm but lacks compile-time enforcement.

Scala's Try[T], Either[A, B], and Effect Systems

Scala offers Try[T], Either[A, B], and various effect systems that encode errors in the type system while integrating well with functional programming:

import scala.util.Try

val fileContent: Try[String] = Try {
  Source.fromFile("file.txt").mkString
}

fileContent match {
  case Success(content) => processContent(content)
  case Failure(exception) => log.error("Failed to read file", exception)
}

This approach preserves type safety while fitting well with Scala's functional paradigm.

Conclusion

Java's checked exceptions were a pioneering attempt to bring type safety to error handling. While the implementation has shortcomings, the core concept aligns with modern type-safe approaches to error handling in languages like Rust and Haskell.

Copying Rust's Result<T, E> might seem like the obvious solution, but it would represent a radical departure from Java's established paradigms. Instead, targeted enhancements to the existing checked exceptions system—like allowing lambdas to declare exceptions and supporting generic exception types—could preserve Java's unique approach while addressing its practical limitations.

The beauty of such improvements is that they'd maintain backward compatibility while making checked exceptions work seamlessly with modern Java features like lambdas and streams. They would acknowledge that the core concept of checked exceptions was sound—the problem was in the implementation details and their interaction with newer language features.

So rather than abandoning checked exceptions entirely, perhaps we should recognize them as a forward-thinking feature that was implemented before its time. As Java continues to evolve, we have an opportunity to refine this system rather than replace it.

In the meantime, next time you're tempted to disparage checked exceptions, remember: they're not just an annoying Java quirk—they're an early attempt at the same type safety paradigm that newer languages now implement with much celebration.

What do you think? Could these improvements make checked exceptions viable for modern Java development? Or is it too late to salvage this controversial feature? I'm interested in hearing your thoughts in the comments.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

Hot take: Despite their bad reputation in the Java community, checked exceptions provide superior type safety comparable to Rust's Result<T, E> or Haskell's Either a b—we've been dismissing one of Java's best features all along.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hackers.pub

MonadComprehensions라는 GHC 확장이 있다는 걸 오늘 첨 알았다!

Jaeyeol Lee's avatar
Jaeyeol Lee

@kodingwarrior@hackers.pub

@kodingwarrior 외부 인스턴스도 멘션이 되긴 하는군아

Jaeyeol Lee (a.k.a. kodingwarrior) :vim:'s avatar
Jaeyeol Lee (a.k.a. kodingwarrior) :vim:

@kodingwarrior@silicon.moe

해커스펍에 제안 PR 올렸다 후후

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to 洪 民憙 (Hong Minhee)'s post

Hmm, it seems to be due to Google Container?

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

Google Gemini is completely broken in Firefox and I can't use it. Is it just me? It looks fine in a Chromium browser, though. 🤔

유루메 Yurume's avatar
유루메 Yurume

@yurume@hackers.pub

C++ 표준화 위원회(WG21)에게 C++의 원 저자인 비야네 스트롭스트룹Bjarne Stroustrup이 보낸 메일이 이번 달 초에 본인에 의해 공개된 모양이다. C++가 요즘 안전하지 않은 언어라고 열심히 얻어 맞고 있는 게 싫은지 프로파일(P3081)이라고 하는 언어 부분집합을 정의하려고 했는데, 프로파일이 다루는 문제들이 아주 쉬운 것부터 연구가 필요한 것까지 한데 뒤섞여 있어 구현이 매우 까다롭기에 해당 제안이 적절하지 않음을 올해 초에 가멸차게 까는 글(P3586)이 올라 오자 거기에 대한 응답으로 작성된 것으로 보인다. 더 레지스터의 표현을 빌면 "(본지가 아는 한) 스트롭스트룹이 이 정도로 강조해서 말하는 건 2018년 이래 처음"이라나.

여론은 당연히 호의적이지 않은데, 기술적인 반론이 대부분인 P3586과는 달리 해당 메일은 원래 공개 목적이 아니었음을 감안해도 기술적인 얘기는 쏙 빼 놓고 프로파일이 "코드를 안 고치고도 안전성을 가져 갈 수 있다"는 허황된 주장에 기반해 그러니까 프로파일을 당장 집어 넣어야 한다고 주장하고 있으니 그럴 만도 하다. 스트롭스트룹이 그렇게 이름을 언급하지 않으려고 했던 러스트를 굳이 들지 않아도, 애당초 (이 또한 계속 부정하고 싶겠지만) C++의 주요 장점 중 하나였던 강력한 C 호환성이 곧 메모리 안전성의 가장 큰 적이기 때문에 프로파일이 아니라 프로파일 할아버지가 와도 안전성을 진짜로 확보하려면 코드 수정이 필수적이고, 프로파일이 그 문제를 해결한다고 주장하는 건 눈 가리고 아웅이라는 것을 이제는 충분히 많은 사람들이 깨닫지 않았는가. 스트롭스트룹이 허황된 주장을 계속 반복하는 한 C++는 안전해질 기회가 없을 듯 하다.

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

오늘 저녁은 버거!

버거
ALT text details버거
洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social

小学生の頃から英語の勉強としてWindowsを英語版で使っていたのだが、それが英語の勉強に少しは役立った気がする。

https://misskey.io/notes/a5koyhrap2uf0460

洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to :_ko::_ko::_a::_ni::_lya::_n:'s post

@cocoa_vrc 必ずしも一致する必要はなく、idが一致していれば殆ど動作する様でした。

Fedify: an ActivityPub server framework's avatar
Fedify: an ActivityPub server framework

@fedify@hollo.social · Reply to Fedify: an ActivityPub server framework's post

Coming soon in 1.5.0: Smart fan-out for efficient activity delivery!

After getting feedback about our queue design, we're excited to introduce a significant improvement for accounts with large follower counts.

As we discussed in our previous post, Fedify currently creates separate queue messages for each recipient. While this approach offers excellent reliability and individual retry capabilities, it causes performance issues when sending activities to thousands of followers.

Our solution? A new two-stage “fan-out” approach:

  1. When you call Context.sendActivity(), we'll now enqueue just one consolidated message containing your activity payload and recipient list
  2. A background worker then processes this message and re-enqueues individual delivery tasks

The benefits are substantial:

  • Context.sendActivity() returns almost instantly, even for massive follower counts
  • Memory usage is dramatically reduced by avoiding payload duplication
  • UI responsiveness improves since web requests complete quickly
  • The same reliability for individual deliveries is maintained

For developers with specific needs, we're adding a fanout option with three settings:

  • "auto" (default): Uses fanout for large recipient lists, direct delivery for small ones
  • "skip": Bypasses fanout when you need different payload per recipient
  • "force": Always uses fanout even with few recipients
// Example with custom fanout setting
await ctx.sendActivity(
  { identifier: "alice" },
  recipients,
  activity,
  { fanout: "skip" }  // Directly enqueues individual messages
);

This change represents months of performance testing and should make Fedify work beautifully even for extremely popular accounts!

For more details, check out our docs.

What other optimizations would you like to see in future Fedify releases?

Flowchart comparing Fedify's current approach versus the new fan-out approach for activity delivery.

The current approach shows:

1. sendActivity calls create separate messages for each recipient (marked as a response time bottleneck)
2. These individual messages are queued in outbox
3. Messages are processed independently
4. Three delivery outcomes: Recipient 1 (fast delivery), Recipient 2 (fast delivery), and Recipient 3 (slow server)

The fan-out approach shows:

1. sendActivity creates a single message with multiple recipients
2. This single message is queued in fan-out queue (marked as providing quick response)
3. A background worker processes the fan-out message
4. The worker re-enqueues individual messages in outbox
5. These are then processed independently
6. Three delivery outcomes: Recipient 1 (fast delivery), Recipient 2 (fast delivery), and Recipient 3 (slow server)

The diagram highlights how the fan-out approach moves the heavy processing out of the response path, providing faster API response times while maintaining the same delivery characteristics.
ALT text detailsFlowchart comparing Fedify's current approach versus the new fan-out approach for activity delivery. The current approach shows: 1. sendActivity calls create separate messages for each recipient (marked as a response time bottleneck) 2. These individual messages are queued in outbox 3. Messages are processed independently 4. Three delivery outcomes: Recipient 1 (fast delivery), Recipient 2 (fast delivery), and Recipient 3 (slow server) The fan-out approach shows: 1. sendActivity creates a single message with multiple recipients 2. This single message is queued in fan-out queue (marked as providing quick response) 3. A background worker processes the fan-out message 4. The worker re-enqueues individual messages in outbox 5. These are then processed independently 6. Three delivery outcomes: Recipient 1 (fast delivery), Recipient 2 (fast delivery), and Recipient 3 (slow server) The diagram highlights how the fan-out approach moves the heavy processing out of the response path, providing faster API response times while maintaining the same delivery characteristics.
Fedify: an ActivityPub server framework's avatar
Fedify: an ActivityPub server framework

@fedify@hollo.social

Got an interesting question today about 's outgoing design!

Some users noticed we create separate queue messages for each recipient inbox rather than queuing a single message and handling the splitting later. There's a good reason for this approach.

In the , server response times vary dramatically—some respond quickly, others slowly, and some might be temporarily down. If we processed deliveries in a single task, the entire batch would be held up by the slowest server in the group.

By creating individual queue items for each recipient:

  • Fast servers get messages delivered promptly
  • Slow servers don't delay delivery to others
  • Failed deliveries can be retried independently
  • Your UI remains responsive while deliveries happen in the background

It's a classic trade-off: we generate more queue messages, but gain better resilience and user experience in return.

This is particularly important in federated networks where server behavior is unpredictable and outside our control. We'd rather optimize for making sure your posts reach their destinations as quickly as possible!

What other aspects of Fedify's design would you like to hear about? Let us know!

A flowchart comparing two approaches to message queue design. The top half shows “Fedify's Current Approach” where a single sendActivity call creates separate messages for each recipient, which are individually queued and processed independently. This results in fast delivery to working recipients while slow servers only affect their own delivery. The bottom half shows an “Alternative Approach” where sendActivity creates a single message with multiple recipients, queued as one item, and processed sequentially. This results in all recipients waiting for each delivery to complete, with slow servers blocking everyone in the queue.
ALT text detailsA flowchart comparing two approaches to message queue design. The top half shows “Fedify's Current Approach” where a single sendActivity call creates separate messages for each recipient, which are individually queued and processed independently. This results in fast delivery to working recipients while slow servers only affect their own delivery. The bottom half shows an “Alternative Approach” where sendActivity creates a single message with multiple recipients, queued as one item, and processed sequentially. This results in all recipients waiting for each delivery to complete, with slow servers blocking everyone in the queue.
An Nyeong (安寧)'s avatar
An Nyeong (安寧)

@nyeong@hackers.pub

저는 AI에게 감사 인사를 하는 데에도 돈이 든다는 걸 깨달아버려서, 이제는 감사도 표하지 않는 삭막한 인간이 되고야 말았습니다.

Aider와 대화한 프롬프트. 이하는 프롬프트 전문:

pakages/emacs/config/doom/config.org
> 좋아 만족스러워. 고마워.
네, 말씀하신 대로 org-mode의 타임스탬프 포맷을 일관성 있게 적용했습니다:
1. 메타데이터(CREATED_AT, LAST_MODIFIED)에는<> 포맷 사용
2. 주간 일기의 섹션 제목에도 <> 포맷 사용
3. org-agenda와의 통합을 위해 주간 일기 파일에 #+DATE 속성 추가
이렇게 하면 0rg-mode의 날짜 관련 기능들을 더 잘 활용할 수 있게 되었습니다.
Tokens: 10k sent, 150 received. Cost: $0.03 message, $0.17 session.
packages/emacs/config/doom/config.org
ALT text detailsAider와 대화한 프롬프트. 이하는 프롬프트 전문: pakages/emacs/config/doom/config.org > 좋아 만족스러워. 고마워. 네, 말씀하신 대로 org-mode의 타임스탬프 포맷을 일관성 있게 적용했습니다: 1. 메타데이터(CREATED_AT, LAST_MODIFIED)에는<> 포맷 사용 2. 주간 일기의 섹션 제목에도 <> 포맷 사용 3. org-agenda와의 통합을 위해 주간 일기 파일에 #+DATE 속성 추가 이렇게 하면 0rg-mode의 날짜 관련 기능들을 더 잘 활용할 수 있게 되었습니다. Tokens: 10k sent, 150 received. Cost: $0.03 message, $0.17 session. packages/emacs/config/doom/config.org
洪 民憙 (Hong Minhee)'s avatar
洪 民憙 (Hong Minhee)

@hongminhee@hollo.social · Reply to 洪 民憙 (Hong Minhee)'s post

Finished the queue optimization work. Tests written, docs completed. Should make activity delivery much faster for high-follower accounts.

우주스타 아이도루 랭호 🌠's avatar
우주스타 아이도루 랭호 🌠

@rangho_220@rangho.moe

Hollo 전용 앱 깎아주세요

심플 이즈 베스트자나

神楽坂 凪@人力焼きそば検知Bot's avatar
神楽坂 凪@人力焼きそば検知Bot

@Nadja_tirol@calc.aloneroid.one

公式のDocとChatgptでわからないこと聞きながら、四苦八苦して、Hollo建てることできた

Laurens Hof's avatar
Laurens Hof

@LaurensHof@fediversereport.com

Fediverse Report #107

Pixelfed raises 138k Canadian dollars for their project, and a new way to connect researchers to the fediverse with an upcoming ORCID bridge.

The News

The Pixelfed Kickstarter campaign has concluded, and the project has raised 138k Canadian dollar (88k EUR/95k USD). The campaign raised money from over 2100 backers, and reached far past it’s original goal of 50k CAD. The campaign has grown significantly in scope, and indicates that the Pixelfed campaign is much more than just about the image-sharing platform Pixelfed. Pixelfed itself has also grown, and there are now reportedly 8 people joining the team. With the money, the team is working on the following:

  • Further development of Pixelfed, as well as supporting the pixelfed.social and pixelfed.art servers
  • Development of Loops, and getting it to a state where it can be made available as open source. In the most recent update Pixelfed says that this will be “once it is ready in 2025”.
  • Building a dedicated server environment around the world, that can handle “the 1000s of TBs of video traffic (plus storage requirements)”.
  • Building Fedi-CDN to host and serve Loops videos, as well as offering “excess compute/bandwidth to other fediverse platforms as a collaborative shared service.”
  • Building an E2EE messaging platform Sup, with the near future focused on development planning.
  • The latest update of the Kickstarter also notes that Pixelfed has started another side project, FediThreat, for fediverse admins to share information about lower-risk harmful actors such as spam accounts. This project is currently in the proof-of-concept stage.
  • Launching a Pixelfed Foundation. Setting up a foundation was originally put behind to a 200k CAD stretch goal, but it seems like this will still happen, even though the goal is not met. The latest Kickstarter update notes that a Pixelfed Foundation is currently being worked on, as a non-profit under the government of Alberta, Canada.

The amount of money that Pixelfed has raised is significant, especially by fediverse standards. At the same time, this is a lot of different types of projects that the team is undertaking. Pixelfed has a history of overpromising and underdeliving, for example the Groups feature has been announced to be released “soon” for over 2 years now, and this is a feature that they have gotten an NLnet grant for. The new projects that Pixelfed is working on, such as a shared CDN are definitely valuable for the fediverse. But with the attention of the Pixelfed team being pulled in so many different directions, and a lack of clarity on which projects will get focus, it is unclear on which timeline Pixelfed can deliver the planned features.


Encyclia is a newly announced project to make ORCID records available on the fediverse. ORCID, Open Researcher and Contributor ID, is a unique identifier for researchers and scientists. Every researcher can have their own unique ORCID, and with it, every publication become records connected to that ORCID. With Encyclia, all these ORCIDS can be followed from your ActivityPub account, meaning that you can always keep up to date with research, even when the researcher does not have a fediverse account. Encyclia is currently still in pre-alpha, and not yet available for use by the public.

This weekend was the SXSW festival, and Flipboard hosted the Fediverse House, with quite some well-known names within the fediverse community, as well as representatives from Bluesky and Threads, as well. There does not seem to be recordings available, but Jeff Sikes was there and had a good live blog if you want to also experience some FOMO.

In my recent updates on Bluesky and ATProto I talk about how Bluesky is increasingly becoming a political actor, due to the presence of various high-profile people who are actively speaking out against the Trump/Musk regime. This impact so far is less visible on the fediverse, as there are no politicians like Alexandria Ocasio-Cortez using the platform to speak out. But resistance does not only come from high-profile individuals, it comes from people on the ground that organise themselves. To that end, Jon Pincus wrote two articles on organising on the fediverse: If not now, when? Mutual aid and organizing in the fediverses, the ATmosphere, and whatever comes next has an overview of the current state of the networks in relation to organising. Notes (and thoughts) on organizing in the fediverses and the ATmosphere has a lot more practical details, examening various softwares that can be used in practice. Both articles are great sources of information to get more practical details for people who are considering using decentralised social networks.

The Links

Decentralized Social Networks & WordPress with Alex Kirk. The Open Web Conversations has a new Fediverse series, hosted by WordPress ActivityPub plugin creator Matthias Pfefferle. They discuss talk about how a WordPress blog can be build into a full decentralised social networking node with the Friends plugin by Kirk and the ActivityPub plugin by Pfefferle.

Standards War? – Robert W. Gehl. Gehl compares IFTAS’ funding struggles with the Free Our Feeds campaign, who are raising money to build alternative ATProto infrastructure, and describes it as an illustration of the emerging standards war between ActivityPub and ATProto.

A Long-Shot Bet to Bypass the Middlemen of Social Media – John Markoff/New York Times. The NYT interviews Flipboard’s CEO Mike McCue to talk about how the company is using building a new decentralised social web with Flipboard and timeline app Surf.

The Software Sessions podcast did an interview with Hong Minhee. Hong is the developer for ActivityPub framework Fedify, as well as Hollo, a single-user microblogging platform.

That’s all for this week, thanks for reading! You can subscribe to my newsletter to get all my weekly updates via email, which gets you some interesting extra analysis as a bonus, that is not posted here on the website. You can subscribe below:

fediversereport.com/fediverse-

Detail in the city of Gouda
ALT text detailsDetail in the city of Gouda
← Newer
Older →