Hacker News Reader

Comment by xiphias2 | Hacker News Reader

Just looking at some .NET programs, they all use unsafe code, which is always faster than using garbage collected memory management, so the results aren't surprising.

View on HN

It's using features of the language. Is that not what this is meant to measure/compare? .NET has unsafe, SIMD, and value types which opens it to a whole world of programming that Java is currently unsuited for.

One would think that a benchmark described in reference to the performance of the virtual machines themselves, would be measuring the performance of pure bytecode (or a JITing thereof), not the performance of native code linked into the runtime.

Using a native regex extension lib is the norm for "scripting" languages so totally reasonable to benchmark like that. How do you define when it's OK to use a native extension and when it's cheating?

You can call C with Java.

For example

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

Benchmarking both C# [Managed] and C# [Unsafe] is preferred.

Like this ?

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

If you want performance, you need to refactor your code into a small performance-critical unsafe core (that has higher test coverage and audit standards), while keeping the rest of the business logic in the safe domain, and relying on the CLR to do the heavy lifting and guarantee memory consistency.

Bonus point: you can write the core in C++ and use COM-like marshaling for absolutely unbeatable performance (no actual COM runtime is needed).

Maybe look at the .NET programs that do not use unsafe code?

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

Gaming the benchmarks for fun and profit.

Reply on news.ycombinator.com