A calculator that doesn't round | Hacker News Reader

A calculator that doesn't round

constructive-calculator.dimview.org

29 pointsdimview24 comments

> Hans Boehm built a Java library for this in the 1980s and 90s

Hans is such a prolific programmer that he wrote a Java library before Java was even invented?

The article has so many of the hallmarks of LLM writing, naturally it will have some hallucinated factoids too. "What that gives you: [bullet list]", "What got ported: [bullet list]", "foo: x, y, just z"

Boehm's Java library has copyright dates of 1999 and 2001-2004. https://www.hboehm.info/new_crcalc/CR.java

The whitepaper was only published in 2005: https://www.sciencedirect.com/science/article/pii/S156783260...

I have a dimview now of the blog, and of this Swift port of the library, indeed.

Thank you. Of course the dates were out of order. The Java library (com.hp.creals) is 1999-2004, paper is from 2005. Unfortunate price of LLM-assisted development. Corrected.

Well, you know, you go into your Smalltalk-80 workstation for a hack sesh, and an all-nighter later you end up with an implementatkon of Swing...

> I built one, by porting Boehm’s engine. > It’s 2026, so I didn’t hand-write the port. I directed Opus 4.8 to translate the source line by line into Swift

I wish I could filter out stuff like this. Cool work by Hans Boehm, but what's the value add in this blog post.

Read. Guy knows what he's doing.

You mean his AI wrote something you like... which is factually incorrect, so it's basically fanfic.

Author here. I could not find a constructive‑reals calculator on iPhone, so I ported the engine Android uses: Hans Boehm's com.hp.creals plus AOSP's UnifiedReal/BoundedRational. Used Opus 4.8 to do the port and Fable 5 for the review.

Fable 5 caught a couple of real concurrency bugs the port introduced while adapting Java's synchronized/AsyncTask to Swift concurrency, including one that was a memory‑safety bug on shared singletons like π, not just a wrong digit. None would have shown up in the unit tests. Writeup has the details.

It's an early iPhone TestFlight beta (link in the post); happy to go deep on either the constructive‑reals side or the AI‑assisted‑dev side.

The old HP calculators, and their emulators, have a computer algebra system, for symbolic maths, that supported this. The user interfaces leave much to be desired, but some also have reverse Polish notation!

This is different from a CAS. For example, if you ask it to do exp(100)+1-exp(100), it does not rearrange and cancel out the two exp(100)s. Instead, it does exp, addition, and subtraction, all with as many digits as you ask for.

Sure. I was speaking to the what, not how.

> that doesn’t round at all. It computes with constructive real numbers: every result is exact, and you can scroll any answer for as many correct digits as you want.

A CAS is a practical way to achieve this, where everything is stored symbolically, with no rounding, until final calculation. Unlimited precision was through Erable [1]. This was included with/as HP49 CAS, but was an add-on with HP48. Many HP48 are on the iOS app store. The one I tried about a decade ago had the add-on, and I see many still there (but not sure if they have the add-on).

The "S" in CAS is "system", so it's a bit open.

[1] https://en.wikipedia.org/wiki/Erable

Maybe you should correct the errors in your post before worrying about the code.

Corrected the dates, thank you! What are the other errors?

I don't get why this is a big deal. All the calculators I've used have been rectangle.

Very interesting, thank you for posting! I'm curious - roughly how many tokens do you think you used during the initial port and subsequent bug hunting and fixes?

For Opus 4.8, 142.3k input, 2.0m output for the entire project. No longer have access to Fable 5 to check.

It would also be fun to have a full computer algebra system (like maxima) on a phone

That should be doable, too. Maybe as a next project.

https://webdev4.gitlab.io/fdroid-website/en/packages/jp.yhon...

btw if you turn the iphone calc into landscape mode and switch you scientific calc it does Ramanujan's constant without rounding, but stops after the twelve 9s.

Correct, but if you don't know the next digit is 2, you can think it's all 9s, another representation of an integer number.

Edit: Whoops. My bad. This must have been a HN "second chance" winner.

Er, that's this submission.

That looks like a link right back to this thread?

Reply on news.ycombinator.com