A simple program reducer for any language | Hacker News Reader

A simple program reducer for any language

comby.dev

129 pointsingve17 comments

Comby looks great. I've been wondering recently why there isn't a mature tool for both language agnostic lining and transforms. I started using semgrep for linting various languages but I just as soon want to run transformations.

Maybe Comby is what I'm looking for for both cases.

Comby looks nice, I miss the "tops" tool from the old MacOSX ecosystem.

For a decade or so I've been using the "delta" tool for this[1,2], although it sometimes requires manual intervention to get an extremely small testcase because it has very little understanding of the source language's syntax. On the other hand, that genericness can be an advantage for custom languages and it's usually pretty fast at getting to the nut of the matter anyway.

(On the other hand, I don't have any evidence to suggest that a compiler bug report with a 5-line testcase gets looked at any sooner than one with a 1000-line testcase.)

Aside from "minimization" or "reduction" this technique is also sometimes called "delta debugging"[3].

[1] https://github.com/dsw/delta [2] https://tracker.debian.org/pkg/delta [3] https://en.wikipedia.org/wiki/Delta_debugging

Is this the same basic principle as the back half of property based testing? Winnowing the example down to the critical components?

The full thing would be an example of property based testing, I believe, based on the property "the compiler doesn't crash".

Very cool stuff!

And exactly what I was looking for when hitting compiler errors in Zig.

Thank you!

I hadn't heard the concept of a reducer. Very cool!

Erlang QuickCheck, and I would guess Haskell as well, offers a reducer. It’s really quite impressive.

Any John Hughes talk is worth watching; here’s one where he demonstrates the functionality.

https://youtu.be/gPFSZ8oKjco

So does Clojure's test.check. Even though Clojure ain't typed you can define "specs" of your data and then have generators generating random data (it's not the only use of specs: they can also be used as a way to verify inputs to functions). The generators data can be used to, say, quickly create fake database with lots of data or fed into test.check which then does fuzzy testing and then tries hard to reduce failing test cases to minimal cases.

There's also DustMite for D. https://dlang.org/blog/2020/04/13/dustmite-the-general-purpo... It's fairly D specific though.

I've submitted dozens of bugs with it. It's pretty great.

C-Vise and C-Reduce are common ones for C/C++:

https://github.com/marxin/cvise http://embed.cs.utah.edu/creduce/

I hadn't either. But I refer to the method as 'code isolation'.

There are a bunch of related stuff that's interesting:

- code slicing (known as code-splitting in JS community).

- round-trip compilation

Yep, comby did good job for me by reducing 10 kB Rust program, which crashes 2 months old nightly compiler, to 5.5 kB minimal one.

Not sure is I should report this bug.

Is anyone else having difficulty viewing this on firefox? I'm getting a HSTS error.

Not an issue on Nightly for Android 11.

No problem.

I will pay good money for a `mergetool` built around Comby.

Does it work for Rust?

Reply on news.ycombinator.com