HN Explorer

Show HN: Civ 1 period accurate decompilation (github.com)

2 points by crc32 ↗ HN
I've used LLMs to decompile Civ 1 and put the reconstructed C source on GitHub, along with a technical book and story.

The reconstructed source builds with what i'm pretty sure is the original Microsoft C compiler and toolchain. It produces a byte-for-byte match with the original executable, aside from a four-byte correction for what appears to be a patch applied after linking. So it should be pretty close to how the original game was made; although it can't be said that this is the "original source code".

I've also used that source to create a modern build of the game, in the `reciv` repository.

Like most people, I'm in two minds about the benefits and downsides of LLMs. This work also builds on OpenCiv1, which probably deserves most of the credit - not only did I direct the LLM to use that work; but it was almost certainly in the training data.

Qualms aside - I couldn't resist having a look inside one of my all-time favourite games. There's a few attempts at Civ 1 including recent LLM ones. But i'm not aware of any that publicise the resulting source code.

6 comments

[ 1.7 ms ] story [ 16.6 ms ] thread
tuomas2
Cool! Is this an approach that you could in theory use to other games / programs too? What about license, is this legal?
crc32
The approach really boils down to just giving the LLM the clear target of 100% binary compatibility with the same toolchain. It makes it easy for it to check its work and iterate towards it.

It's probably a couple of months of intermittently using £15pm subscriptions between OpenAI, Anthropic and Google; nowhere near maxing them out.

MSC 5.1 is a relatively simple compiler; and there's lots of prior art on Civ 1. I've set them off on a slightly more obscure and recent game now; so we'll see how well they manage on that!

Regarding legality - IANAL so perhaps we'll see. I just wanted to see how the sausage was made.

tuomas2
So essentially the source code is LLM's best guess what the original might look like, as long as it compiles the same binary the goal is met.
crc32
Exactly yes.

Interestingly; the way that MSC51 and LINK 5 lay out the binary depends on identifier names. There's also length restrictions due to memory limitations. So although the names aren't going to be historically accurate; the possibilities are at least bounded!

muzani
I love the curiosity despite the possible legal consequences. It's in the spirit of Hacker News.