Hacker News new | ask | show | jobs
by CompanyGardener 25 days ago
Apologies for the slow reply, my lack of karma is throttling how often I can comment.

I appreciate you sharing your thoughts, the whole point of a public preview is to find out how things are received, what’s working, what isn’t.

We more or less agree on the vibe-coded website (especially the copywriting). It’ll improve over time, but this is a side-project, time is limited, and I wanted to release something on 7/4.

Tungsten’s grammar and design are mine.

Tungsten’s implementation, is almost entirely LLM (lowering, escape analysis, SSA), with a good bit of prodding, benchmarking, troubleshooting; in order to have the performance land remotely close to where I knew it could.

While my primary motivation was to build the language that I wish existed, if I can make something that others enjoy or find useful, even better.

To address your other point:

Whitespace is almost certainly getting tokenized whether it’s significant to the language or not. And I would expect most folks are indenting their blocks anyway, even when they are delimited.

I think you’ll find the token reduction comes from more than just dropping delimiters.

I won’t try to litigate the point here, I’ll save that for a longer-form write up that is lacking on the website.

Thanks again for the feedback.

Related (in the repo):

WValue layout https://github.com/tungsten-lang/tungsten/blob/d5f505acca245...

Ruby Lexer (regex-based) https://github.com/tungsten-lang/tungsten/blob/d5f505acca245...

Ruby Lexer (spec) https://github.com/tungsten-lang/tungsten/blob/d5f505acca245...

1 comments

Thanks for the in-depth reply, appreciate the elaboration on where you did and didn't use LLMs. Looking forward to reading the blog posts and other future updates!