Assuming the semantics of programming languages are well-defined, I'm wondering if there's any software projects that we can confidently say are entirely bug-free and not requiring any new feature additions to be considered complete.
If not, how about only considering the projects' core source code, not counting compiler/build-chain bugs and dependency bugs?
If not, what would be the closest?
TeX (not LaTex) is quite close https://en.wikipedia.org/wiki/TeX#TeX82
The Hewlett Packard calculators had their binary code inserted into the ROMs as part of the fabrication of the chip. The code was written in assembly language, then heavily tested on hardware simulators. The original HP-35 had a bug in the logarithm code. The bug had to be found and fixed, retested, new ROM chips fabricated and packages. Then owners that wanted it fixed had to mail their calculator back to HP. The unit was opened up, a ROM chip removed, the new ROM soldered in, unit closed up, tested, mailed back. All for free. There was a very strong demand that this never happen again. I am unaware of any other HP calculator recalls. Each calculator code was considered complete.
The early video games, both console and arcade, were delivered in ROM chips. Also early embedded systems like DVD/CD disk players, televisions, modules in cars, routers, etc.
I consider the static site generator [0] I wrote done. I don't plan to add new features (can't think of any). I have no idea if there are bugs left in it; I have been using it for 476 days to generate a tumblelog [1].
I doubt they'd claim to be perfectly "bug free" but this came to mind reading your post: https://github.com/boltdb/bolt
I think the GNU coreutils are mostly done.
Daniel Bernstein's qmail has a standing offer from 1997 to find a bug https://cr.yp.to/qmail/guarantee.html and the last version released was in 1998, I guess it's done. He wrote a paper about why it succeeded https://cr.yp.to/qmail/qmailsec-20071101.pdf