[weinzierl]

I used it a bit recently and it is brilliant. Instead of lugging a huge TeX installation around you have a single reasonably sized binary. When you run ’tectonic file.tex’ it will download everything it needs on the fly. If you want to run it without internet access you can sort of freeze the downloaded files and run tectonic with your *.tex plus the frozen bundle.

It also claims reasonable error messages but I only ran it on error-free files, so I cannot tell.

One small downside is that in my direct comparison with lualatex it was always marginally slower.

Another one is that it is not a rewrite but still based on the old TeX code.

I think tectonic is a great project but generally speaking I find it quite disappointing that we do not have 100% compatible LaTeX with a modern codebase. And that is not even because the old code is bad - not at all. It's just that a pure Rust or Java implementation would make it so much easier to integrate into other projects. An audited TeX core we could run on untrusted input would be another benefit.

Which brings me to my last point: When researching tectonic I found a fork that attempted a pure Rust port and it nearly worked with examples. There is hope...*

[alerighi]

Why rewrite TeX? To me TeX is one of the example of what a software project should aim to be, arrive to a stable version where it does what it should do and it work reasonably well. That was the initial idea of Knuth (in fact the version is a number that approaches PI one decimal at a time).

This is the contrary of the "modern" philosophy that says that we need to constantly modify, update or rewrite a software in "modern" languages even if it's a huge effort and risks introducing bugs that were not present in the original version.

Well, at least if we want to rewrite it let's just have to wait 10 years, then we can fed all the codebase to ChatGPT and ask for a new fancy Rust version!

Regarding of integration, TeX follows the (to me still relevant) UNIX philosophy that a software can be called by another software and the input/output communicates in a pipe. Nowadays we have even containers that let us distribute the software as a single unit so we don't have to worry about installing multiple files on a system. Not a big deal, in the end.

[jhanschoo]

I don't think you realize that

1. Most people aren't compiling with tex->dvips->ps2pdf anymore, most people with basic requirements compile with pdftex for more robust hyperlink and navigation pane, etc. features

2. People who want Unicode support and support for modern font files are using xetex or luatex

3. What has been proposed isn't a new tex compiler but more like a package manager for 3rd-party tex packages, using the already established xetex compiler to compile. Whereas historically you would download the whole distribution (or a predefined subset thereof) onto your system versioned by year (e.g. TexLive 2020) eating up GiBs of space, this allows you to download just what you need for your document.

[convolvatron]

I've been looking at https://typst.app

not tex based, but uses CM fonts. I really think that trying to update the programming parts of TeX is necessary. sadly I'm struggling with fonts in svg figures.

but to suddenly see applying effort in this area and getting real traction is treat. so many thanks for the luminaries that put up the last one, but maybe its time for some rework after 40 years.

[nsajko]

Sadly it still doesn't support vector graphics. A TeX-killer must also include a replacement for something like PGF, Tikz, Metapost, etc.

[jcparkyn]

It supports SVGs. Not a replacement for all the options in latex, but that certainly qualifies as "vector graphics".

[nsajko]

I'm talking about vector graphics as an integral part of the document, not mere inclusion of external graphics.

[nsajko]

> I find it quite disappointing that we do not have 100% compatible LaTeX with a modern codebase

That would be a huge effort. Your time would be better spent designing and creating a new typesetting system.

[bobbylarrybobby]

A new typesetting system will be very hard to get people to adopt. There are a million packages and existing documents that would have to be ported over. At this point, compatibility with TeX is a critical feature, not just a nice-to-have.

[weinzierl]

New typesetting systems are a dime a dozen and none of them is getting anywhere. Also in almost all cases HTML/CSS is what you want to use anyway. TeX is pretty niche but it fills that niche perfectly. A 100% compatible LaTeX that plays nicely in today's infrastructure is all I want.

[mhh__]

Integrating TeX into a program inside exactly impossible now though. Pandoc gets along fine.

Also the way to trust code is isolation rather than verification, even rust has unsafe.

[weinzierl]

Isolation is a good second line of defense. It's not going to prevent malicious input to produce malicious output. At least not until you go as far as Qubes OS and you squeeze everything through a rasterized bitmap.

[lejalv]

Not really what you asked for -- but there is an interactive modern typesetter that is programmable and works on trees. It's called TeXmacs - you may want to check out what it can do -- https://www.youtube.com/watch?v=H46ON2FB30U