Real-Time LuaTeX: Recompiling Large Documents in 1ms [pdf] | Hacker News Reader

Real-Time LuaTeX: Recompiling Large Documents in 1ms [pdf]

tug.org

66 pointsamichail23 comments

the purpose (from the paper):

> texlode, the browser-based book editor built on this architecture, is scheduled for public release in October 2026. It provides collaborative editing via conflict-free replicated data types (CRDTs), manu- script import from Word, proceedings management, cover design tools, and print-ready PDF output iden- tical to standard LuaLATEX. The accompanying con- ference talk demonstrates the editor in action. For details: texlode.com.

and also:

> (section) 6 Comparison to Typst

This was presented earlier today, and you can view the recorded video at [0].

[0]: https://www.youtube.com/live/6riWbnT0grs?si=zIePpHrC6TEZpc_I... (starts at 2:52:46) [1]

[1]: Once the conference is over, I'll separate each presentation out into separate videos, but I'm busy at the moment with filming the current presentations :).

> Benchmarks show that per-paragraph recompilation achieves O(1) latency, constant regardless of document size, whereas Typst’s [3] incremental compilation scales linearly (O(n)).

> The tradeoff is temporary inconsistency: pages the user is not viewing may lag until a background compile converges, [...]

There doesn't seem to be any reason functionality like this couldn't also be added to Typst though. In general the authors of this paper seem dismissive of typst, but Typst also fixes so many other things about LaTeX, like the awful syntax. Not sure why they act like that.

> Not sure why they act like that.

They seem to launch paid competitor.

The answer is fairly obvious, Typst (like ConTeXt) is not LaTeX. Typst is a perfectly fine solution if you're giving the typeset output (i.e. a PDF) to someone, it not a solution if you need to give the document source to someone (if the person expects LaTeX, or .doc(x), or some other format). Many of the issues people raise with LaTeX are due to needing to pass on the document source to someone (otherwise you could use whatever packages or engines you wanted), but that is fundamental to interoperability and why LaTeX (and Markdown) stick around.

> There doesn't seem to be any reason functionality like this couldn't also be added to Typst though.

If I have to guess it's because the temporal inconsistency tradeoff can actually affect the current page too, since it might depend on the previous pages for layout, references, etc etc.

Typst on the other hand aims to have no inconsistencies due to incremental compilation.

> There doesn't seem to be any reason functionality like this couldn't also be added to Typst though.

Correct, and the paper author even suggested that they could add that in their presentation earlier today [0].

> In general the authors of this paper seem dismissive of typst, but Typst also fixes so many other things about LaTeX, like the awful syntax. Not sure why they act like that.

Well this was a presentation at a (La)TeX conference [1], and this is a preprint for a (La)TeX journal [2], so this specific audience is more interested in hearing about extensions to LaTeX than about Typst. (This certainly isn't a requirement though—we've happily accepted articles and presentations about non-TeX software, since we always like to hear about what's going on elsewhere)

[0]: https://www.youtube.com/live/6riWbnT0grs?si=2ls7VrAIE9WCU5kK... (starts at 3:12:09)

[1]: https://tug.org/tug2026/

[2]: https://tug.org/TUGboat/

This is great but I feel like the potential is underused.

I know its a lot of work but I think we all could desperately use an interactive LaTeX notebook instead of limiting it to only the PDF backend and this work helps a lot with achieving that endeavour. Though it'll be limited to LuaTeX, it'll still be quite something.

Back in the 90's, one of my professors had such software on Solaris.

The experience was just like using Word, but with LaTeX.

Unfortunately not the kind of stuff the vi and emacs folks care about.

Although LyX comes close.

Was it BaKoMa TeX [0] perhaps?

[0]: https://bakoma-tex.daniel-aldrich.ca/home

I don't remember the name, this one also looks cool.

No mention of TikZ. Realtime TikZ would be really useful to me. I will wait and see what their https://texlode.com service is like for that.

The author said that TikZ works correctly in a message during their presentation [0]:

> 3:27:36 @lodepublishing : I have tested tikz but not yet tcolorbox. TikZ works, with some wiring.

But a different tool was posted here a few weeks ago with real-time support for TikZ [1] if you're interested in something that you can use immediately.

[0]: https://www.youtube.com/live/6riWbnT0grs

[1]: https://news.ycombinator.com/item?id=48645437

Oh good, thanks.

[1] is a totally different kind of tool that generates unmaintainable code full of absolute positions.

Looks like layout/format of this PDF is heavily broken - line overflows and out of the context/layout line at the end...

This is (surprisingly) intentional; see my other comment [0].

[0]: https://news.ycombinator.com/item?id=48966387

Typst taking 300ms to do 300 pages cannot be seen as 4FPS - you don't treat the sum of 300 pages as single frame, LOL. And laying out a single paragraph is not 1ms for a whole document either. The paper is otherwise pretty good, but these rhetorics are probably not necessary, IMO.

It is no rhetoric. It is the time spent when working interactively on a thesis project with one software vs the other.

Personally, I am enjoying using Typst.

Same, so much in fact that I have used it for my website now and it works beautifully for graphics and formulas.

What packages have you found useful?

unfortunately the pdf itself has some typesetting issues. Whats about all those overflowing boxes and the small black boxes all over the document?

You can't tell from the URL, but this link is to a preprint [0], and the bad boxes there are intentional, since we compile all the preprints in draft mode [1]. The final article will be published some time in August [2], and should be perfectly typeset.

[0]: https://tug.org/tug2026/program.html#:~:text=Clemens%C2%A0Lo...

[1]: https://tug.org/tug2026/cfp.html#preprints

[2]: https://tug.org/TUGboat/#:~:text=vol%2E47%2C%20no%2E2%20%28T...

Reply on news.ycombinator.com