alt Hacker NewsReal-Time LuaTeX: Recompiling Large Documents in 1ms [pdf]
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 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.
> 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.
Looks like layout/format of this PDF is heavily broken - line overflows and out of the context/layout line at the end...
unfortunately the pdf itself has some typesetting issues. Whats about all those overflowing boxes and the small black boxes all over the document?
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.
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.
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 :).