SILE Typesetting System

No math is pretty much a dealbreaker for most people, I guess.

Yes, it made me recall this, though I first saw it in FOSDEM '15 talks [1]. I thought it'd be nice to see what people think about this as I'm in the looks for a TeX alternative.

[1] https://www.youtube.com/watch?v=5BIP_N9qQm4

And a glaring "overfull hbox" between the two columns "means".

And shouldn't the "grid layout" feature assure that the baselines of the two columns match?

And the kerning problem in "T able of Contents"

;)

So, I'm really sorry about the examples. There are two issues here: first, I'm not good at coming up with compelling examples. Second, I'm not very good at keeping the examples on the web site up to date with the current progress of the code. In fact, I haven't regenerated those examples, like, ever. Try the PDF examples in the repository as they should be better; but at the same time we should still auto-regenerate them periodically, maybe on release. (And convert them for the web site.) Really sorry about this.

Of the issues you've mentioned: I now justify the columns; the overfull box is gone as I tightened up the linebreak tolerance on narrow columns; kerning most definitely happens automatically if the font provides kerning pairs; grid layout now works better.

Any other issues, please file bug reports and I'll fix them!

TeX by Topic was a much clearer resource to me than the TeXBook although one eventually has to go back and read that. TeX seems like a reasonable approximation of it, but it's definitely slanted towards text-heavy and mathematical stuff, which is pretty far from what something like InDesign is mostly about. I'm just an appreciator, not an implementer, but this is my overview of typesetting text:

First, you have to build words from letters; this is where kerning comes into it.

Second, you have to build lines. TeX calls this "horizontal mode" and this is where inter-word spacing and justification matter. TeX uses a "boxes-and-glue" model, so during line building it's essentially putting stretchable glue between each word, and then when the line is built, it spreads the space evenly between all the pieces of glue to achieve justification. Glue has a natural size as well as minimum and maximum values; this along with hyphenation has to do with calculating the line "badness."

Third, you have to build paragraphs. This is what TeX calls "vertical mode." Between each line of text, it inserts vertical glue that works similarly to the horizontal glue mentioned above. Another glue is used between paragraphs. TeX worries about widow and orphan lines here; other calculations are made to prevent or allow them. This is like badness but I think it has another name.

Finally, you have to build pages. With TeX, this happens periodically as it notices it has enough material (or you force it by calling the right kind of eject). By default, this invokes a page building macro which you can customize. This part of TeX feels pretty 1970s and hackish, but essentially, there is a variable that contains a box with the supposed contents of the page, your job is to assign a box to a certain location and remove some of the contents of the input box. LaTeX's "float" mechanism is actually built on top of this mechanism, but as it's pretty procedural, you can certainly achieve other effects here.

So that's the birds-eye-view I acquired a few years ago when I did some in depth playing with TeX. First you build words, then lines, then paragraphs, then pages. There are different considerations at each stage of that but TeX attempts to subsume everything in that overall flow.

I think that process probably generalizes nicely for some things and is almost inadequate for others. I'm not sure how ConTeXt achieves the grid layout; I imagine they are relying on some complicated Lua code to pull it off, because TeX really doesn't care unless you make it.

Note that I use paper, because I am an old person. They're very different problems.

In print, you have to deal with page breaks and positioning of figures, but you know everyone will see the same thing. Doing this right requires human attention, but you can produce beautiful layouts, even with multiple columns and oddly-shaped text boxes.

On the web, everyone sees a single "page," but those pages are all different. Even simple two-column justified text is mostly garbage on the web, and forget trying to float figures close to their references. Web designers have mostly given up, resorting to either huge margins, or an ugly wide wall of giant-font text. Print-quality layout and typesetting on the web is very far away.

Looking at the FOSDEM'15 talk linked above, it appears to support this.

1.0.0 will be reached when, given an appropriate style sheet, SILE can perform unsupervised typesetting of an arbitrary USX file (XML-based Bible translation document) to publication standard.