LaTeX was not built for the web | Hacker News Reader

LaTeX was not built for the web

authorea.com

54 pointsapepe75 comments

LaTeX was built to typeset documents. You have pages in documents and pagination is of critical importance in LaTeX. In the web you don't have pages, it's evolved to be seamless and scrollable. Natuarally, LaTeX was not designed for such an environment.

That doesn't mean that LaTeX is worthless on the web. The gluing algorithms that Knuth used for creating sentences and paragraphs do (for the most part) work, even in documents without pages. In all honesty, the web could benefit greatly from LaTeX, pagination aside.

I'm glad that the author seems to have a practical perspective on this: >We think LaTeX is still the best programming language to tell a computer how to place text on a page. But the TeX project started pre-web, in 1978, and its scope and function are tightly linked to the printed page, not the webpage.

He goes on to give an example of constructing a table in LaTex which requires (or assumes by default) a suggested location for LaTeX to place the table in the page; something which is currently nonsensical in all but a few web environments. Like the author, I also find CSS appealing for a potential "LaTeX" on the web solution.

However, LaTeX was designed from the very first to be the next standard in type setting and is a turing complete language. CSS is still under debate for being turing complete, and it lacks nearly all of the features that are iconic of LaTeX, except for the ability to format math equations ( and even then...).

I'm not sure what the next standard in typesetting will be, but it will be designed to be device agnostic (with book existing as a supported format), it will target the web and similar digital media, it will be turing complete, and it will be created specifically to fulfill each of those goals, not to have them retroactively attached in a ham-fisted way.

"We think LaTeX is still the best programming language to tell a computer how to place text on a page."

No. The output might be beautiful, but the language is most certainly not. Awful to debug.

I consider a more modern language that compiles to LaTeX and leverage its rendering engine for paper docs (while giving you sufficient control over the output) and also gives you a nice web output more pragmatic and a nicer way to approach the problem.

Strongly agree. It's taboo to criticize TeX because of who wrote it, but let's be honest: from a usability perspective it's atrocious. Cryptic error messages, code that rivals assembler for readability ... hell even the directory layout makes no sense to me, and I've been using TeX daily for over ten years. I hold Knuth in high regard as a scientist, and there's no arguing that TeX is a technical masterpiece which revolutionized the field of mathematics--but this is not the typesetting language that's going to carry us through to the next century (I hope).

> Strongly agree. It's taboo to criticize TeX because of who wrote it,

I assume you're referring to Knuth, but it's important to distinguish between TeX and LaTeX (the latter was written by Leslie Lamport).

The distinction between TeX and LaTeX is actually relevant here.

> The distinction between TeX and LaTeX is actually relevant here.

I know very little about both. Could you elaborate on the relevance?

TeX is the low-level typesetting system and programming language. LaTeX is a set of macros on top of that that try to support a more semantic approach to document layout. If you see long command names that more or less spell out what they do, it's usually LaTeX. If you see short, cryptic abbreviated commands it's usually TeX.

And if you use either TeX or LaTeX instead of Lyx, you're a masochist.

> I consider a more modern language that compiles to LaTeX and leverage its rendering engine for paper docs (while giving you sufficient control over the output) and also gives you a nice web output more pragmatic and a nicer way to approach the problem.

I'm currently looking for exactly such a solution. Have you found any?

Sidenote: Ideal would be a language that comes with either Word or E-Pub conversion tools (in order to migrate existing word documents over). I say E-Pub because there is a relatively nice migration path from Word to there: Get on a Mac, open the Word in Pages, export to E-Pub. Can also be used to get a sane xhtml output, since it's just a zipped folder with xhtml and some images.

Emacs with org-mode works very well this way.

yep... we're all ears to alternatives. i often have to publish technical works academically and i really think the current crop of non-LaTeX solutions produce painful results. but i'm up against "we all use MSWord™ because we all use MSWord™" and there will be no getting overworked academics to invest the energy in anything that doesn't look like the tools with which they are already familiar. so... for gawdsake, find me something which isn't LaTeX which produces good equations and good page positioning without crashing and -looks- like MSWord to the non-technical computer user. i've had no luck. (and my cobbled together web nightmare rtf2text, git merge, meta LaTeX rube-goldberg device is leaking oil rapidly)

oh, and it has to play well with EndNote™ too. -sigh-

This post, right here, is one reason I'm glad I'm not in academia any more. I added Sweave to my rube-goldberg device though, and it leaked much, much more.

Have you tried Lyx[1]? It might not be as straightforward as MS Word, but if you care about the way the page looks, aren't afraid of using writing a little bit of LaTeX by hand, it works great.

[1]: http://www.lyx.org/

yes indeed. and it's certainly worthy at the postdoc level. but (quote from the LyX site): "Can I read and write Word files? Yes, but not trivially, and you should not expect to be able to collaborate with someone using Word to edit your paper. You can import Word files, and you can export Word files, but going 'roundtrip' is not workable." ...it's that "round-trip" that is currently vital.

what i really would love to see is some stable fusion of LyX and a web-collaborative-writing program like 'Gobby' (http://gobby.0x539.de/trac/) ... google-docs has been tried, but folks got scared between google-wave suddenly dying and that whole issue of corporate leaking to sinister third parties and corporate scraping

"you should not expect to be able to collaborate with someone using Word to edit your paper"

This, this is death to every alternative I look at.

> but i'm up against "we all use MSWord™ because we all use MSWord™" and there will be no getting overworked academics to invest the energy in anything that doesn't look like the tools with which they are already familiar.

MSWord has the additional problem that it can't be used together with my favourite Version Control System.

A system that takes ideas from http://docs.racket-lang.org/scribble/ would be most welcome.

"No. The output might be beautiful, but the language is most certainly not. Awful to debug."

Absolutely agree but, as far as I'm aware, this "more modern alternative" does not exist.

Lout is quite a bit saner but doesn't have much traction. It seems to be in "deep maintenance" mode now.

What about this ? http://dlmf.nist.gov/LaTeXML/

Having some XML output seems to make it easier to add CSS rules ?

Combines the complexity of LaTeX with the hideousness of XML. Great.

There's Patoline[^1], which is still in a somewhat early state, and its language is quite LaTeX-ish.

[^1]: http://patoline.com/index.html

> In the web you don't have pages

It can be done. For example Google docs supports headers, footers, and footnotes, but you are still free to decide whether to use the "print layout" or more webpage-like seamless layout for both editing and publishing ("publish to the web").

Be careful what you wish for re. Turing completeness. One of the things we're discovering is that, when implemented properly, basic modern CSS (i.e. 2.1) parallelizes beautifully across multiple threads and cores for large speedups, with only a few exceptions in wide use, like floats. But if CSS hadn't been so declarative, we'd be stuck in the computing model of the 90's era it was designed in when implementing it.

Turing completelness does not mean that the execution paths are specified.

Or, in a simpler way, do you know Haskell?

"CSS is still under debate for being turing complete"

So far as I'm aware, that's not the case. CSS is not Turing complete, in terms of what it can calculate in a single calculation. It is Turing complete if you string those calculations together, feeding the output of the last into the input of the next, which can only be done with some external source of events, but that is a larger system than "CSS". A UTM run for up to 1000 steps is not Turing complete.

Turing completeness is arguably a bug, not a feature, when the goal is producing something quickly.

Wouldn’t Turing-complete file formats be difficult to search because the output could have arbitrary syntax and the crawler would have to run the code. On the other hand, why not make use of faster machines.

Give me something that

  * is a simple and logically consistent as Markdown

  * has the ability to embed LaTeX formulas

  * has a functional #include for larger documents

  * can refer back to headings, equations and citations

  * can compile to PDF or webpage

  * compiles FAST

and I will give you a lot of money.

LaTeX is a horrible language, inconsistent and badly designed and unpleasant to look at. TeX is good at what it does but the whole system is horribly slow to compile documents. Documents will compile fine with some frontends but not with others. Markdown is not feature complete enough.

The scientific community is unfortunately stuck with LaTeX for the foreseeable future. Seems like every day someone asks me how to do something that should be very simple, which inevitably involves loading some obscure package.

Pandoc has almost all these features: http://johnmacfarlane.net/pandoc/

I think it's lacking the #include/#input feature, and I'm not sure whether it supports labels and internal references, but apparently it has some form of bibliography.

Pandoc doesn't have an "include" feature, but it's easy enough to add this with a preprocessor. See http://randomdeterminism.wordpress.com/2012/06/01/how-i-stop....

Pandoc does provide a way to link to section headings, but it doesn't yet have a generic system for autogenerating numbers and producing references to these, like LaTeX's \label{} and \ref{}. It does, however, have a system for creating running example lists: http://johnmacfarlane.net/pandoc/README.html#numbered-exampl.... This can be used for numbering equations and referring back to them, but it is considerably less flexible than LaTeX.

Pandoc has extensive support for automatic citations using CSL stylesheets: http://johnmacfarlane.net/pandoc/README.html#citations. BibTeX and BibLaTeX files can be used as the database, or YAML citations can be included in the markdown document itself.

LaTeX formulas can be embedded in markdown. They can even use macros defined in the document. Formulas can be converted to MathML or native Word equation objects, depending on the output format.

Pandoc can also be extended using "filters" that operate directly on the parsed AST. Here's an example of a filter that finds tikz diagrams and converts them to embedded images that can be displayed on the web: https://github.com/jgm/pandocfilters/blob/master/examples/ti....

hi fiddlosopher, we love Pandoc! We use it extensively and find it to work very very well with most TeX content. The only problematic part is tables (e.g. captions do not render in the latest pandoc version).

Pandoc does a lot of that (but I haven't used it).

At a minimum, it does markdown with embedded LaTeX and outputs pdf.

What do you find complicated/logically-inconsistent about LaTeX? The job it is supposed to do is nontrivial, so it seems appropriate that the tool is at least somewhat nontrivial, too, especially if you don’t like defaults.

It admittedly doesn’t compile to HTML and the compile-time can be on the order of minutes for very large documents, but are these really deal-breakers?

If you really don’t like LaTeX, have a look at org-mode, which seems to be able to fulfil most, if not all, of your requirements (not sure about compile times).

Org is really good about letting you unobtrusively embed LaTex into a document and can export to a bunch of formats.

I was doing an algorithms course with lots of mathematical formulas last year. Org was a godsend for completing the homework assignments.

reStructuredText (http://docutils.sourceforge.net/rst.html) seems have all those qualities. It's simple and IMHO more consistent than Markdown, math and include directives, cross-references and can be compiled to almost anything with tools like Pandoc (http://johnmacfarlane.net/pandoc/).

I believe that the Softcover typesetting system I'm working on meets all of the criteria you set out. It's currently in private beta; you can sign up for an invitation here:

http://softcover.io/

Send an email to the address in my profile and I'll fast-track it.

> horrible language, inconsistent and badly designed and unpleasant to look at.

I can see how you might not like the language itself, I don't either. But inconsistency is rather orthogonal quality. I think it is actually pretty consistent, which is respectable in its own right.

Asciidoctor: http://asciidoctor.org/docs/asciidoc-writers-guide/ . There are forthcoming plans to include a LaTeX backend (https://github.com/asciidoctor/asciidoctor/issues/681 ), but it can already output to HTML and Docbook, which can in turn output to other formats including PDF. There's also MathJax-based equation support for the HTML5 backend: https://github.com/asciidoctor/asciidoctor/issues/492 .

Obviously the big step is for it to be able to output LaTeX so that the equation capabilities and such will be directly exposed, but Asciidoc is pretty awesome already in my experience.

orgmode

I made my own markup language (post: http://breuleux.net/blog/quaint.html, docs: http://breuleux.net/quaint/, source: https://github.com/breuleux/quaint) and I think it could eventually fit the bill. I designed the syntax to be extremely regular, easy to extend and to stay out of your way. You can embed Python code to extend it (I don't get why you'd make a formatting language Turing complete instead of just embedding a real one). It generates documents in multiple passes and lets you specify what parts depend on what other parts, so back references are easy to implement. Unfortunately, I'm the sole developer and I've only developed it enough to be useful to my own purposes, so it only compiles to HTML and lacks a lot of features.

I'm not sure why you want the ability to embed LaTeX formulas. LaTeX formats math gloriously, but its syntax is not any less horrid for math than it is for anything else. I'd rather write "sum(i <- 1..10, frac(t_i, sigma^2))" than "\sum_{i=1}^{10} \frac{t_i}{\sigma^2}" but maybe that's just me.

Perhaps I'm just old and tired, but I loathe reading technical papers on a screen, whether it's my LCD monitor, my iPad, or my Kindle. One thing that irks me about the WWW specifications, for example, is that you cannot get a nicely formatted hard copy. I'm find with technical material being put on the web, but for the love of $DEITY please make a decently formatted PDF available too.

Agreed. That is one reason why most technical and scientific documents are still perused in PDF. In Authorea you can export each document to a journal format (Export). Soon we will have a printer-friendly style which is basically CSS.

From the article:

This table command instructs TeX to put the table in the page, here, where the table is declared (h) AND at the top of the page (t).

PrinceXML, an XML/HTML + CSS to PDF renderer, can do this using a "float:top" style: http://www.princexml.com/doc/9.0/properties/float/

Also, before they abandoned their bespoke browser engine, Opera released an experimental build that could render web pages as paged media (see http://dev.opera.com/articles/view/opera-reader-a-new-way-to... ). It, too, supports floating to top (as vendor-prefixed "-o-top").

Paged media is as alive as ever (it's just moving from dead trees to tablets), so I wouldn't count this stuff out just yet.

The main problem with LaTeX is that it does not have a good hyperlink mechanism. So, I am thinking about how to switch from LaTeX to HTML. I have a markup system based on Racket:

http://andreimikhailov.com/slides/bystroTeX/slides-manual/in...

The main application was supposed to be slides, to replace the Beamer. But I think it is good for generic math-oriented web publishing. It does not work on Windows, though. Comments/suggestions/testing are welcome!

Is there anything specifically wrong with the hyperref package? It seems to work rather decently for me.

I want to put a hyperlink to Eq. (4.4) of http://arxiv.org/abs/hep-th/9907164 What is the right format? How can it even in principle work? If I click on the link:

http://arxiv.org/abs/hep-th/9907164/Eq_4_4

what is supposed to happen?

Is http://arxiv.org/pdf/hep-th/9907164v2.pdf#page=30 a reasonable approximation? It works in Chrome and should work with the Adobe plugin, but doesn't work in Safari. I don't know if PDF has finer scale references than page.

I dont know what is Adobe plugin. I will investigate. However, just based on its name, I would probably recommend removing it. The combination of these two words, ``Adobe'' and ``plugin'', makes me feeling uneasy :) Which brings us to the second point: there is no good open-source PDF viewer. If you care about using open source, then go HTML.

Yeah, I don't have the Adobe PDF plugin either. But Firefox and Chrome use PDF.js which is open source AFAIK, and works with page references.

Also, if they submit a revised version, it will be a different page. Anyway, I don't think we will ever get a good PDF reader, because there is no demand. People read HTML.

In academia though, most scholars tend to stick to PDF for reading papers, see e.g. http://www.quora.com/Why-do-scientists-tend-to-prefer-PDF-do...

After a little more digging, I've found that pdf files can have named destinations, like eqrefs, so you don't have to use page numbers as references. See http://www.tug.org/pipermail/pdftex/2007-October/007383.html for how to make them in LaTeX.

This does require whoever generates the pdf to include the labels, but then so does html. It shouldn't be too hard to generate a reference to any named equation in a file.

What exactly is your problem with current PDF readers? PDF.js is pretty nice for browsers, Preview.app is standard on OS X, Linux has several that were fine and I know that Windows has many that are considered good alternatives to Acrobat Reader. Even phones and tablets do a good job, though e-readers do not.

You are right, and thank you for the link. Actually, let me use this to advertise my own fork of PdfViewer:

https://github.com/amkhlv/pdfviewer

I started that fork because I could not find a viewer which would keep vertical position, and jump back after following an internal link, and have bookmarks with charhints.

There is a conceptual problem, however: as I said, I feel that PDF is getting deprecated because of insufficient demand. If this is true, then it does not make much sense to invest effort into TeX + PDF. Maybe I am wrong. Sometimes I think that I am simply allergic to TeX :)

To me, that looks more like a limitation in PDF rather than the tool that is used to create the PDF – that is, hyperref allows you to create arbitrary links to be handled by your system, but the problem is that PDF does not work all that well when you want to reference a link within a given PDF.

If there was a reliable way to advise a PDF reader to ‘open document X and jump to reference Y’ and to specify both X and Y, then I am sure you would have no problem using hyperref to create a link to X#Y.

"I have a markup system based on Racket"

Interesting. I just discovered the excellent online book "Practical Typography"[0], which the author prepared using a system he created that he calls "Pollen," and which he wrote in Racket.

[0]http://practicaltypography.com/

Thanks for the link, I will have a look! Although, in terms of typography you cannot beat LaTeX. That's what LaTeX is good at. But the thing is, I don't care so much about the typographic quality. All I need is a quick way of inserting and hyperlinking math formulas in HTML pages.

In terms of pure mathematical typesetting, it's great, and, as they mention, Mathjax does a fantastic job of handling equations in webpages. Yes, TeX takes some learning, but it's a let less verbose than MathML for the same equation, and being distinct isn't necessarily an issue: there is precedent within HTML, such as CSS and Javascript code being embedded within pages. A browser with native support for (a subset of) TeX would allow for a lot of technical papers to be presented in a fully searchable, device agnostic manner, and would avoid the additional learning curve or conversion to MathML.

For layout, though, HTML with CSS is the way to go if your prime target is web.

> What does the future hold for academic writing? We like to think that a few years from now we will format our research papers with the web version in mind, rather than the printed PDF.

I've been thinking about this for almost a year. I believe as well that webview-first is the future of technical publications. The problem is how to either 1) get traditional publishers to adopt newer and better technology (a lot of publishers are still using systems seemingly from 90s that doesn't even support features that have been stablized in TexLive for years); or 2) build new publishers that gain enough reputation fast enough, so that the academia would consider them as good communication channels.

Personally I like the second way better. It's more convenient (and easier to think out of box) to start from scratch than to change an existing system (by system I mean organizations, publishers, rather than a computer system).

Academia is somehow like a trust chain. People tend to follow reputable researchers/professors. If a platform can get most reputable researchers, it can be adopted soon.

What about 3) compile to LaTeX?

I really hope we don't optimise for a webview-first future. I personally find reading large pieces of text on a screen uncomfortable at best, and painful at worse. I want to be able to print out any scientific or academic documents, and at the moment LaTeX and PDF do that for me the best.

And HTML was not built for GUIs. Does it stop the ongoing madness? No.

These guys are well worth a look https://www.writelatex.com if you're looking for web-based, collaborative LaTeX tools.

Along with mendeley and peerj there's the makings of little cluster of companies trying to innovate in the academic publishing space here in London.

Thanks, yes! We know and are in touch with all these guys (writelatex, mendeley, and peerj). Great stuff coming out of London in this space. Digital Science and Figshare are also there. We are very excited that there's a movement to change/improve academic publishing.

One tricky bit is that I think a good solution medium-term will need to handle both media. Right now, a typical academic paper's print format transfers relatively poorly to the web. But the reading experience you get when you print out a web-formatted document to paper is also typically pretty poor, especially with things like figure layout.

It's possible e-readers will someday obsolete print entirely, but I personally still find it difficult to read longer-form stuff on a screen, so I'd like to see better options for print layout of web documents. Technologically these are possible, e.g. PrinceXML shows quite a bit of print-layout stuff you can do with the web-technology stack (though it's unfortunately proprietary), but the bits and pieces don't quite plug together well yet.

Correct, tangle was built for the WEB. Well TeX was built written in WEB.

(sorry for the bad knuth toolchain jokes)

There is another problem.

Some time ago, I thought about integrating LaTeX into user-facing software, not to render an entire web page but to generate a few formulas and images containing formulas in real-time or close to real-time. It breaks my heart to say that it seems impossible to do so, because LaTex is soooo slow.

The codebase is written in an obscure language (that Knuth created for the purpose?) and would have to be converted to something optimizable such as C in order to become faster.

Since I won't be re-implementing LaTex soon, I'll stick with MathJax for simple formulas and must look for something else if I want to create a complicated formula-containing image.

Web2c is typically used these days. The issue is not as simple as "It needs to be written in C".

Well, I gave up when I saw that it wasn't written in anything I could work with, so this is news to me.

Also, having had some bad experiences with to-C translated code (looking at you, Matlab), I'd say that knowing that Web2C exists doesn't cause me great optimism.

There is recent work on a TeX engine with a Lua interface:

http://tex.stackexchange.com/questions/36/differences-betwee...

But still LaTeX sits on top of that, so it probably doesn't move you along very much.

TeX is written in WEB, a "literate programming" language designed by Knuth, which a pre-processor turns into Pascal. There should be nothing inherently slow about Pascal. In practice, it's compiled using "web2c" which does generate C code.

http://en.wikipedia.org/wiki/WEB

A lot of the time overheads with LaTeX are in the initial starting up. If you look into, for example, the preview environment, you can look at processing multiple snippets of LaTeX much quicker. Of course, such batching doesn't help achieve the goal of real-time LaTeX.

If you're willing to eschew the luxuries that PDFTeX provides (microtype and pdf output) and drop back to DVI, it is possible to interpret the DVI yourself in real time using the IPC hack (which allows dumping of dvi on each page flush). The file format is specified in the dvitype manual[1].

Unfortunately, once you've processed the DVI output, you'll discover that many modern TeX fonts do not render, because they are virtual fonts and you will need to implement these. You'll also have to process the TeX font metric files.

Once you've done all this, you can easily get TeX to produce in excess of 1000 snippets per second. The rest is up to your rendering backend, but you'd have to be doing a lot wrong to end up with less than 100 per second.

[1] http://texdoc.net/texmf-dist/doc/generic/knuth/texware/dvity...

I think the guys at http://substance.io are trying to solve this problem. You should take a look !

thanks! I did come across substance.io a few months ago. Seems like they made a lot of progress. Looks beautiful. It's time to get in touch with them again.

On a related note, has there been any recent word on LaTeX3?

Should I learn to finally do my resume in LaTeX?

I did, and would recommend it to others. The final product is just so much nicer looking than I feel the reader subconsciously finds one to be a more appealing candidate.

You almost certainly want to use a better font than Computer Modern, but that's easy.

Reply on news.ycombinator.com