I think some kind of working group might be in order here. There seems to be some real interest in making something usable, and thanks to Simon we have not just ideas but code as well.

Joseph

Simon Dales is working on some ideas using Doxygen, and has made some progress on an initial testing system. Might be worth talking directly with him.

Joseph

It is convenient to have the source of LaTeX packages that are on CTAN. It is especially useful that all LaTeX packages come as source, and are not distributed in compiled form. Because of that I can just start to discuss how to make the index from plain text.

Any pattern matcher should be able to spot \def, \let and \newcommand. A good pattern matching tool especially if it has dynamically created patterns, not evaluated until actually required, should be able to track catcode changes and so identify @commands. In the discussion at the meeting, it was noted that end-users of (La)TeX should not know about @commands and not use them directly, but use the package interface to invoke them indirectly in the intended manner.

I think it would be a challenge to accurately index definitions created from \csname and \expandafter commands. Not necessarily impossible, but difficult.

It would be an immense job to index all of the packages on CTAN, and deal with all the name clashes. Also, CTAN is updated frequently, so the indexing tool would have to run on CTAN frequently. That’s a potential load on the CTAN server.

But as far as I know, no-one actually loads all the packages on top of LaTeX. And if a user seeks help, they need help with the particular combination of packages they actually loaded on the LaTeX run that went wrong.

So a specific index for that user’s run of (LaTeX) would be more help to them and involve less indexing for us.

Rather than having the problem of updating a central index continuously, instead I propose we build a tool that reads the user’s preamble, indexes the files loaded, in the order they are loaded and offers that much smaller index to the user. It might rely on the users’ package files being virtually or actually on the same workstation as the documentation tool. This is almost always the case.

It’s a much smaller task than cataloguing CTAN, it would be accurate and uptodate for the user. Indeed, this tool will give users helpful results even if they use outdated packages or packages not available from CTAN e.g. style packages from LMS, Elsevier and other publishers. And I believe with a suitable choice of language, the index would be created very rapidly in the majority of cases.

As programming languages that have suitable pattern matching, I know of Snobol, Icon, Ada, and another- can’t remember which just now.
I believe all are freeware. Snobol and Icon are true open source. Snobol is very fast, and only about 100k of runtime. The associated data (such as a precompiled LaTeX kernel index?) could be larger than that? So the footprint is tiny too. And there are versions for PC & Linux. There may still be a Mac version. So it covers most of the workstations that run TeX.

Given that powerful pattern matching is only a “nice to have”, what other languages should we consider?

What languages do most TeX (at least the windows and linux users) nearly always have (apart from TeX Macro language, please )

What about Perl, Ruby, Lua, php ? Others?

What could we get users to add to their worksatation that would be small & effective (like Snobol)? Would any of these tools run (with adequate security) on a server running Mathtran or another “I imageset your TeX formulae” tool?

Initially we could prototype this tool, trial it with windows and linux users, and if successful, make it more accessible, faster by choosing a different language.

I’d be willing to prototype in Snobol or Icon/Unicon.
Any other offers? Anyone want to help, e.g draft a spec? Or just tell me what command defining words exist in Plain TeX, LaTeX and ConTeXt ?

I’d not noticed that LaTeXRender retains the TeX source: thanks for that.

The point about MathML (or OpenMath) content is that it is defined. The problem with (La)TeX for that type of thing is the definitions alter depending on what is loaded.

Joseph

I understand that development of incorporating mathML into Firefox has stopped though that may be more rumour than fact.

By the way, LatexRender and similar do retain the content – the underlying LaTeX code is in the alt and title tags which can be read by a screen reader and show when images are turned off. For example, hover over the graph at Plotting Graphs and click on it to open a popup with the code.

Thanks very much for the comment: glad I’m interesting. I’ve not come across freenode, but it sounds like the approach is pretty sensible. I’ll take a look.

Joseph

Documentation was a strong topic at the meeting. The comparison to other programming systems does not necessarily flatter LaTeX: if you look at things like Python, you can get access to the same docs online or in a printed form. But the problem is that everyone does their own thing for LaTeX packages. It would be nice to imagine a better ltxdoc (or something) that made it easier to produce multiple output formats. You could then imagine having web pages of the documentation to complement the existing PDFs. You could then, I think, imagine having something like:

http://latexdocs.org/

To open the documentation directly.

Joseph ]]> http://www.texdev.net/2009/11/16/answering-questions/comment-page-1/#comment-1589 Joseph Wright Tue, 17 Nov 2009 08:16:38 +0000 http://www.texdev.net/?p=540#comment-1589 Hello Steve, Thanks for the information. One of the discussion points was that things like MathML/OpenMath have the advantage that the "content" remains available and scalable in the browser. The thing with the various image-based solutions is that the content is not there for the PC (which includes problems for partially-sighted users). As to TUG, I'm sure that supporting projects would be possible, but the usual problem is finding people to do things. There are always more good ideas than people with time to do them, I think. Joseph Hello Steve,

Thanks for the information. One of the discussion points was that things like MathML/OpenMath have the advantage that the “content” remains available and scalable in the browser. The thing with the various image-based solutions is that the content is not there for the PC (which includes problems for partially-sighted users).

As to TUG, I’m sure that supporting projects would be possible, but the usual problem is finding people to do things. There are always more good ideas than people with time to do them, I think.

Joseph

Next I want to mention that on the irc channel #latex on freenode we usually try to help people help themself:

If they ask how to write an integral, we orcourse tell them it is \int but also point them to lshort.pdf (or the tug FAQ) since we then belive that they need an introductory doc.

If they ask a more specific question, say on howto have multiple lines in headers, we point them to a specific package (and its doc).

After this, if they still do not manage, there is a bot (written by Troy Henderson) that runs LaTeX (and MetaPost) in an interactive way that is very helpful. There is also a pastebin (on http://pastebin.tlhiv.org ) where one can give larger snippets and see how they compile.

Most users get help very fast this way, and are happy with it. Some do not like to be pointed at some doc, but rather just want the answers.

If noone knows the answer c.t.t. is a very good resource for finding answers.