It’s not a standard but still its an interesting software so I’ll post this here:

Joking aside, I love and hate it. Its paradigm is almost like using the C preprocessor to build a really awkward Turing-machine. TeX/LaTeX does a great job of what it was intended to do; it applies high quality typesetting rules to complex material and produces really good results. I love the output I can get with it and I will be eternally grateful that Donald Knuth decided to tackle this problem. And despite my complaints below, that gratitude is genuine. Being able to redefine something in a context-sensitive way, or to be able to rely on semantics to produce spacing appropriate to an operator vs a variable etc; these are beautiful things.

The problem is, at least once a day I’m left wishing I could just write a callable routine in a normal language with variables, types, arrays, loops and so on. You can implement all those things in TeX, but TeX doesn’t have a normal notion of strings, numbers or arrays, so it is rare that you can do a complicated thing in an efficient way, with readable code. So as a language, TeX frequently leads to cargo-cult programming. I’m not aware that you can invoke reflection after a page is output, to see what decisions on glue and breaks were made; but at the same time you can’t conditionally include something that is dependent on those decisions, since the decision will depend on what is included. This leads to some horrible conditionals combined with compiling twice, and the results are not always deterministic. Sometimes I find it’s quicker to work around things like that by writing an external program that modifies the resulting PDF output, but that seems perverse.

At the same time, there’s really nothing else out there that comes close to doing what LaTeX does, and if you have the patience, the quality of documents it can produce is essentially unbounded. The legacy of encodings, category codes, parameter limits, stack limits etc. just makes it very hard for package writers, and consumes a great deal of time for a lot of people. But maybe I am picky about things that a saner person would just live with.

A lot of very talented people have written a lot of very complex packages to save the user from these esoteric details, and as a result LaTeX is alive and well, and 99% of the time you can get the results you want, using off-the-shelf parts. The remaining 1% of the time, getting the result you want requires a level of expertise that is unreasonable to expect of users. (For comparison, I wrote an optimising C compiler and generally found it far easier to make that work as expected, than some of the things I’ve tried, and failed, to do properly in LaTeX. I now have a rule; if getting some weird alignment to work takes me more than an hour, I just fake it with a postscript file, an image, or write an external program to generate it longhand, in order to save my sanity.)

I think (and certainly hope) that LaTeX is here to stay, in much the same way that C and assembly language are. As time moves forward I think we’ll see more and more abstractions and fewer people dealing with the internals. But I will be forever grateful to the people who are experts in TeX, and who keep providing us with incredible packages.