[ajross]

Beware! That's the DSL trap.

It works here so well because it's limited to 20 lines and each macro does exactly what it needs to for the problem at hand.

Take that DSL and use it over a year to write a bunch of code to do normal things as your app grows into its problem domain and spills over into a few more, and it melts. New developers will show up to onboard to your and be like "WTF is this 'on()' thing I'm looking at all over the place, and why isn't it used over here?!". Some enterprising developer will introduce "map2()" to indirect based on keysym and not keycode, etc...

Domain Specific Languages are a mistake, almost every time they're used. And the only exceptions are the ones that grow into first class languages for well-defined problem areas (I'm thinking about things like VHDL or Mathematica here), and even there they tend not to be that much better than well-crafted domain-specific APIs in true programming languages (think numpy, pytorch, et. al.)

DSLs: Just say no.

[convolvatron]

there really isn't a fundamental difference between DSLs and libraries for the points that you brought up. where it really starts to get sketchy is when you do really funny things with the base syntax (looking at you lisp and rust). if not well thought out they can be fragile, confusing, and a real burden for new contributors.

I guess here's a question - do you consider regex libraries to be DSLs?

[ajross]

Interestingly I'd say a regular expression is absolutely not a DSL. It's sort of the opposite. A DSL is a tightly crafted interface for the configuration needed to support solutions to one somewhat unique problem. And the "trap" above is that the fact that the problem area is narrow means that only experts will really understand the DSL.

A regex is a tool implementing a solution (albeit a tightly crafted one) to an extremely broad set of problem areas. I mean, sure, in some sense it's a "DSL for string matching", but that doesn't really capture the essence of the DSL trap mentioned above. I mean, almost everyone needs string matching. So there's no trap: basically everyone knows regex syntax, even if they like to complain about it.

[silon42]

personally, I'd only consider a DSL to be good and useful if it can be implemented in different ways/languages, etc... It's not good if a DSL is language specific leaky abstraction.

Regex is a good example of a DSL.

[90s_dev]

Yeah exactly, this is why I stopped liking DSLs about 15 years ago, shortly after using Ruby extensively (probably not a coincidence) and converting to Clojure, where there was a large movement away from macros despite it being lisp. They're good in very isolated situations, and only when designed very carefully. This wm is quite possibly one of them; if you need more complexity than the macros here allow, and adding/changing macros only makes it worse, just use another wm.