SQL is an absolutely amazing domain specific language, but people keep building ORMs & other SQL abstractors. I get building one as a learning or resume project. I don't see how it's useful in most scenarios. Like, there's some difficulty breaking up the queries in a logical way (which leads to duplication), so the solution is another library that you need to write or install and learn how to use?

Even if it's easy to use, I can't imagine it'd take less time to read the documentation than to write some SQL. And then there's another dependency; Another thing to check if there's performance issues, another attack vector.

The issue with plain SQL is simply that you can not compose queries at runtime.

For example let user decide which columns to select, in which order to fetch the rows, which table to query etc. Or not even let the user decide but decide based on some config file or the system state.

You end up with a bunch of string manipulation that is fragile and does not compose well.

What solution is there except from simulating SQL via your languages nestable data structures?

Still fundamentally manipulating SQL text (which is a feature as I don't want to learn a full DSL), but it handles wrangling embedded placeholders while you're composing stuff and some other common compositional tasks. It's worked well for me anyway but I'm under no illusions it'd be right for everyone.

Not an original concept regardless; my original version of this was in Node: https://github.com/bdowning/sql-assassin, but a few years after I wrote that (and mostly didn't use it) I found https://github.com/gajus/slonik which was very similar and much more fleshed-out; I rolled _some_ of its concepts and patterns into sql-athame.