> ERROR: DomainError with -1.0:
> sqrt was called with a negative real argument but will only return a complex result if called with a complex argument. Try sqrt(Complex(x)).
This is the sort of thing that drives me absolutely nuts. In no possible world that I can imagine is generating an error a more useful result than returning a complex value. I could forgive it if the system didn't know about complex numbers, but in this case it is clear that the system is perfectly capable of producing a correct and useful result, but it just refuses to do so in service of some stupid ideology. I have zero tolerance for this sort of nonsense.
I'm not sure how familiar you are with Julia, but it is very spiritually aligned with lisp. The reason for the DomainError is "type stability" and it's not stupid ideology unless you consider high-performing numerical code stupid ideology.
I know very little about Julia. It is on my list of Things To Look Into Some Day.
> type stability [is] not stupid ideology unless you consider high-performing numerical code stupid ideology.
"Stupid ideology" may have been putting it a bit strongly, and there are times when I want numerical code to be performant. But I don't always want that, and I don't want a language that makes me pay for it even when I don't want it. Almost always I prefer high-fidelity [1] over speed. But even when I want speed, I always want to start with high-fidelity, get that working, and then make it fast if I need to. 99% of the time I don't need to because 99% of my code turns out to be I/O bound or memory-latency bound. It's extremely rare for my code to be compute-bound, and even in those cases I can almost always just find a C library somewhere that someone else has written that I can call through an FFI. So for me, a programming language that forces me to pay for run-time performance whether I want it or not has negative value.
But honestly, I can't actually think of a single instance in my entire 40 year career when the run-time performance of my CL code was a limiting factor. The limiting factor to my productivity has almost always been my typing speed and the speed of my internet connection.
---
[1] By "high fidelity" I mean how well the semantics of the language reflects the domain model, which, in the case of numerical code, is mathematics. Fixnums and floats are a fast but low-fidelity model because they don't actually behave like the integers and reals that they purport to model. Bignums and rationals are a high(er)-fidelity model. A really high fidelity model would let me do something like (* (sqrt 2) (sqrt 2)) and get back exactly 2 by having some sort of exact representation of algebraic numbers. I don't know of any programming language that provides that natively, but some languages, like CL and Python, let me extend the language to add that kind of capability myself. For me, that kind of extensibility is table stakes.
This is the sort of thing that drives me absolutely nuts. In no possible world that I can imagine is generating an error a more useful result than returning a complex value. I could forgive it if the system didn't know about complex numbers, but in this case it is clear that the system is perfectly capable of producing a correct and useful result, but it just refuses to do so in service of some stupid ideology. I have zero tolerance for this sort of nonsense.