I've found their comments, and the constructive responses to them, interesting because I am interested in Lisp but have never learned it.
What I don't find interesting is seeing troad's observations strongly challenged, as you are doing, as if somehow by sharing their observations, they now owe us detailed solutions as well.
It's not cool.
It's a backwards compat breakage, but there is a mechanism in the language which helps with such situations.
txr -C 237
or setting the TXR_COMPAT environment variable to 237 or lower will restore the destructive behavior of sort and shuffle. People who don't want to fix their code can deploy that way.I hate compatibility breaking changes, but it was a real thorn in my side how sort is destructive without any indication in its name, in a language where destructive versions of functions should be separately named.
Now the thing is that sort is not required to be destructive. It is allowed to be. Therefore code written to the language spec will not break if sort becomes pure. But we have to worry about breaking all programs though, not just programs written to the spec. (Unless we are GCC maintainers; different rules apply.) You can't just say, "Oh, programs that call sort on a vector and ignore the return value, expecting the vector to be sorted, are just nonportable junk; let them break."
What alternative would you suggest?
Giving paragraph-length names to the functions? Common Lisp already has some of the more descriptive function names around and they can be inconvenient if the function is used often. (By the way, there is something that helps to remember what those functions do: shorter names are correlated with more primitive and restrictive checks.)
Having a single function that performs different checks depending on a parameter? If you don’t know which function to use you don’t know which parameter to use either.
Removing the ability to perform different kinds of checks entirely?