feat: New sort argument to rel_order()

[toppyy]

Related to tidyverse/duckplyr#92

Changes:

1. rapi_rel_order now accepts a vector of boolean values that describe whether each expression should be sorted in ascending order
2. rel_order checks that the number of expressions and boolean values match or defaults to TRUE if a null-value is passed
3. tests for the relation api test the rel_order(), not rapi_rel_order(). The tests break without this change as the default value is handled in rel_order.

I will mark this as a draft due to these notes:

• This does not work for nested expressions. For example, assume someone writes arrange(someotherfunction(desc(column))) in duckplyr. I don't know if this is an issue as the interpretation of such an expression in terms of sort order is a bit ambiguous(?).
  • note: this problem exists also without changes proposed here?
• I was unsure what function should to check that length(orders) == length(ascending). It is now rel_order(), rapi_rel_order(). As a result, using rapi_rel_order directly might result indexing out of array bounds. Perhaps the check should be in rapi_rel_order as the vector is indexed there?
• Related to the previous note. Should we aim to test rel_order() or rapi_rel_order? There are examples of both for other relation expressions

[krlmlr]

Thanks, looks great! Triggering checks.

[krlmlr]

@Antonov548: What's the idiomatic C++ way to iterate over two "parallel" vectors of the same length, orders and ascending ?

[Antonov548]

As I know there is no specific function in STL to do this. Usually it's range based for loop with : for single iterable objects and index based loop for multiple vectors.

[krlmlr]

Can we use zip_view() or similar?

https://codereview.stackexchange.com/q/283638/10691

[Antonov548]

ranges require C++20 and higher. Not sure that it's worth of changing to C++20 if it's even possible.

[krlmlr]

Testing rel_order() is fine, rapi_rel_order() is low-level and will never be exported. The arrange(f(desc(x))) issue is not much of a problem -- duckplyr will only translate desc() if it's the outermost expression and fall back to dplyr otherwise.

[krlmlr]

Thanks!