Negative numbers are not syntactic literals

[moodymudskipper]

Since they use -.

We dismiss complex numbers with a real part, presumably because of the +, so it would make sense to return FALSE for negative integer, double, and imaginary numbers.

x <- quote((-1))
y <- substitute((X), list(X = -1))
x
#> (-1)
y
#> (-1)
typeof(x[[2]])
#> [1] "language"
typeof(y[[2]])
#> [1] "double"
rlang::is_expression(x) # correct
#> [1] TRUE
rlang::is_expression(y) # incorrect
#> [1] TRUE

Related: rlang::is_syntactic_literal(NA + 1i) is currently TRUE, to be a syntactic literal a complex number must have a real part of 0.

[lionel-]

I'm tempted to deprecate this functionality as we don't use it for anything important and it sounds like a pain to keep maintaining it.

Would you have any interest in taking over these functions and maintaining them in another package?

Edit: let's keep this in rlang for now

[moodymudskipper]

In constructive I have already is_expression2() and is_syntactic_litteral2(), both unexported, which are mostly similar, though I don't think I based the former on yours directly. I need those so i can generate the code to construct language objects faithfully. I can remove the "2", make sure they pass rlang's own tests and export them there, but don't you need those in rlang for expr_deparse() and friends ?

[jonthegeek]

For rlang::is_syntactic_literal(NA + 1i), isn't that equivalent to is_syntactic_literal(as.integer("fish")) (also TRUE)? We aren't capturing the call, so there are a lot of calls you could make that aren't syntactic literals themselves but result in a syntactic literal. is_syntactic_literal(quote(NA + 1i)) returns FALSE (at least in my branch, see incoming PR) as it should.

[moodymudskipper]

@jonthegeek NA + 1i prints as NA but is truly complex(real = NA, imaginary = 1), not the same as NA_complex_, which is complex(real = NA, imaginary = NA). The latter is a syntactic literal, but the former isn't, and is.na() return TRUE for both.

On the other hand as.integer("fish") is identical to NA_integer_ so it's a syntactic literal.

[jonthegeek]

@jonthegeek NA + 1i prints as NA but is truly complex(real = NA, imaginary = 1), not the same as NA_complex_, which is complex(real = NA, imaginary = NA). The latter is a syntactic literal, but the former isn't, and is.na() return TRUE for both.

@moodymudskipper I recommend splitting that out into a new issue. I can probably knock out a fix after confirming that it's identifiable inside the function. Fortunately I think the current form of the function will be relatively easy to update to deal with that case.

[jonthegeek]

@moodymudskipper never mind, I pulled this out into #1828. I'll try to knock it out real quick!

[jonthegeek]

NA + 1i case fixed in #1829.

[moodymudskipper]

I'm sorry that I dragged you into R version hell.