[R] Cryptic error when creating Arrow array from POSIXct with invalid time zones

[Wainberg]

Describe the bug, including details regarding any error messages, version, and platform.

POSIXct is tricky because you can construct one with an invalid timezone and it won't give an error until you print it:

> a = as.POSIXct(1, NA)
> print(a)
Error in as.POSIXlt.POSIXct(x, tz) : invalid 'tz' value

But at least the error itself is easily understandable. When you try to create an Arrow array from such a POSIXct, you instead get a cryptic error:

> arrow::Array$create(as.POSIXct(1, 1))
Error in vec_to_Array(x, type) :
  STRING_ELT() can only be applied to a 'character vector', not a 'double'
> arrow::Array$create(as.POSIXct(1, NA))
Error in vec_to_Array(x, type) :
  STRING_ELT() can only be applied to a 'character vector', not a 'logical'

Component(s)

R

[amoeba]

Thanks for the report @Wainberg. I do think it's ideal that arrow could give a better error even if R's doesn't. I think we could safely make this section more restrictive, possibly throwing an error or converting non-character inputs to character first:

arrow/r/src/type_infer.cpp

Lines 88 to 94 in 7288bd5

	auto tzone_sexp = Rf_getAttrib(x, symbols::tzone);
	if (Rf_isNull(tzone_sexp)) {
	auto systzone_sexp = cpp11::package("base")["Sys.timezone"];
	return timestamp(TimeUnit::MICRO, CHAR(STRING_ELT(systzone_sexp(), 0)));
	} else {
	return timestamp(TimeUnit::MICRO, CHAR(STRING_ELT(tzone_sexp, 0)));
	}

I'll get some eyes on this and plan to put a fix up if we think that's appropraite. Thanks again.