Neal Richardson created ARROW-6338:
--------------------------------------
Summary: [R] Type function names don't match type names
Key: ARROW-6338
URL: https://issues.apache.org/jira/browse/ARROW-6338
Project: Apache Arrow
Issue Type: Improvement
Components: R
Reporter: Neal Richardson
Assignee: Neal Richardson
Fix For: 0.15.0
I noticed this while working on documentation for ARROW-5505, trying to show
how you could pass an explicit schema definition to make a table. For a few
types, the name of the type that gets printed (and comes from the C++ library)
doesn't match the name of the function you use to specify the type in a schema:
{code:r}
> tab <- to_arrow(data.frame(
+ a = 1:10,
+ b = as.numeric(1:10),
+ c = sample(c(TRUE, FALSE, NA), 10, replace = TRUE),
+ d = letters[1:10],
+ stringsAsFactors = FALSE
+ ))
> tab$schema
arrow::Schema
a: int32
b: double
c: bool
d: string
# Alright, let's make that schema
> schema(a = int32(), b = double(), c = bool(), d = string())
Error in bool() : could not find function "bool"
# Hmm, ok, so bool --> boolean()
> schema(a = int32(), b = double(), c = boolean(), d = string())
Error in string() : could not find function "string"
# string --> utf8()
> schema(a = int32(), b = double(), c = boolean(), d = utf8())
Error: type does not inherit from class arrow::DataType
# Wha?
> double()
numeric(0)
# Oh. double is a base R function.
> schema(a = int32(), b = float64(), c = boolean(), d = utf8())
arrow::Schema
a: int32
b: double
c: bool
d: string
{code}
If you believe this switch statement is correct, these three, along with float
and half_float, are the only mismatches:
[https://github.com/apache/arrow/blob/master/r/R/R6.R#L81-L109]
{code:r}
> schema(b = float64(), c = boolean(), d = utf8(), e = float32(), f = float16())
arrow::Schema
b: double
c: bool
d: string
e: float
f: halffloat
{code}
I can add aliases (i.e. another function that does the same thing) for bool,
string, float, and halffloat, and I can add some magic so that double() (and
even integer()) work inside the schema() function. But in looking into the C++
side to confirm where these alternate type names were coming from, I saw some
inconsistencies. For example,
https://github.com/apache/arrow/blob/master/cpp/src/arrow/type.h#L773-L788
suggests that the StringType should report its name as "utf8". But the ToString
method here
https://github.com/apache/arrow/blob/master/cpp/src/arrow/type.cc#L191 has it
report as "string". It's unclear why those should report differently.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
