vec_ptype2() finds the common type for a pair of vectors, or dies trying.
It forms the foundation of the vctrs type system, along with vec_cast().
This powers type coercion but should not usually be called directly;
instead call vec_ptype_common().
# S3 method for logical vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for integer vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for double vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for complex vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for character vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for raw vec_ptype2(x, y, ..., x_arg = "", y_arg = "") # S3 method for list vec_ptype2(x, y, ..., x_arg = "", y_arg = "") vec_ptype2(x, y, ..., x_arg = "", y_arg = "") vec_default_ptype2(x, y, ..., x_arg = "", y_arg = "")
Arguments
| x, y | Vector types. |
|---|---|
| ... | These dots are for future extensions and must be empty. |
| x_arg, y_arg | Argument names for |
Coercion rules
vctrs thinks of the vector types as forming a partially ordered set, or poset. Then finding the common type from a set of types is a matter of finding the least-upper-bound; if the least-upper-bound does not exist, there is no common type. This is the case for many pairs of 1d vectors.
The poset of the most important base vectors is shown below:
(where datetime stands for POSIXt, and date for Date)
S3 dispatch
vec_ptype2() dispatches on both arguments. This is implemented by having
methods of vec_ptype2(), e.g. vec_ptype2.integer() also be S3 generics,
which call e.g. vec_ptype2.integer.double(). vec_ptype2.x.y() must
return the same value as vec_ptype2.y.x(); this is not enforced
for reasons of efficiency, but should be tested.
vec_ptype2() are not inherited, classes must explicitly implement
the methods. There are two reasons for this:
The coercion hierarchy is often different from a class hierarchy. For instance the richer type between a tibble and a data frame is a tibble. Grouped data frames inherit from tibble, and so would by default inherit from tibble's
vec_ptype2()method if inheritance was allowed. The method would then indicate that the richer type between a grouped data frame and a data frame is a tibble, which is wrong.vec_ptype2()should be symmetric, i.e. it should return the same type no matter the order of the inputs. With inheritance, this isn't the case when two classes have a common parent class. For instance the richer type between a tsibble and a tibble is tsibble. Similarly, grouped data frames are a richer type than tibble. Both of these classess have tibble as common parent class. With inheritance,vec_ptype2(tsibble, gdf)would return a tsibble via the tsibble-tibble method.vec_ptype2(gdf, tsibble)would return a grouped data frame via the gdf-tibble method.
Because of the way double dispatch is implemented, NextMethod()
does not work inside vec_ptype2() methods.
See vignette("s3-vector") for full details.