vec_unique() fails on lists of objects containing non-vectors

[mikmart]

In the current dev version, vec_unique() doesn’t work with lists of objects that are not vectors, or that contain non-vector elements in a recursive structure:

m <- glm(mpg ~ wt, data = mtcars)
vctrs::vec_unique(list(m, m))
#> Error: `x` must be a vector, not a `glm/lm` object

o <- list(x = expression(x))
vctrs::vec_unique(list(o, o))
#> Error: `x` must be a vector, not an expression vector

glm objects are an example of both, and can frequently appear in list columns in data frames. As vec_unique() gets more use behind the scenes in packages such as tidyr, this threatens to break code that works with list columns: e.g. tidyverse/tidyr#735 seems to be a symptom of this behaviour.

^{Created on 2019-11-04 by the reprex package (v0.3.0)}

[DavisVaughan]

Part of this will be fixed by #634, but I'm not quite happy with that solution yet.

While working on that, I also noticed that vec_equal() (and through this, the dictionary functions which call equal_scalar()) doesn't work correctly with lists containing scalar objects.

vctrs::vec_equal(list(lm(mpg~cyl, data = mtcars)), list(lm(mpg~cyl, data = mtcars)))
#> Error: `x` must be a vector, not a `lm` object

^{Created on 2019-11-04 by the reprex package (v0.3.0)}

This comes from equal_object() calling vec_size().

vctrs/src/equal.c

Line 241 in b4f7be4

R_len_t n = vec_size(x);

This is a bit tough because we are using this as the fallthrough for most vector types, and want it to work for data frames too, which is why we use vec_size(). I think we might need to separate out the VECSXP case and use Rf_length() when it is a true list, and vec_size() when it is a data frame.

[lionel-]

I think this lm object should be checked with an obj_ function, which should use base R primitives to descend through all objects, including data frames.

[DavisVaughan]

Ah I see. A data frame would not have to be checked row wise when it is inside a list like vec_equal(list(mtcars), list(mtcars)) because we just need to return a single TRUE/FALSE. That makes a lot of sense.

I think everything is already set up correctly. equal_object() is being called on the two lm objects (which I think is the obj_ function you want). I think that equal_object() just needs to be rewritten to use the base R primitives.

[hadley]

This is another a case where we should be using inherits(x, "list") rather than typeof(x) == "list".

[lionel-]

I don't think so because there is no dispatch involved in these equality checks / hashing.

[DavisVaughan]

With the dev version:

m <- glm(mpg ~ wt, data = mtcars)
vctrs::vec_unique(list(m, m))
#> [[1]]
#> 
#> Call:  glm(formula = mpg ~ wt, data = mtcars)
#> 
#> Coefficients:
#> (Intercept)           wt  
#>      37.285       -5.344  
#> 
#> Degrees of Freedom: 31 Total (i.e. Null);  30 Residual
#> Null Deviance:       1126 
#> Residual Deviance: 278.3     AIC: 166

o <- list(x = expression(x))
vctrs::vec_unique(list(o, o))
#> [[1]]
#> [[1]]$x
#> expression(x)

^{Created on 2019-11-13 by the reprex package (v0.3.0.9000)}