`filter()` silently accepting matrices

[romainfrancois]

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(x = 1:2, y = 3:4)

this fails, as it should

filter(df, c(TRUE, TRUE, FALSE, FALSE))
#> Error: Problem with `filter()` input `..1`.
#> ℹ Input `..1` is `c(TRUE, TRUE, FALSE, FALSE)`.
#> x Input `..1` must be of size 2 or 1, not size 4.

but this does not fail, and worse it silently uses the first column of the matrix:

filter(df, matrix(c(TRUE, TRUE, TRUE, FALSE), nrow = 2))
#>   x y
#> 1 1 3
#> 2 2 4

however if we convert the matrix to a data frame, then
we get an auto splicing behavior

filter(df, as.data.frame(matrix(c(TRUE, TRUE, TRUE, FALSE), nrow = 2)))
#>   x y
#> 1 1 3

^{Created on 2021-08-06 by the reprex package (v2.0.0)}

[romainfrancois]

This is related to #5972

[romainfrancois]

we already have plans for disabling, or at least discouraging data frame results inside filter(), which were only useful before we had if_any() and if_all():

    } else if(Rf_inherits(res, "data.frame")) {
      // disable the warnings entirely for now, so that we can first release
      // if_any() and if_all(), will start warning later
      bool warn = false;

      if (first && warn) {
        SEXP expr = rlang::quo_get_expr(VECTOR_ELT(quos, i));
        bool across = TYPEOF(expr) == LANGSXP && CAR(expr) == dplyr::symbols::across;
        if (across) {
          Rf_warningcall(R_NilValue, "Using `across()` in `filter()` is deprecated, use `if_any()` or `if_all()`");
        } else {
          Rf_warningcall(R_NilValue, "data frame results in `filter()` are deprecated, use `if_any()` or `if_all()`");
        }
      }

      const SEXP* p_res = VECTOR_PTR_RO(res);
      R_xlen_t ncol = XLENGTH(res);
      for (R_xlen_t j = 0; j < ncol; j++) {
        reduce_lgl_and(reduced, p_res[j], n);
      }
    }

Not sure what to do with matrices ? Forbid them ? Treat them as data frames ?

This feels like a reach for natural use of filter() but the matrix might come from some other function, as in #5972 where we get a matrix by applying is.na() to a data frame:

library(dplyr, warn.conflicts = FALSE)

tibble(
  y = tibble(
    b = c(1, 2),
    a = c(NA, 1),
  )
) %>% 
  filter(is.na(y))
#> # A tibble: 0 x 1
#> # … with 1 variable: y <tibble[,2]>

^{Created on 2021-08-06 by the reprex package (v2.0.0)}

[romainfrancois]

relatedly, slice() just lets matrices pass:

library(dplyr, warn.conflicts = FALSE)

df <- data.frame(x = 1:2)
slice(df, matrix(c(1, 2, 1, 2), nrow = 2))
#>   x
#> 1 1
#> 2 2
#> 3 1
#> 4 2

^{Created on 2021-08-06 by the reprex package (v2.0.0)}