filter() forbid matrices

[romainfrancois]

closes #5973

original code from #5972 gives:

library(dplyr, warn.conflicts = FALSE)

tibble(
  y = tibble(
    b = c(1, 2),
    a = c(NA, 1),
  )
) %>% 
  filter(is.na(y))
#> Error: Problem with `filter()` input `..1`.
#> ℹ Input `..1` is `is.na(y)`.
#> x Input `..1` must be a logical vector, not a logical[,2].

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

[romainfrancois]

Relatedly, is it time to be stricter about data frame results in:

dplyr/src/filter.cpp

Line 116 in 8e1ca14

} else if(Rf_inherits(res, "data.frame")) {

and just allow logical vectors full stop. Allowing data frame results made sense before we had if_any() and if_all() but not so much now. We could have the error message guide towards if_any() / if_all() for some time.

} 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);
      }
    }

    UNPROTECT(2);
  }

[hadley]

Change looks good and I think your proposed warning makes sense.