bind_rows() rejects POSIXlt columns

[krlmlr]

instead of segfaulting.

Fixes #1789.

Also fixes another instance with the same pattern, but I'm not sure if this is strictly necessary.

NEWS line:

* `bind_rows()` cleanly aborts if one of the data frames contains a `POSIXlt` column
  (#1789, #1875, @krlmlr).

[codecov-io]

Current coverage is 20.00%

Merging #1875 into master will decrease coverage by <.01%

@@             master      #1875   diff @@
==========================================
  Files           188        188          
  Lines          7422       7425     +3   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
  Hits           1485       1485          
- Misses         5937       5940     +3   
  Partials          0          0          

Powered by Codecov. Last updated by a1c4faf...d978987

[hadley]

Does the logic in order_visitor() look right to you?

                    case VECSXP:
                    {
                        if( Rf_inherits( vec, "data.frame" ) ){
                            return new OrderVisitorDataFrame<true>( vec ) ;
                        }
                        break ;
                    }

I think that will fall through to the return 0. I think the break should be removed, since we don't have a general list orderer (and POSIXlt doesn't need to be a special case)

[krlmlr]

No, the logic looks odd. I'll submit a pull.

Fall-through in switch() blocks (removing the break) is an anti-pattern and may even induce a compiler warning.

[hadley]

@krlmlr good to know - better to be explicit anyway

[krlmlr]

In clang: -Wimplicit-fallthrough (http://stackoverflow.com/a/27965827/946850), work in progress for gcc.

[hadley]

Thanks!

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/