Closed benjcunningham closed 8 years ago
Still, why does this only happen in Travis?
Oops. Didn't realize my box hadn't updated to dplyr 0.5.0
yet, which apparently makes the hard switch to erroring on non-existent columns. Hopefully now I can cut down on all the needless commits / reverts / builds while I debug.
Finally got to the root of the problem in 634caad. Based on some of the issues / conversations currently happening in the dplyr
, tibble
, and purrr
threads, it seems that the best fix for now is just using unclass()
inside of as.body()
to make sure we can still take advantage of df$xyz
returning NULL
when xyz
doesn't exist in the object.
Need to take a look at how the depends are currently aligned between all of the packages, but with tibble 1.1, it looks like this might be able to be re-rewritten after all. Probably low priority, though.
Here's the error from the failed Build #41:
Can't reproduce outside of Travis, but here are some notes from ad1bd66:
Looks like the entire test schedule fails out as soon as it hits the
as.body()
call inside ofgc_event_import()
. I still think this is an issue with tibbles, and now realize thatpurrr::by_row()
probably coerces data frames too, so the edits in ad1bd66 didn't cover everything.Still, why does this only happen in Travis? :angry: