Breaking (and other) changes for dplyr 0.9.0

[krlmlr]

I'd like to suggest that we break the things we want to break very early in the release cycle. We should only break things where there is a clearly better alternative in the existing CRAN version. This could be done in a separate pull request first, with early notification to the downstream dependencies.

New interfaces should not break existing ones, we can mark them as deprecated in the docs and start breaking them in the version after that. This includes changes in behavior.

This meta-issue could collect the scheduled breakages. Off the top of my head, this includes:

• all.equal() (all.equal() problems #2751)
• Warn if vars attribute is a list (Compatibility fix for deserialized grouped dfs #2786 (comment))
• Deprecation warnings in summarise_each() and mutate_each() (Deprecation messages #2791)
• Remove dbplyr from .travis.yml and appveyor.yml
• Finally remove rbind_list() and rbind_all() (Correctly handle empty names in arguments to bind_rows() #2828 (comment))
• Add api_roclet to DESCRIPTION (and pkgapi to Suggests) (Update API description #2805 (comment))
• Revert Revert unintended API change #3055
• Change warning to error in throw error when providing invalid column to distinct() #3140
• Change default of na_matches argument
• Throw an error when trying to access variables through a leaked data mask (Avoid cleaning the data mask #3443 (comment))
• Order of variables, Order of variables #3481
• Raise error when accessing hybrid functions not on the search path (no search with findFun when we know the function comes from dplyr. #3458 (comment))
• Implement only methods for data.frame, with an internal reconstruct() that restores the "tbl_df" class if the input had it (Use vctrs::vec_restore to preserve data frame subclass and attributes #3429)
• Tidy grouped data attributes (tidy grouped data attributes #3489)
• Fail if matrix is passed to if_else() and case_when() (if_else converts matrix into vector (unlike base::ifelse) #3256)

[hadley]

We'll reserve breaking API changes for 0.9.0. Maybe it would be better to have a label than a meta issue?

[krlmlr]

I'm not sure. They lie around here peacefully, waiting patiently for their time to come...

[cboettig]

• Change default of na_matches argument

👏 🙏

[hadley]

• all.equal() (all.equal() problems #2751) - done.
• Warn if vars attribute is a list - likely to be out of date since we'll be reconsidering the class structure anyway
• Deprecation warnings in summarise_each() and mutate_each() - now Deprecation updates #4655
• Add api_roclet to DESCRIPTION - still not convinced this technique is mature enough
• Revert Revert unintended API change #3055 - now distinct needs to error on unknown variable names #4656
• Change warning to error in throw error when providing invalid column to distinct() #3140 - also distinct needs to error on unknown variable names #4656
• Change default of na_matches argument - I no longer believe this is a good idea
• Throw an error when trying to access variables through a leaked data mask (Avoid cleaning the data mask #3443 (comment)) - too vague
• Order of variables, Order of variables #3481 - milestone added
• Raise error when accessing hybrid functions not on the search path (no search with findFun when we know the function comes from dplyr. #3458 (comment)) - hybrid functions are no more
• Implement only methods for data.frame, with an internal reconstruct() that restores the "tbl_df" class if the input had it (Use vctrs::vec_restore to preserve data frame subclass and attributes #3429) - @lionel is working on it
• Fail if matrix is passed to if_else() and case_when() (if_else converts matrix into vector (unlike base::ifelse) #3256) - now problem of funs package

[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/