No plyr #3013
No plyr #3013
Conversation
|
r-devel complains about plyr in tests: (https://travis-ci.org/tidyverse/ggplot2/jobs/459483350#L1765) It seems you need to replace two more plyr things:
|
| @@ -0,0 +1,334 @@ | |||
| # Adds missing elements to a vector from a default vector | |||
| defaults <- function(x, y) c(x, y[setdiff(names(y), names(x))]) | |||
There was a problem hiding this comment.
Isn't this basically modify_list?
There was a problem hiding this comment.
It is more or less the opposite... in set operation terms modify_list() overwrites the intersection of elements, while defaults adds the complement elements
|
I've unexported the plyr substitutes... Unless there are other comments @hadley I'll merge and wave goodbye to plyr |
|
Sounds good. |
|
Thank you for this @thomasp85! I was profiling one of my ggplots to try to speed it up, and noticed a bunch of time spent in plyr, but when I came to look for the source code on master I saw that you'd made this change. Just by installing from master, the time dropped from 1020ms to 720ms!
|
|
Great to hear, though a lot of the performance improvements is probably due to text dimension caching implemented in #2993... You should get an additional speedup by installing the development version of gtable |
|
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/ |
This PR writes out all ply dependencies from ggplot2. The following strategies has been used:
**ply
Except for
ddplyall use of**plyfunctions has been substituted with base equivalents.ddplyhas been substituted with the newdapplyfunction that mimics its API and behaviour. It is ~40% faster in normal use by avoiding theas.quotedcall.Utility functions
Use of
split_indices,summarise, andcolwisehas been substituted with base equivalent approaches.quickdfhas been substituted with the already existingnew_data_frameemptyis now exported so the vignette doesn't need to depend on the plyr versiondefaults,unrowname,rename,revalue,id,count,join.keys,as.quoted,round_any, andrbind.fillhas all been reimplemented. In a few circumstances they have been lifted as-is from plyr, but often they have been rewritten to reduce complexity by only targeting use in ggplot2. All reimplementations are either on par or slightly faster than the plyr counterpartsCurrently, a selection of reimplementations has been exported as internal functions to make it easier for extension developers to write out plyr as well.
All rewritten functions are currently collected in
compat-plyr.Rbut I think it makes sense to distribute them into files guided by their use rather than keep them there for historical reasons.There are no unit tests yet but these will come once we have agreed on the implementations and approaches.