ARROW-11704: [R] Wire up dplyr::mutate() for datasets

[nealrichardson]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-11704

[nealrichardson]

I'll add a couple more tests, particularly around error handling, but I'd like to move on to other issues and get this merged so that others can test it out more widely. cc @ianmcook @jonkeane

[ianmcook]

Looks like mutate() on datasets errors when expressions are literals:

ds %>% transmute(x=42) %>% head(1)
## Error in dataset___ScannerBuilder__ProjectExprs(self, cols, names(cols)) : 
##  Invalid R object for std::__1::shared_ptr<arrow::dataset::Expression>, must be an ArrowObject

ds %>% transmute(x="foo") %>% head(1)
## Error in dataset___ScannerBuilder__ProjectExprs(self, cols, names(cols)) : 
##  Invalid R object for std::__1::shared_ptr<arrow::dataset::Expression>, must be an ArrowObject

[nealrichardson]

Thanks @ianmcook, fixed in 3ac4086, PTAL.

[ianmcook]

Thanks @ianmcook, fixed in 3ac4086, PTAL.

Works fine now 👍 Thanks!

[ianmcook]

transmute() with no arguments should return no columns. Currently it returns all columns. This is true for Tables and RecordBatches too.

[ianmcook]

So now we have scalar recycling in mutate() for Datasets but not yet for Tables and RecordBatches, correct?

[nealrichardson]

So now we have scalar recycling in mutate() for Datasets but not yet for Tables and RecordBatches, correct?

Correct, that's ARROW-11705