summarise() works for empty data frames

[krlmlr]

Fixes #3071.

[hadley]

It’d be useful to add a test for mean() too. In general, I’m not sure if the hybrid handlers have been written with an eye to the zero length case.

[krlmlr]

Indeed, mean(na.rm = TRUE) doesn't handle empty vectors. Will review all handlers.

[krlmlr]

I'm seeing an average 10% slowdown with this PR with flights, note the different scales for has_na: TRUE and na_rm: FALSE due to #3288.

I've started a writeup, http://rpubs.com/krlmlr/dplyr-spec-template, would that be suitable as an internal dplyr vignette?

[krlmlr]

Update: A second measurement with times = 100 suggests performance is about equal:

## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  0.987559   0.011399  86.635  < 2e-16 ***
## typereal    -0.002408   0.009307  -0.259  0.79883    
## has_naTRUE   0.005208   0.009307   0.560  0.58271    
## funvar       0.010561   0.011399   0.927  0.36643    
## funsum       0.034494   0.011399   3.026  0.00726 ** 
## na_rmTRUE   -0.035563   0.009307  -3.821  0.00125 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

[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/