Feature request: consistent validation of test set

Dear colleagues,

I have trained a forecasting model for a grouped dataframe as follows:

model <- testing_data  %>% filter(a==70, b!=3) %>% filter(year(snsr_dt)<2019) %>% 
    model(prophet3 = fable.prophet::prophet(snsr_val_clean ~ season("month", 3, type = "multiplicative")),
          prophet4 = fable.prophet::prophet(snsr_val_clean ~ growth("linear") + season("week", 2, type = "multiplicative") + season("month", 2, type = "multiplicative")),
          prophet5 = fable.prophet::prophet(snsr_val_clean ~ growth("linear") + season("week", 2, type = "multiplicative") + season("year", 2, type = "multiplicative")),
          prophet6 = fable.prophet::prophet(snsr_val_clean ~ growth("linear") + season("week", 2, type = "multiplicative") + season("month", 2, type = "multiplicative") + season("year", 2, type = "multiplicative")))
fc <- model %>% forecast(h=52, testing_data  %>% filter(a==70, b!=3) %>% filter(year(snsr_dt)>2019)) 

And I asses the accuracy as follows:

test <- accuracy(fc, testing_data  %>% 
filter(a==70,b!=3) %>% 
filter(year(snsr_dt)>2019))

And I am getting the following warning (nice that this is so transparent btw):

Warning message:
The future dataset is incomplete, incomplete out-of-sample data will be treated as missing. 
104 observations are missing between 2019-01-01 and 2019-12-23 

However this presents a problem in the sense that most of the time series they do not have equal length. For processing my data as was using:

ts_tibble <- as_tsibble(df, 
                            key=c(a,b,c), index = snsr_dt)
    print("Filling gaps for not breaking groups")
    ts_tibble <- ts_tibble %>% fill_gaps()

Which definitely helped for the training of multiple models but should I understand that with the current API I need to align my timeseries to the same end_date to be able to assess the accuracy? Is there any way of making this more consistent?

BR
/Edgar

The issue names the R entry points as as_tsibble(), fill_gaps(), forecast(), and accuracy(), but no repository files or tests. Start by reproducing the grouped-data example and tracing how accuracy() handles incomplete future data; done should be a documented or implemented consistent validation behavior for unequal series lengths.