Facet_wrap and scale="free" unexpectedly centers y-axis at zero in ggplot2

[trichterhb]

From this dataframe

     df <- data.frame(cat=c(rep("X", 20),rep("Y", 20), rep("Z",20)), 
                         value=c(runif(20),runif(20)*100, rep(0, 20)), 
                         var=rep(LETTERS[1:5],12))

i want to create facetted boxplots. One facet would contain only Zeros.

library(ggplot2)
  

  p1 <- ggplot(df, aes(var,value)) + geom_boxplot() + facet_wrap(~cat, scale="free") 
  p1

The results is aesthetically dissactisfactory as it centers the y-axis of the empty panel at zero. Note, that "scale="free"" causes this. I want to start all y-scales at zero. I tried several solutions found on the web:

    p1 + scale_y_continuous(expand = c(0, 0)) # not working
    p1 + expand_limits(y = 0) #not working
    p1 + scale_y_continuous(limits=c(0,NA)) ## not working
    p1 + scale_y_continuous(limits=c(0,100)) ## partially working, but defeats scale="free"
    p1 + scale_y_continuous(limits=c(0,max(df$value))) ## partially working, see above
    p1 + scale_y_continuous(limits=c(0,max(df$value))) + expand_limits(y = 0)## partially working, see above

One solution would possibly be to replace the zero's with very tiny values, but maybe you can find a more straightforward solution. Thank you.

[teunbrand]

Please note that an identical question has (per time of writing) two answers at SO.

[trichterhb]

Please note that an identical question has (per time of writing) two answers at SO.

Thank you. i was indeed first posting at at SO, but was recommended to here at first. the two answers were added later.

The question would may be interesting if this behaviour is working as intended.

[thomasp85]

The proposed solution with passing in a function is the correct approach. There is currently no way to directly control the different scales used in facets with scales = "free" outside of setting scale arguments to functions that can dynamically respond to the calculated values