Change in ggsave behaviour

[ecorreig]

I used to "add" ggsave("something.png") to a ggplot object but this has stopped working after version 3.3.3. In case it's been decided that this "feature" wasn't a good idea and the removal was on purpose, please better document that change (even if the possibility to do + ggsave() wasn't documented before; I think many people used it because it might the pipes very elegant).

In short, this worked on 3.3.3:

library(ggplot2)
data.frame(a = rnorm(100), b = rnorm(100)) |> 
  ggplot(aes(a, b)) +
  ggsave("temp.png")

But throws this error on 3.3.4:

Error: Can't add `ggsave("temp.png")` to a ggplot object.

[clauswilke]

Honestly, I've never seen anybody do this, and it's definitely not a supported feature to my knowledge. The only reason that it worked is probably that ggsave() returned NULL, and you can add anything to a plot if it returns NULL. But that doesn't mean it's sensible.

You were relying on the fact that ggsave() uses the last plot that was made if none is provided as an argument, and somehow it found the plot that you just generated. (This could just as well have gone wrong and you would have gotten one plot prior.)

The valid way to do what you want to do is this code:

library(ggplot2)
data.frame(a = rnorm(100), b = rnorm(100)) |> 
  ggplot(aes(a, b))
ggsave("temp.png")

This is the commit that caused the change in behavior: 23baab0

[thomasp85]

Well, that certainly never was a feature — I'm surprised it ever worked in the first place... it is hard to anticipate when someone's hack stops to work which is also why it hasn't been communicated.

I don't suspect many have used this hack based on the fact that this is the first time I've seen it, but I could be wrong

[ecorreig]

Honestly I never thought I was doing a hack, but I understand this was the case.

But still, with the new changes in R 4.1 and ggplot 3.3.4 I can do:

data.frame(a = rnorm(100), b = rnorm(100)) |> 
  ggplot(aes(a, b)) +
  geom_point() |>  
  ggsave(file = "temp.png")

But the mixing of pipes and + is prone to errors, so I don't know if it would be a good idea to add a feature of "adding" to ggsave.

You find more information about what I mean in the response I got in stackoverflow.

[yutannihilation]

Can we just add a NEWS item and close this issue? If so, I'll address this in #4518.

[thomasp85]

Fine by me. There's already a bullet about the invisible return value of ggsave so maybe just add it there ?

[yutannihilation]

Thanks, added. Does this sound good?

Note that, as a side effect, an unofficial hack <ggplot object> + ggsave() no longer works (#4513).

[ecorreig]

I still think that the hack adheres more to the adding/piping philosophy than having a ggsave floating around somewhere after the plot, but you can close it, yes, thanks.

[thomasp85]

I think that sounds fine, yeah

[yutannihilation]

@ecorreig
I'm sorry, but I'd say it's not the part of "the adding philosophy" as none of the other output functions (i.e. print() and plot()) is allowed to add. For piping, yeah, we all wish ggplot2 had been born after magrittr...

@thomasp85
Thanks, could you just approve #4518?

[acircleda]

I used + ggsave() not as a hack but because that is how I learned from others to save ggplot graphics. I think a lot of people do this to pipe in/replace the plot= argument. Any way to make this a feature instead of calling it a hack? I feel like this will break a lot of people's scripts, including mine.

[thomasp85]

@acircleda when we say that it is a hack we do not necessarily imply any ill-intend on the user, just that it is an accidental functionality that was never promised to work. The fact that some have used it does not really change our stance on it.

adding ggsave did not stop working because we decided to remove the functionality, but because internal wanted changes resulted in the hack stopping to work - such is the nature of hacks.

I can certainly emphasise with your situation but I hope you can understand that development would be completely locked down if we were to take all unintended uses of our code into consideration when fixing bugs etc

[DSLituiev]

This is indeed very anti-intuitive. Implicit arguments (i.e. taking last plot) are also a prone to problems when working interactively.

Not sure if developers follow how most users believe the code should behave, but if so,
ggplot() %>% ggsave("filename.png") would be much cleaner and intuitive than ggplot() %>% ggsave(file="filename.png", .)