Unexpected blank page in pdf about cowplot

[JSanzR]

Doing:
pdf("test.pdf", onefile=FALSE)
seems to get rid of the blank page. See: http://stackoverflow.com/questions/12481267/in-r-how-to-prevent-blank-page-in-pdf-when-using-gridbase-to-embed-subplot-insi)

[clauswilke]

Yes, this happens. I'm not sure why. However, fundamentally pdf() is not the supported mechanism to save ggplot2 figures. The supported mechanism is ggsave(). (I'm pretty certain that ultimately, this is a ggplot2 bug, not a cowplot bug.)

[jscamac]

I don't think this a ggplot bug.

Problematic cowplot example giving a blank page:

library(couplet)
library(ggplot2)
p <- ggplot(mpg, aes(displ, hwy)) +
geom_point() +
facet_wrap(~class)
pdf('test.pdf',width=6.69, height=6.69)
plot_grid(p, p, labels=letters[1:2], ncol = 1, label_size = 11, hjust = -1.5)
dev.off()

However, if I replace plot_grid with grid.arrange I don't have the problem,

library(gridExtra)
p <- ggplot(mpg, aes(displ, hwy)) +
  geom_point() +
  facet_wrap(~class)
pdf('test.pdf',width=6.69, height=6.69)
grid.arrange(p, p, ncol = 1)
dev.off()

[clauswilke]

It is a ggplot2 bug.

This code produces one page of output:

library(ggplot2) # load only ggplot2, not cowplot
p <- ggplot(mpg, aes(displ, hwy)) +
geom_point() +
facet_wrap(~class)
pdf('test.pdf',width=6.69, height=6.69)
p
dev.off()

But this produces two:

pdf('test.pdf',width=6.69, height=6.69)
g <- ggplot2::ggplotGrob(p)
p
dev.off()

So calling the function ggplot2::ggplotGrob has the side-effect of creating an extra page in the pdf device, even if the result of that call is assigned to a new object. Cowplot calls ggplot2::ggplotGrob to create the grobs it works with.

[marcus1487]

I have been struggling with this issue myself within my own hacky version of cowplot for while and have not been satisfied with work arounds I have found. I have usually used the pdf(NULL) solution when calling ggplotGrob in my own scripts. From what I can tell in discussions about this issue within ggplot2 forums it has to do with the definition of line and other plotting values do not have a meaning until a device is opened, so the default device is opened when ggplotGrob is called. I therefore have two comments on this matter.

First, I think that the current correct workaround to use cowplot should be the following:

p <- ggplot(mpg, aes(displ, hwy)) + geom_point() + facet_wrap(~class)
p <- ggplotGrob(p)
ggsave(p, 'test.pdf')
## or
pdf('test.pdf')
p
dev.off()

Also this is not a pdf versus ggsave issue as both of the following produce 2 plots.

p <- ggplot(mpg, aes(displ, hwy)) + geom_point() + facet_wrap(~class)
ggsave('test.pdf', p)
## and 
pdf('test.pdf')
p
dev.off()

Instead it is an issue with any device that can save multiple plots to a single file (e.g. PDF).

Second, even though ggplot needs an open device when a call to ggplotGrob is made, I don't believe anything within cowplot depends on the device opened when ggplotGrob is called. Thus I believe this bug (whether it is in gggplot2 or not) could be resolved in cowplot by wrapping the only ggplotGrob call in cowplot with a null device call. This may be bad form to open a NULL device, but seems like the only solutions as @hadley indicates that there is no plan to fix this issue as noted in tidyverse/ggplot2#809 . I have opened a fork to resolve this issue (https://github.com/marcus1487/cowplot/tree/blankpage) and am happy to open a pull request for this issue unless this will not be integrated as it is seen as bad form to use a NULL device.

[clauswilke]

Yes, please set up a pull request. Two requests from my side:

1. Comment the code, so it is clear why these lines are added.
2. Verify that this is the only place where we call ggplot2::ggplotGrob().

Thanks!

[marcus1487]

I have opened the pull request (with comment). The only ggplotGrob call in the R directory scripts is wrapped in a NULL device call within the draw.R script in the pull request. There are many ggplotGrob calls in the vignettes and docs, but I don't think these need to be fixed with this pull request.

P.S. Thanks for saving me from porting all of my hacky code now that cowplot is here! Great work!