layout and plot.stars

Coming back to the example which initiated the issue, I have found the 
trick following your hints Edzer, but it makes a strange rational. Here 
we go:

nf <- layout(matrix(c(1,2),2,1,byrow = TRUE), c(3,3), c(3,1), TRUE)
layout.show(nf)

plot(st_geometry(mydept),col="grey",border="grey90")
plot(st_geometry(fzone1),col=mypal(12)[1],border=NA,add=TRUE)
plot(st_geometry(fzone2),col=mypal(12)[2],border=NA,add=TRUE)
plot(st_rasterize(ztot),col=mypal(12),main="",breaks="equal",key.pos=NULL,reset=FALSE,add=TRUE)
plot(st_geometry(mydept),col=NA,border="grey90",add=TRUE)

mybox<-bbox2sf(n=0.5,s=0,w=0,e=10,crs=2154) # crs is just crap here, I 
do not need it thenafter
mypoly<-st_make_grid(mybox,n=c(12,1),what="polygons")
plot(mypoly,col=mypal(12),border=mypal(12))
plot(mybox,add=TRUE)

Makes exactly what I was intending to make. The critical point was to 
explicitely keep reset=FALSE in the 4th plot (actually a plot.stars). If 
not, the 6th plot is plotted in region #1 erasing the previous plots...

Some personal remarks:
- in the first plot, I ommited the argument reset=FALSE, however this 
does not make a problem (maybe because I just plot a geometry ?)

- the 4th plot is definitely strange with reset=FALSE and add=TRUE 
together, isn't it ?

Thanks Roger and Ezder for bailing me out once again... :-)
Patrick



Le 15/03/2022 ? 16:02, Edzer Pebesma a ?crit?:

In addition to what Roger wrote, you can use e.g.

layout(matrix(1:4,2))
library(stars)
s = st_as_stars(L7_ETMs)
image(s[,,,1])
plot(s[,,,1], key.pos = NULL, reset = FALSE)
plot(s[,,,1], key.pos = NULL, reset = FALSE, main = NULL)

to fill the sub-plots of layout() incrementally. Note the key.pos and 
reset arguments to plot.stars(): they make sure plot.stars doesn't 
mess with the layout settings.

On 15/03/2022 11:45, Patrick Giraudoux wrote:

Great ! Thanks Roger.? On this basis, I have a way to explore the issue
now. Will give a feed-back on the list once done.
Best,
Patrick

Le 15/03/2022 ? 11:27, Roger Bivand a ?crit?:

On Tue, 15 Mar 2022, Patrick Giraudoux wrote:

Hi,

I have a trouble with the combination of layout and plot.stars. e.g.

nf <- layout(matrix(c(1,2),2,1,byrow = TRUE), c(3,3), c(3,1), TRUE)
layout.show(nf)

plot(st_geometry(mydept),col="grey",border="grey90")
plot(st_rasterize(ztot),col=mypal(12),main="",breaks="equal",add=TRUE)

You have already noticed that sf and stars, like raster and terra,
modify the assumptions of base plot methods, as
graphics::filled.contour(), unless some ordering and argument
conditions are met, crucially the non-base reset= argument. I do not
think that you can use layout() at all.

library(stars)
nc <- st_read(system.file("gpkg/nc.gpkg", package="sf"))
bir74_rast <- st_rasterize(nc["BIR74"])
plot(bir74_rast, reset=FALSE)
plot(st_geometry(nc),border="grey90", add=TRUE)
gridGraphics::grid.echo()
library(grid)
g <- grid.grab()
plot(nc["BIR74"])
gridGraphics::grid.echo()
gv <- grid.grab()
grid.newpage()
gridExtra::grid.arrange(g, gv, ncol=2)

Grabbing the base graphics device state lets you use
gridExtra::grid.arrange() to place multiple graphics objects; here I
haven't tried to constrain aspect or relative sizes. I don't think
that the plot methods in sf and stars play well with layout, because
they use it themselves internally.

Hope this helps,

Roger

plot(mypoly,col=mypal(12),border=mypal(12))

I expect that the first two plots display in region #1, the second
added to the first, and the third plot in region #2. However, this is
not what happens: actually, the third plot displays in region 1
erasing the others. I understand that plot.stars when not "added"
does not respect the layout definition (and displays its own
regions), and that my problem comes from the way plot.stars deals
with that.

Has anyone an idea about a workaround ?

????[[alternative HTML version deleted]]