scale_*_reverse() is ignored when limits are set on Coord

[vitor-mendes-iq]

Since I updated R to version 4.0 the scale_x_reverse () function has stopped working together with renderPlot () or renderPlotly( ), which I use in a shiny app that I develop ...
However when I test scale_x_reverse () out of the render it works fine. It is worth mentioning that in all other versions of R it was working normally in both Windowns (7, 8 and 10) and Linux (Ubunto, mint, Debian, Suse). Now I am use Windowns 10, R (4.0), RStudio (1.2.5042).

Can someone help me?

---

Brief description of the problem

# ` # Variables
  peran_multi <<- 0
  ysup_multi <- max(NMRData[1,])
  yinf_multi <- min(NMRData[1,])
  ysup_multi <- ysup_multi + ysup_multi*0.03
  chkzoom_multi <<- 1
  idb_multi <<- 0


  # Data.frame
  testy_multi <<- data.frame(Chemical_Shift=CS_values_real[1,],Spectrum=NMRData[1,])
  ranges_multi <- reactiveValues(x = c((min(testy_multi$Chemical_Shift)), max(testy_multi$Chemical_Shift)), y = c(-100,ysup_multi))
  # ranges_multi_plot <- reactiveValues(x = sort(c((min(testy_multi$Chemical_Shift)), max(testy_multi$Chemical_Shift)), decreasing = TRUE), y = c(yinf_multi,ysup_multi))
  spectrums_multi <- reactiveValues(dat = data.frame(Chemical_Shift=CS_values_real[1,],Spectrum=NMRData[1,]))

  # Plot
    output$plot_multi <- plotly::renderPlotly({

      ggplot2::ggplot(spectrums_multi$dat,ggplot2::aes(Chemical_Shift,Spectrum)) + ggplot2::geom_line(color='blue') + ggplot2::scale_x_reverse() +

        ggplot2::coord_cartesian(xlim = ranges_multi$x, ylim = ranges_multi$y, expand = FALSE) +

        ggplot2::theme(axis.text.x = ggplot2::element_text(size = 12, color = "#000000"),
                       axis.text.y = ggplot2::element_text(size = 12, color = "#000000"),
                       title = ggplot2::element_text(face = "bold", color = "#000000", size = 17),
                       axis.title = ggplot2::element_text(face = "bold", color = "#000000", size = 15)
                       ) +

        ggplot2::labs(x = "Chemical Shift", y = "Intensity")

    })
`

[yutannihilation]

Could you ask on plotly's repo (with a minimal reproducible example)? I'm not sure if this is what should be fixed on plotly's side, but we have too little knowledge about plotly's internals to identify the cause of the problem.

[vitor-mendes-iq]

Sorry yutannihilation for not put a reproducible example. I am a new programmer and this is my first question on forum.
try to see this new reproducible example now

`library(shiny)
library(ggplot2)
library(plotly)


ui <- fluidPage(
        fluidRow(align = "center",
          mainPanel(
           plotly::plotlyOutput("plot_multi")
          )
        )
)
  

server <- function(input, output) {




# Data.frame
z <<- runif(1001, min=0, max=1000000)
f <<- seq(0,10,0.01)
testy_multi <<- data.frame(Chemical_Shift=f,Spectrum=z)
ranges_multi <- reactiveValues(x = c((min(testy_multi$Chemical_Shift)), max(testy_multi$Chemical_Shift)), y = c(-100,1000000))
spectrums_multi <- reactiveValues(dat = data.frame(Chemical_Shift=f,Spectrum=z))

# Plot
output$plot_multi <- plotly::renderPlotly({
  
  ggplot2::ggplot(spectrums_multi$dat,ggplot2::aes(Chemical_Shift,Spectrum)) + ggplot2::geom_line(color='blue') + 
    
    ggplot2::coord_cartesian(xlim = ranges_multi$x, ylim = ranges_multi$y, expand = FALSE) + ggplot2::scale_x_reverse() +
    
    ggplot2::theme(axis.text.x = ggplot2::element_text(size = 12, color = "#000000"),
                   axis.text.y = ggplot2::element_text(size = 12, color = "#000000"),
                   title = ggplot2::element_text(face = "bold", color = "#000000", size = 17),
                   axis.title = ggplot2::element_text(face = "bold", color = "#000000", size = 15)
    ) +
    
    ggplot2::labs(x = "Chemical Shift", y = "Intensity")
  
})

}

shinyApp(ui = ui, server = server)
`

[yutannihilation]

Thanks for the reproducible example, I confirmed your example works with ggplot2 v3.2.1 but won't with v3.3.0 (or the current master). But, again, could you file this issue on plotly's repo? I wasn't complaining about the lack of reproducibility, I'm just saying we cannot fix this (at least without any help from plotly's maintainers).

[vitor-mendes-iq]

Thanks for responding yutannihilation , I agree with what you said about going to plotly and now I go to the forum, but I have the same problem using renderPlot () which is not from plotly. This was the reason for looking for this forum before.

[yutannihilation]

@vitor-mendes-iq
Ah, OK. I misunderstood your point, sorry... Here's a minimal reprex:

library(ggplot2)

d <- data.frame(x = 1:2)

p <- ggplot(d, aes(x, x)) +
  geom_point() +
  scale_x_reverse()

p

p + coord_cartesian(xlim = range(d$x))

^{Created on 2020-05-23 by the reprex package (v0.3.0)}

The reversed x scale is ignored because you specifies xlim in the order.

[yutannihilation]

@thomasp85
Sorry, at the time of implementing #3958, I thought this is the correct behaviour, but I found it didn't behave so in the past. Do you think I need to fix this before releasing v3.3.1?

[clauswilke]

I would expect all these plots to produce the same result. That is not currently the case. I'm particularly confused by the cases that remove the points. Note that this is not a rounding error; if I make the limits broader the points are still removed.

library(ggplot2)

d <- data.frame(x = 1:2)

p <- ggplot(d, aes(x, x)) + geom_point()
  
p + xlim(2, 1)

p + scale_x_continuous(limits = c(2, 1))
#> Warning: Removed 2 rows containing missing values (geom_point).

p + coord_cartesian(xlim = c(2, 1))

p + scale_x_reverse()

p + scale_x_reverse(limits = c(1, 2))
#> Warning: Removed 2 rows containing missing values (geom_point).

p + scale_x_reverse() + coord_cartesian(xlim = c(1, 2))

^{Created on 2020-05-22 by the reprex package (v0.3.0)}

[yutannihilation]

The difference between xlim() and scale_x_continuous() is that the former automatically sets the trans to reverse when the limits are specified in the reverse order.

ggplot2/R/limits.r

Lines 118 to 122 in 4826838

	if (!any(is.na(lims)) && lims[1] > lims[2]) {
	trans <- "reverse"
	} else {
	trans <- "identity"
	}

This heuristic made me think that the order of the limit do matter, and maybe the user is supposed to specify them in the same order of the scale...? Sorry, I'm still confused.

[clauswilke]

I guess this is also a consistent principle: If explicit limits are provided, they always supersede the natural direction of the scale.

[yutannihilation]

Ah, thanks, makes sense. Then, we have two issues to consider:

1. When scale limits are provided in the opposite order to the direction of the scale, the limits are ignored and all data are removed (This should be fixed).
2. When coord limits are provided in the opposite order to the direction of the scale, the limits supersede the natural direction of the scale. This seems to do the right thing, but it didn't behave so before v3.2.1 (v3.3.0 was broken here).

The examples below shows issue 2. What I was wondering is whether we should fall back to the behaviour of v3.2.1, but now I feel the current behaviour is OK.

d <- data.frame(x = 1:2)

ggplot(d, aes(x, x)) +
  geom_point() +
  scale_x_reverse() +
  coord_cartesian(xlim = range(d$x))

v3.2.1

v3.3.0

current master