Setting anchor = "free" in R subplot() causes the graph to disappear

[duncanrellis]

I'm trying to build a subplot of stacked scatter plots, where there are multiple axes on each scatter plot.

Below is an example of the type of plot. To position the yaxis side-by-side, I have layout(yaxis(anchor = "free")) set am using position to place the axis.

All my attempts to set anchor = "free" in subplots left my plot space completely blank. Some testing revealed that even a subplot constructed from a single plot will not display if anchor="free"

Please see below REPREX.

require(plotly)
df <- data.frame(x = 1:10, y = 1:10, y2 = 10:19)
plt <- plot_ly(data = df) %>%
  add_trace(x = ~x, y = ~y,
            mode = "markers", type = "scatter",
            yaxis = "y") %>%
  add_trace(x = ~x, y = ~y2,
            mode = "markers", type = "scatter",
            yaxis = "y2") %>%
  layout(
    xaxis = list(
      domain = c(.12, .88)
    ),
    yaxis = list(
      title = "y_1"
    ),
    yaxis2 = list(
      title = "y_2",
      anchor = "free",
      position = .10,
      overlaying = "y"
    ),
    margin = list(
      pad = 60
    )
  )
subplot(plt)

[cpsievert]

I think this is a plotly.js bug...that R code generates this fatal JS error:

Uncaught TypeError: t.match is not a function
    at r.cleanId (plotly-latest.min.js:61)
    at Object.r.cleanLayout (plotly-latest.min.js:61)
    at Object.r.plot (plotly-latest.min.js:61)
    at Object.renderValue (plotly.js:162)
    at htmlwidgets.js:653
    at Array.forEach (<anonymous>)
    at forEach (htmlwidgets.js:55)
    at htmlwidgets.js:576
    at Array.forEach (<anonymous>)
    at forEach (htmlwidgets.js:55)

[r2evans]

@cpsievert, I don't have experience with plotly outside of R, do you know if this is reproducible in a non-R context? (I understand your point that the error is coming from plotly and not R.) In your experience, is plotly (the organization) responsive to bugs like that?

[cpsievert]

I just filed an issue here plotly/plotly.js#4844. Please follow that thread for updates (then, if and when it gets fixed in plotly.js, you can poke me about updating plotly.js in the R package :) )

[cpsievert]

Oh actually, this is not a plotly.js problem. And now that I look closer, why do you need subplot(plt) instead of plt?

[duncanrellis]

Oh actually, this is not a plotly.js problem. And now that I look closer, why do you need subplot(plt) instead of plt?

For this particular example I don't, I was just trying to provide a minimal set of code that didn't function. My use case that uncovered this issue was trying to create a subplot of two "stacked" plots using subplot()

Try 1.

• This approach works, but the axis are overlapped.

require(plotly)
df <- data.frame(x = 1:10, y = 1:10, y2 = (10:19)^2)
plt1 <- plot_ly(data = df) %>%
  add_trace(x = ~x, y = ~y,
            mode = "markers", type = "scatter",
            yaxis = "y") %>%
  add_trace(x = ~x, y = ~y2,
            mode = "markers", type = "scatter",
            yaxis = "y2") %>%
  layout(
    xaxis = list(
      domain = c(.12, .88)
    ),
    yaxis = list(
      side = "left",
      title = "y_1"
    ),
    yaxis2 = list(
      side = "left",
      title = "y_2",
      ## anchor = "free",
      ## position = .10,
      overlaying = "y"
    ),
    margin = list(
      pad = 60
    )
  )

plt2 <- plot_ly(data = df) %>%
  add_trace(x = ~x, y = ~y,
            mode = "markers", type = "scatter",
            yaxis = "y") %>%
  add_trace(x = ~x, y = ~y2,
            mode = "markers", type = "scatter",
            yaxis = "y2") %>%
  layout(
    xaxis = list(
      domain = c(.12, .88)
    ),
    yaxis = list(
      side = "left",
      title = "y_1"
    ),
    yaxis2 = list(
      side = "left",
      title = "y_2",
      ## anchor = "free",
      ## position = .10,
      overlaying = "y3"
    ),
    margin = list(
      pad = 60
    )
  )

subplot(plt1, plt2, nrows = 2)

##Try 2

• When you un-comment out the anchor = "free" argument the entire plot disappears, regardless of where position is set.

Require(plotly)
df <- data.frame(x = 1:10, y = 1:10, y2 = (10:19)^2)
plt1 <- plot_ly(data = df) %>%
  add_trace(x = ~x, y = ~y,
            mode = "markers", type = "scatter",
            yaxis = "y") %>%
  add_trace(x = ~x, y = ~y2,
            mode = "markers", type = "scatter",
            yaxis = "y2") %>%
  layout(
    xaxis = list(
      domain = c(.12, .88)
    ),
    yaxis = list(
      side = "left",
      title = "y_1"
    ),
    yaxis2 = list(
      side = "left",
      title = "y_2",
      anchor = "free",
      position = .10,
      overlaying = "y"
    ),
    margin = list(
      pad = 60
    )
  )

plt2 <- plot_ly(data = df) %>%
  add_trace(x = ~x, y = ~y,
            mode = "markers", type = "scatter",
            yaxis = "y") %>%
  add_trace(x = ~x, y = ~y2,
            mode = "markers", type = "scatter",
            yaxis = "y2") %>%
  layout(
    xaxis = list(
      domain = c(.12, .88)
    ),
    yaxis = list(
      side = "left",
      title = "y_1"
    ),
    yaxis2 = list(
      side = "left",
      title = "y_2",
      anchor = "free",
      position = .10,
      overlaying = "y3"
    ),
    margin = list(
      pad = 60
    )
  )

subplot(plt1, plt2, nrows = 2)

In my testing I found that even if only a single plot obj was added to the subplot() function, the plot would be blank, which is why I constructed the REPREX in the above manner.

Hope that's clear.

[duncanrellis]

• Not quite related, but I find it confusing from a plot construction perspective that the overlaying argument on the second plot is indexed at a "global" level, that is, it must be set to overlaying = "y3" instead of overlaying = "y" to appear in the right place.

Not sure if that's a design feature, or unintended consequence of subplot()