Long legend with horizontal orientation got truncated when converting ggplot object using ggplotly but it's fine when using plot_ly

Hi all, as a work-around approah to display a very long multi-column legend, I tried to use the layout(legend = list(orientation = "h")).

Here is the data_frame I was working with:

When I first built a ggplot object and then used ggplotly to convert it, I got this:
p <- ggplot(df, aes_string("cluster", "mCH", fill="cluster"))
p <- p + geom_boxplot() + scale_fill_manual(values = colors2, name="Cluster") + theme_bw() + ggtitle("someGene")
p.ggplotly <- p %>% ggplotly(tooltip = c("cluster", "mCH", "density")) %>% layout(legend = list(orientation = "h"))
p.ggplotly

The legend had only one row and it just got out of the margin... But when I did a similar thing using plot_ly the legend was fine:

p.plotly <- plot_ly(df,x=~cluster, y=~mCH,type="box", color=~cluster, colors=colors2) %>% layout(legend=list(orientation = "h"))
p.plotly

And I thought this is somehow control by the layout, so I tried copying the layout from p.plotly to p.ggplotly:
p.ggplotly.bak <- p.ggplotly
p.ggplotly$x$layout <- p.plotly$x$layout
But the legend was still truncated...
p.ggplotly

However if I did it the other way around, I got a something like this:
p.plotly$x$layout <- p.ggplotly.bak$x$layout
p.plotly

The legend was displayed correctly with multiple rows(even though the position was not right, but I know how to adjust for that).

I'm not sure if this is because the ggplotly and plot_ly is inherently different, or there're other options that can controll this behavior (or this is just a bug?). Any idea? I somehow prefer to use ggplotly...

FYI, ggplot2 version is 2.2.1 and plotly version is 4.5.6.9000. Thanks!

Reproduce the issue using the ggplot2 and plotly R examples in the report, comparing ggplotly conversion with the equivalent plot_ly chart and horizontal legend layout. Trace the conversion and rendering behavior to identify why the ggplotly legend is truncated, then verify that long legends display across multiple rows without breaking the reported plot behavior.