Fill and color guides for factor with drop=FALSE don't show color.

[ltierney]

In ggplot2 3.5.1 the fill guide for a factor where not all level sare present but drop=FALSE is used shows the label for the missing level but not the color. A simple example:

library(ggplot2)
d <-data.frame(x = factor(1:3, levels = 1:4), y = 1:3)
ggplot(d, aes(x, y, fill = x)) + 
    geom_col() +
    scale_fill_discrete(drop = FALSE)

This was different in the version available a year ago. Compare
this from 2024 to this from 2023.

[teunbrand]

Thanks for the report! This appears to be a duplicate of #5728. The summary of that issue is: we have no intention of changing this behaviour.

[ltierney]

It's a pretty clear regression, something I try to avoid in software I write and maintain. But it's your software and your call.

[teunbrand]

I don't think of it as a regression but as trade-off that trades the complex problem of having to tailor guide_legend(override.aes) for the mild inconvenience of having to put show.legend = TRUE.

[ltierney]

When existing working code no longer works correctly after a change that is, by definition, a regression. Something we try pretty hard to avoid in maintaining R — not that we always succeed.

Thanks for the show.legend = TRUE hint; putting that in the right places works for me. I'm glad you told me since there is no way I can see I could have figured out from the documentation of show.legend in ?geom_polygon or ?geom_sf that that would give me what I needed.

[teunbrand]

When existing working code no longer works correctly after a change

Argueably, the code was working incorrectly before by showing legend keys that weren't present in the data. I get that this might be intended, but this intent has to be explicit now.

there is no way I can see I could have figured out from the documentation

As mentioned in #5728, it was displayed in the changelog, described in the release blog and we've since included it in the documentation of the drop argument.

[ltierney]

When existing working code no longer works correctly after a change

Argueably, the code was working incorrectly before by showing legend keys that weren't present in the data. I get that this might be intended, but this intent has to be explicit now.''

We sometimes try to make that argument as well; doesn't usually work very well for us either :-).

there is no way I can see I could have figured out from the documentation

As mentioned in #5728, it was displayed in the changelog, described in the release blog and we've since included it in the documentation of the drop argument.

That looks good. might be worth having a cross-reference from the show.legend docs as well.

The current behavior with just drop = FALSE still seems very odd to me. The guide does include all labels, just not all colors. It may make sense with the internal design, but still seems odd.

In any case I think we've both spent enough time on this.

[teunbrand]

That looks good. might be worth having a cross-reference from the show.legend docs as well.

Thanks for this suggestion, I'll reopen to remind us to document this.