Hiroaki Yutani

> a histogram plot containing two groups with a log10 transformed y-axis. I cannot see any valid use cases when a log-scale bar chart can be useful... What plot did...

Thanks. While I'm not sure if this can be fixed, it makes sense to some extent. Btw, could you please render your example with the reprex package? (c.f. https://www.tidyverse.org/help/)

Thanks. Honestly, I'm not sure if this is the best way to record the dropped columns. Any suggestion is welcome.

I think the idea of using an env as a hashmap is not so uncommon. I came up with several ideas including yours, and I just couldn't figure out which...

Thanks, makes sense. I'll use the first one. Another one I attempted was returning a list rather than a data.frame (and I forgot to remove `result

> What is `unique0()`? It's defined here. After the vctrs integration (#4868), most of the code uses `unique0()`, so I followed it. https://github.com/tidyverse/ggplot2/blob/c86ed7bc59060d94fb070fe1bb1574e8b6026ff8/R/utilities.r#L355 > Would it make sense to add...

Added tests. > 2\. What happens when different columns get dropped in different groups. How does the data frame get assembled again? What do you mean by "what happens" here?...

> But now looking it up, it seems like `vec_rbind()` can handle this scenario, so I guess that's why it works. Yes, that's why we see this kind of incomplete...

I added many comments and rewrote the code a bit. I hope this will make it a bit easier to understand the logic...

Thanks so much for reviewing this complicated pull request! I hope I can do better next time...!