Histograms are different between ggplot 3.5.2 and rc/4.0.0 on Windows

[rogerssam]

Histograms are (visually) substantially different between the current released version 3.5.2 and the version in the 4.0.0 release candidate (3.5.2.9002, commit ce8d250). I'm not entirely sure where the issue comes from, but possibly the data is being binned differently? I believe this is what's causing revdepcheck failures in my package biometryassist.

Note this issue only appears on Windows. I have verified this appearing across multiple computers under Windows, but is fine on macOS or Ubuntu.

Here is in minimal reprex:

# Version 3.5.2
ggplot(iris, aes(Sepal.Length)) + geom_histogram()

# Version 3.5.2.9002
ggplot(iris, aes(Sepal.Length)) + geom_histogram()

See for example values around x = 5.

[teunbrand]

Thanks for the report! This is announced in the news file as a breaking change:

ggplot2/NEWS.md

Lines 28 to 30 in bed8e32

	* In binning stats, the default `boundary` is now chosen to better adhere to
	the `nbin` argument. This may affect plots that use default binning
	(@teunbrand, #5882, #5036)

I'd suggest the solution to accept any new snapshots or update tests that are sensitive to how default breaks are computed, or fix the breaks argument with manual values.

[rogerssam]

Ah, thanks for the clarification. I will have a look at the snapshots and update, or change the function to use manual breaks.

[rogerssam]

Just letting you know that I've just had an updated version of biometryassist accepted by CRAN. Hopefully that will resolve any remaining issues on my part preventing the release of the next version of ggplot2.

[teunbrand]

Great, thank you!