Esoteric bug with specific text

[teunbrand]

Hi Thomas,

This is a transfer of tidyverse/ggplot2#6539. Essentially, when trying to render the text 'Sr' and 'Fe', it renders 'Sr' twice instead of each once. In the reprex below, this is why we only see 'Sr' and not 'Fe' on top per the instructions. Other graphics device do not show this problem and so far the only text for which this bug observed is exactly 'Sr' and 'Fe'.

library(grid)

# Render as text grobs
vals <- c("Sr", "Fe")
grobs <- lapply(vals, textGrob)

# Confirm label is correct
grobs[[1]]$label
#> [1] "Sr"
grobs[[2]]$label
#> [1] "Fe"

# Create device for temp file
tmp <- tempfile(fileext = ".png")
ragg::agg_png(tmp)

# Draw the grobs
grid.newpage()
grid.draw(grobs[[1]])
grid.draw(grobs[[2]])

dev.off()
#> png 
#>   2
knitr::include_graphics(tmp)

^{Created on 2025-07-02 with reprex v2.1.1}

[jmclawson]

This is most likely related to what I was seeing in Positron for other text options besides “Sr” and “Fe”.

[thomasp85]

Thanks for this. Something must be messing up the caching and retrieval of shaped text. I'll look into it

[r2evans]

I just came across this in an admittedly-contrived example, but it sounds similar to a 2013 bug in Xerox scanners where caching of image segments (and pdf's use of JBIG2 decompression) was responsible for silently changing numbers on the scanned documents.

Clearly it's different (unless you're using jbig2 internally :-), but ... if labels or titles or some grobs are silently or inadvertently cached, I am concerned that artifacts produced by ragg could be poisoned with something else from the same R session. The only way to measure the frequency of occurrence requires quite a bit of manual labor, and since I use ggsave() dozens of times each week (with plots that contain lots of text grobs), I really do not relish the idea of having to visually check all for numerical correctness.

The example here was just the title, do you have confidence that this is "just" an issue with titles, subtitles, captions, or axes, and not with in-plot text grobs? Do you have an indication of where in the dependencies or code this is being introduced?

Thanks for all the effort getting here, and for trying to resolve the issue.

[thomasp85]

Fixed in textshaping