I was happy to discover that read_stars supports reading any file as stars_proxy object, significantly easing my memory constraints (quite the contrary occured to my surprise, see #708)! However, st_bbox() when applied via st_set_bbox() to a stars_proxy does not outlast the conversion into a stars object, e.g.
library(stars)
fl = system.file("tif/L7_ETMs.tif", package = "stars")
rs = read_stars(fl, proxy = TRUE)
rs2 = rs |>
st_set_bbox(value = st_bbox(rs)*2) |>
st_as_stars()
# The two bboxes should be different, i.e. TRUE TRUE TRUE TRUE
st_bbox(rs) != st_bbox(rs2)
This entails problems for intermediate down-sampled output such as plots as well as repetitive statements after each obtained stars objects if the st_bbox of the raw data is not set correctly.