Inconsistent treatment of units in st_as_sf()

[fgoerlich]

Hi,
H1084 is an SpatRaster. Executing

H1084 |>
  crop(filter(MTN50grid, MTN50 == "1084-1086")) |>
  st_as_stars() |>
  st_xy2sfc(as_points = TRUE) |>
  st_as_sf() |>
  rename(H1084 = 1)

produces numeric, without units, output

Simple feature collection with 155 features and 1 field
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 607385 ymin: 3180900 xmax: 617162.5 ymax: 3182725
Projected CRS: ETRS89 / UTM zone 28N
First 10 features:
   H1084                 geometry
1    4.1   POINT (615690 3182520)
2    2.8 POINT (615692.5 3182520)
3    4.0   POINT (615685 3182518)
4    3.9 POINT (615687.5 3182518)
5    4.0   POINT (615690 3182518)
6    4.1 POINT (615692.5 3182518)
7    0.5 POINT (615697.5 3182518)
8    4.3 POINT (615682.5 3182515)
9    4.0   POINT (615685 3182515)
10   3.8 POINT (615687.5 3182515)

Without the cropping (from terra) instruction,

H0184 |>
  st_as_stars() |>
  st_xy2sfc(as_points = TRUE) |>
  st_as_sf() |>
  rename(H1084 = 1)

I get numeric, with units, output

Simple feature collection with 238404 features and 1 field
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 484497.5 ymin: 4686732 xmax: 512062.5 ymax: 4705362
Projected CRS: ETRS89 / UTM zone 29N
First 10 features:
       H1084                 geometry
1  4.735 [m]   POINT (510505 4705295)
2  5.330 [m] POINT (510507.5 4705295)
3  2.100 [m] POINT (510492.5 4705293)
4  6.137 [m]   POINT (510505 4705293)
5  6.900 [m] POINT (510507.5 4705293)
6  2.723 [m]   POINT (510490 4705290)
7  2.005 [m] POINT (510492.5 4705290)
8  2.789 [m]   POINT (510450 4705288)
9  4.000 [m] POINT (510452.5 4705288)
10 5.750 [m] POINT (510502.5 4705288)

I wonder what makes the difference, since croping produces an SpatRaster also.
PS: I can provide the data if necessary.
Thanks

[edzer]

Speculating: H1084 is a pretty large file, which terra doesn't read into memory. On convertion to stars, stars gets to read the file and figures out it has units (while, for instance, it is a NetCDF file or so). If terra crops this, it reads a section of the file, and ignores the units. So the question is: is this a stars issue or a terra issue?

[fgoerlich]

Probably you are rigth.
If I read H1084, which is a raster (tif) object, as star object, I always get units.
stars reads the units, but apparently terra does not. I am not sure if this is a stars issue or a terra issue.

Find the data an small example.

Units.zip

[edzer]

Confirmed.