st_perimeter works in R, but not with knitr

[josephholler]

I've got the st_perimeter() function working fine in RStudio running Rmarkdown chunks one by one, but when I try to knit the Rmd file to html using Knitr, it throws an error:

error in `st_perimeter_lwgeom()`:
! for perimeter of longlat geometry, cast to LINESTRING and use st_length

I'm using lwgeom_0.2-14 on Windows and the CRS of the simple features is epsg:4269.

The work-around to cast to LINESTRING and use st_length does work, and gives the same result as st_perimeter() run from the code chunk.
I'm mostly curious why st_perimeter() might work while running code chunks, but not while knitting?

[edzer]

The only think I can think of is a difference in package versions between both R sessions; can you run an update.packages() and try both again?

[josephholler]

I am running both environments with the same groundhog code to load packages as of 2025-02-01 and confirm with sessionInfo() that I'm using lwgeom_0.2-14 in both instances.

I'm stumped on this one. It doesn't matter much at the end of the day because st_length() of the polygons cast to MULTILINESTRING does work.

[edzer]

My guess (based on known and unresolved problems in renv) is that this is caused by groundhog, which may work with pure R packages but not work well with packages that have system requirements (such as sf and lwgeom).

[josephholler]

Thanks for your thoughts. After testing a bit more with a simplified case, these two work while knitting with groundhog:

• st_length(st_cast(geom, "MULTILINESTRING"))
• sf::st_perimeter(geom)

While these throw the error above about using st_length():

• lwgeom::st_perimeter(geom)
• st_perimeter(geom)

So, when I have sf and lwgeom both loaded, the lwgeom version of the function seems to take precedence and stops the knitting with an error, while the sf version works.

Outside of knitting with the same code & environment, lwgeom::st_perimeter(geom) gives the error while st_perimeter(geom) works.

Using environment() I can confirm that while running code blocks manually, st_perimeter is in the sf namespace, but while knitting, it's in the lwgeom namespace.

These were the package versions:
lwgeom_0.2-14
sf_1.0-19

Feel free to close this, unless there's something you can do on your end to manage the conflict between sf and lwgeom when both are loaded.