Error saving sce with metadata: Error in value[[3L]](cond) : failed to stage 'metadata(<SingleCellExperiment>)'

[whtns]

Hello, I am encountering an error (below) while saving an sce to alabaster due to inclusion of metadata. I get no error if I set metadata to an empty list. I am consulting OSCA for guidance on the use of object-level metadata.

library(SingleCellExperiment)
example(SingleCellExperiment, echo=FALSE) 
metadata(sce) <- list("experiment" = list("SingleCellExperiment_version" = packageVersion("SingleCellExperiment")))
library(alabaster.sce)
tmp <- tempfile()
saveObject(sce, tmp)

# Error in value[[3L]](cond) : 
#  failed to stage 'metadata(<SingleCellExperiment>)'
#  - failed to stage list element 'experiment'
#  - failed to stage list element 'SingleCellExperiment_version'
#  - failed to stage list element 1

[LTLA]

For now, can you coerce the packageVersion to a character string? saveObject() doesn't know how to deal with package_version class instances.

[whtns]

Thanks, will do. Is there any guideline for what kind of data can be stored and saved safely in metadata? Are tibbles alright, for example?

[LTLA]

The uzuki2 specification describes everything that can be saved in the list format. Anything that isn't supported is passed through to saveObject(), so as long as your object does something sensible with saveObject(), it should be fine.

For the package version: it is not inconceivable that we add a "format": "version" option to the specification. Probably too close to the current release to do so, though.

In the specific case of tibbles, I would imagine that these are treated as list objects, so it'll just get saved as a list of vectors. Probably best to coerce it to a DFrame object if you intend for it to be saved as a DF. We could also add a saveObject,data.frame-method to save it more directly as a DF, but note that it will be reloaded as a DFrame.

Edit: turns out past @LTLA had a bit more foresight than I thought. A data.frame inside the metadata will be auto-converted to a DFrame and saved correctly. As a tibble is just a data.frame, it should also end up being saved as a DFrame. That said, when you read it back in, it will be as a DFrame.

[whtns]

Thanks, converting to character works. I'll consult uzuki2 in the future