I'm wondering what the best practices are for handling the descriptorfile (foo.desc
) and backingfile (foo.bin
) that are created when reading foo.csv
using read.big.matrix(...)
As far as I can find out the backingfile is an RDS file by default, and the descriptorfile is a plain text file. So in theory those are both cross-platform and could be checked in, but my guess is that the backingfile will change a lot and should be ignored by version control. Does anyone have experience here?
The backingfile is not an RDS file. It is just a simple binary file and should be cross-platform.
It is generally not recommended to version binary files, because they change a lot. Moreover, this is a large file and so it's even less recommended.
The descriptor file is useless by itself.
So, I would recommend to not version those files. The only case would if it's only small files that you use for examples (see e.g. https://github.com/privefl/bigstatsr/tree/master/inst/extdata).