On Thu, 28 Sep 2023 23:46:45 +0800 Shu Fai Cheung <shufai.che...@gmail.com> wrote:
> In my personal work, I prefer using saveRDS() and loadRDS() as I > don't like the risk of overwriting anything in the global > environment. There's the load(file, e <- new.env()) idiom, but that's potentially a lot to type. Confusingly, ?save also says: >> For saving single R objects, ‘saveRDS()’ is mostly preferable to >> ‘save()’, notably because of the _functional_ nature of ‘readRDS()’, >> as opposed to ‘load()’. > The files produced by save > <http://127.0.0.1:18888/library/base/help/save> have a header > identifying the file type and so are better protected against > erroneous use." This header is also mentioned elsewhere in ?saveRDS: >> ‘save’ writes a single line header (typically ‘"RDXs\n"’) The difference between the save() header and the serialize() header is that the save() header is designed to be read independently from the machine running the code: it's exactly 5 bytes; some precisely defined combinations of those 5 bytes identify how the rest of the file should be interpreted (nowadays, it's likely either "XDR format version 2" or "XDR format version 3"), and the rest of them cause an error. The serialize() header does contain enough information describing it (there's the first byte choosing between ASCII/XDR/native binary and a number of encoded integers describing the format version and the version of R you need to parse it), but it's stored in terms of serialized objects, so if you cannot for some reason decode them properly, you won't be able to read the header. A little bit of Catch-22. > When will the problem mentioned in the warning occur? That is, when > will a file saved by saveRDS() not be read correctly? One example I can offer is when a dataset is saved using serialize(xdr = FALSE) (which is not reachable using saveRDS()). The resulting file format would be dependent on the native byte order of the CPU in your computer. (Nowadays it's really hard to encounter a CPU that doesn't use little-endian byte order, so this is doubly unlikely to happen in practice.) Both save() and saveRDS() set xdr = TRUE and convert the data to "network byte order" (big-endian) when saving and back - when loading. The warning is relatively fresh (May 2021). Perhaps Prof. Brian D. Ripley (who made that change) will be able to explain it better. -- Best regards, Ivan ______________________________________________ R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
