WISH: Simple class for files / pathnames

[HenrikBengtsson]

It would be useful to have a class for files (e.g. File) and basic functions for creating objects of such classes. I can imagine that this file class is a simple extension of character, because that is currently how files/pathnames are currently represented, e.g.

File <- function(...) { x <- c(...); structure(x, class=c("File", class(x))) }

Then existing functions returning files/pathnames, e.g.

dir <- function(...) File(base::dir(...))

and instead of c() users can have, say, p() for very brief syntax, e.g. p <- File.

Examples:

> pathname <- p("R/zzz.R")
> pathnames <- p("R/000.R", "R/zzz.R")
> pathnames <- dir("R/")

[mtmorgan]

• 'isa' is not such a good model here, e.g., R's connections are not character vectors.
• presumably one would like to transparently manage a collection of files, so File() has length(), [ and names() methods
• It would be helpful to define the API, e.g., open(), close(), isOpen() and it's relation to the existing connection API.
• Finalizers seem important
• a 'registry' of file types seems useful, to map extensions (e.g., '.csv') to appropriate subclasses.

[lawremi]

Need to keep the notions of path (File) and stream (connection) distinct. We open() a path to get a stream. There would only be close() and isOpen() methods for the stream.

Also, the path represents a low-level format, but the stream might have richer semantics. For example, an HDF5 file might be represented by an HDF5File object. Lets say someone stored genotypes in the HDF5 file. The open() function might return an object that implements the same API as the VCF class. We could have a dual-dispatch openWithScheme generic that dispatches on the path and some object representing the VCF scheme.

[mtmorgan]

@lawremi Path versus stream is a useful distinction; sometimes its useful to create a stream without opening it, or to have a stream that exists in a state where it can be opened after being closed. Hence open() on the stream.

[leeper]

And what potentially intrigues me would be filetype-specific classes that would allow method dispatch. I'm mainly think about this for data import/export (in the context of rio), but I can imagine there might be other contexts, such as calling graphics devices based on file type without the need for jpeg(), png(), etc.