I found it took a while to get to the point.
In the end, I actually agree with this! I have also been thinking about filesystems, which are trees of of dictionaries (directories) and blobs (files), and that there are many other examples of tree-like data. Data structures in our programs are tree shaped, perhaps with references/pointers to other parts of the tree. JSON is a tree of dictionaries (objects), arrays, and data (the primitive string/number/null). The arrays are ordered, much like the content inside a HTML/XML block.
I agree that adding an "array" style of directory to our file systems would be really cool. I've been toying with the idea of writing a FUSE driver that holds some structured data (possibly including arrays) and just converts the (integer) index into a string. The idea is that you could e.g. view and edit some JSON tree with the file explorer. And not just JSON - basically any piece of structured data that we have in our programs can be "viewed" as a FS this way (e.g. just convert structs into directories of fields). It could even be a pretty cool and universal debugger - a breakpoint could make the program pause and serve a FUSE driver and later continue when it is unmounted :)
And yes, exactly what follows from this is some program could be written to open some "directory" and render a "document" based on the contents. The filesystem supports links, so we have the "web" like experience.
The "document" angle does require adding a kind of directory with ordered array semantics rather than dictionary/map semantics. It's the first missing ingredient listed in the article. Though some filesystems use sorted dictionaries (b-trees or whatever) for directory maps so you could maybe hack this ordered semantics in that way.
The second missing ingredient listed in the article is the hypermedia part. I mean my computer is actually OK at inferring if a file is a movie or photo or text document, so we kind of have a way of dealing with that, too. The "blob of bytes is a narrow waste" thing is quite powerful. That said, sum types could be useful to demark different kinds of "stuff", and there's no reason an implementation of this idea couldn't support sum types as part of its data model.
In the end, I actually agree with this! I have also been thinking about filesystems, which are trees of of dictionaries (directories) and blobs (files), and that there are many other examples of tree-like data. Data structures in our programs are tree shaped, perhaps with references/pointers to other parts of the tree. JSON is a tree of dictionaries (objects), arrays, and data (the primitive string/number/null). The arrays are ordered, much like the content inside a HTML/XML block.
I agree that adding an "array" style of directory to our file systems would be really cool. I've been toying with the idea of writing a FUSE driver that holds some structured data (possibly including arrays) and just converts the (integer) index into a string. The idea is that you could e.g. view and edit some JSON tree with the file explorer. And not just JSON - basically any piece of structured data that we have in our programs can be "viewed" as a FS this way (e.g. just convert structs into directories of fields). It could even be a pretty cool and universal debugger - a breakpoint could make the program pause and serve a FUSE driver and later continue when it is unmounted :)
And yes, exactly what follows from this is some program could be written to open some "directory" and render a "document" based on the contents. The filesystem supports links, so we have the "web" like experience.
The "document" angle does require adding a kind of directory with ordered array semantics rather than dictionary/map semantics. It's the first missing ingredient listed in the article. Though some filesystems use sorted dictionaries (b-trees or whatever) for directory maps so you could maybe hack this ordered semantics in that way.
The second missing ingredient listed in the article is the hypermedia part. I mean my computer is actually OK at inferring if a file is a movie or photo or text document, so we kind of have a way of dealing with that, too. The "blob of bytes is a narrow waste" thing is quite powerful. That said, sum types could be useful to demark different kinds of "stuff", and there's no reason an implementation of this idea couldn't support sum types as part of its data model.