Deserializing to OsString with serde

[jwodder]

Serde's documentation says that it supports deserializing to OsString, albeit in some sort of indirect way that I'm not clear how to use. Trying to deserialize to OsString naïvely with:

use serde::{Serialize, Deserialize};
use std::error::Error;
use std::ffi::OsString;
use std::path::PathBuf;

#[derive(Serialize, Deserialize, Debug)]
struct Config {
    #[serde(default)]
    dirpaths: Vec<PathBuf>,

    #[serde(default)]
    filenames: Vec<OsString>,
}

fn main() -> Result<(), Box<dyn Error>> {
    let input = r#"{
    "dirpaths": ["/usr/src", "/home/me"],
    "filenames": ["foo", "bar"],
}"#;
    let cfg: Config = serde_json::from_str(input)?;
    println!("{cfg:#?}");
    Ok(())
}

fails with:

Error: Error("unknown variant `foo`, expected `Unix` or `Windows`", line: 3, column: 23)

What exactly do I need to do to be able to deserialize to an OsString?

[quinedot]

First a clarification: the docs you linked to are just an import of std docs/types.

OsString is an enum in the model, and this works:

    let input = r#"{
        "dirpaths": ["/usr/src", "/home/me"],
        "filenames": [{ "Unix": [102, 111, 111] }, { "Unix": [98, 97, 114] }]
    }"#;

PathBuf will only actually work with paths that could be Strings, which is quite dishonest IMO. (Supporting non-UTF8 paths is a main motivator for supporting OsStrings.)

n.b. I'm not a serde expert and there may be a better way.

[jwodder]

I'm not about to change the input format for my program to look like {"Unix": [102, 111, 111]}. So far, the only way I can see to get strings in JSON or TOML to deserialize to OsStrings is to use an intermediate struct with String fields and an impl From<IntermediateConfig> for Config implementation plus serde's #[serde(from = "IntermediateConfig")]. Does anyone know of a better way?

[Michael-F-Bryan]

@jwodder

Does anyone know of a better way?

That's what I would do. Almost all JSON strings are encoded as UTF-8 anyway, so it's not really possible for your end users to put a non-UTF-8 string in the file unless they want to write the bytes out as an array of numbers. If you want to support both, then I would probably create my own serialize() and deserialize() implementations using the #[serde(with = ...)] attribute that fits what you want.