Read host target triple from rustc -vV

[messense]

Fixes #478, closes #486

cc @al626

[konstin]

I think the best solution for both maturin build --target xxx --universal2 from #429 and the musl handling from #478 would be reverting #429 and manually failing on the --target xxx --universal2 case. I.e. I'd prefer #486 and add an extra error path. This also has the advantage that we save a subprocess call.

If there are other reasons I've missing and we still need to go for reading the target triple, I'd do what cargo does and read rustc -vV:

use std::process::Command;

use anyhow::{format_err, Context, Result};
use std::str;

fn get_target() -> Result<String> {
    let output = Command::new("rustc")
        .arg("-vV")
        .output()
        .context("Failed to run rustc to get the host target")?;
    let output = str::from_utf8(&output.stdout).context("`rustc -vV` didn't return utf8 output")?;

    let field = "host: ";
    let host = output
        .lines()
        .find(|l| l.starts_with(field))
        .map(|l| &l[field.len()..])
        .ok_or_else(|| {
            format_err!(
                "`rustc -vV` didn't have a line for `{}`, got:\n{}",
                field.trim(),
                output
            )
        })?
        .to_string();
    Ok(host)
}

fn main() -> Result<()> {
    let host = get_target()?;
    println!("target triple: {}", host);
    Ok(())
}

[messense]

There is also #423 affected by target_env

[messense]

Anyway I think it's good to get rid of Platform::guess_current(). Reading the host target triple from rustc is a better solution than using inaccuratePlatform::guess_current() IMO.

[konstin]

Since we're only shipping musl anyway, wouldn't it make more sense to only run the tests for musl?

[messense]

Just in case someone needs to compile maturin from source instead of using prebuilt wheels, for example on some BSD systems?

[konstin]

Looks good, thank you!