podup 2.1.0

Translate and run docker-compose files on rootless Podman
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44

//! Interactive `exec`: a pseudo-terminal and a live stdin, on Unix.
//!
//! `podup exec -it db psql` is the most-used shell-into-a-container habit in
//! compose, and it did not work: podup allocated no TTY and attached no stdin,
//! so `-i` set a flag nothing read and `-T` disabled something that never
//! existed (#1079).
//!
//! Unix only for now. podup reaches `podman machine` over a named pipe on
//! Windows, and both raw mode and resize events there are a different API — two
//! implementations, not one behind a `cfg`. The non-interactive path is
//! unchanged everywhere, so `podup exec` in a script behaves exactly as before
//! on every platform.

#![cfg(unix)]

use crate::error::{ComposeError, Result};
use crate::libpod::{urlencoded, API_PREFIX};

use super::Engine;

impl Engine {
	/// Run an exec session with a pty and a live stdin.
	///
	/// The terminal is restored by `RawMode`'s `Drop`, so every exit path —
	/// command exits, socket dies, `?` on an unrelated error, panic — leaves the
	/// caller's shell usable. That is the one thing this must not get wrong: a
	/// terminal left raw has no echo and no line discipline, and the user cannot
	/// even see what they type to fix it.
	pub(super) async fn exec_interactive(&self, exec_id: &str) -> Result<()> {
		let start_path = format!("{API_PREFIX}/exec/{}/start", urlencoded(exec_id));
		let body = br#"{"Detach":false,"Tty":true}"#;
		let hijacked = self
			.client
			.post_hijack(&start_path, body)
			.await
			.map_err(ComposeError::Podman)?;

		// Raw mode, sizing and the byte pump all live in `terminal_pump`, shared
		// with interactive `run` so the loop that must not get raw mode wrong
		// exists once.
		self.pump_terminal(hijacked, &format!("exec/{}", urlencoded(exec_id)))
			.await
	}
}