rust/cargo-miri/bin.rs

549 lines
21 KiB
Rust
Raw Normal View History

2020-04-23 10:46:49 -05:00
use std::env;
2020-05-21 04:32:12 -05:00
use std::ffi::OsString;
2019-02-15 19:29:38 -06:00
use std::fs::{self, File};
use std::io::{self, BufRead, BufReader, BufWriter, Write};
2019-08-04 03:14:51 -05:00
use std::ops::Not;
2019-12-23 05:56:23 -06:00
use std::path::{Path, PathBuf};
use std::process::Command;
use std::collections::HashMap;
use serde::{Deserialize, Serialize};
use rustc_version::VersionMeta;
2020-07-28 07:00:33 -05:00
const XARGO_MIN_VERSION: (u32, u32, u32) = (0, 3, 22);
2019-02-07 06:00:42 -06:00
const CARGO_MIRI_HELP: &str = r#"Interprets bin crates and tests in Miri
Usage:
cargo miri [subcommand] [<cargo options>...] [--] [<miri options>...] [--] [<program/test suite options>...]
Subcommands:
run Run binaries (default)
test Run tests
setup Only perform automatic setup, but without asking questions (for getting a proper libstd)
Common options:
-h, --help Print this message
--features Features to compile for the package
-V, --version Print version info and exit
2020-02-24 07:17:19 -06:00
Other [options] are the same as `cargo check`. Everything after the first "--" is
passed verbatim to Miri, which will pass everything after the second "--" verbatim
to the interpreted program.
Examples:
cargo miri run -- -Zmiri-disable-stacked-borrows
cargo miri test -- -- test-suite-filter
"#;
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
enum MiriCommand {
Run,
Test,
Setup,
}
/// The inforamtion Miri needs to run a crate. Stored as JSON when the crate is "compiled".
#[derive(Serialize, Deserialize)]
struct CrateRunInfo {
/// The command-line arguments.
args: Vec<OsString>,
/// The environment.
env: HashMap<OsString, OsString>,
}
fn show_help() {
println!("{}", CARGO_MIRI_HELP);
}
fn show_version() {
2019-12-23 05:56:23 -06:00
println!(
"miri {} ({} {})",
env!("CARGO_PKG_VERSION"),
env!("VERGEN_SHA_SHORT"),
env!("VERGEN_COMMIT_DATE")
);
}
fn show_error(msg: String) -> ! {
eprintln!("fatal error: {}", msg);
std::process::exit(1)
}
2019-02-15 19:29:38 -06:00
// Determines whether a `--flag` is present.
fn has_arg_flag(name: &str) -> bool {
let mut args = std::env::args().take_while(|val| val != "--");
args.any(|val| val == name)
}
2019-02-15 19:29:38 -06:00
/// Gets the value of a `--flag`.
fn get_arg_flag_value(name: &str) -> Option<String> {
2019-02-15 19:29:38 -06:00
// Stop searching at `--`.
let mut args = std::env::args().take_while(|val| val != "--");
loop {
let arg = match args.next() {
Some(arg) => arg,
None => return None,
};
if !arg.starts_with(name) {
continue;
}
2019-02-15 19:29:38 -06:00
// Strip leading `name`.
let suffix = &arg[name.len()..];
if suffix.is_empty() {
2019-02-15 19:29:38 -06:00
// This argument is exactly `name`; the next one is the value.
return args.next();
} else if suffix.starts_with('=') {
2019-02-15 19:29:38 -06:00
// This argument is `name=value`; get the value.
// Strip leading `=`.
return Some(suffix[1..].to_owned());
}
}
}
/// Returns the path to the `miri` binary
fn find_miri() -> PathBuf {
if let Some(path) = env::var_os("MIRI") {
return path.into();
}
2020-02-24 07:17:19 -06:00
let mut path = std::env::current_exe().expect("current executable path invalid");
path.set_file_name("miri");
path
}
fn miri() -> Command {
Command::new(find_miri())
}
fn version_info() -> VersionMeta {
VersionMeta::for_command(miri()).expect("failed to determine underlying rustc version of Miri")
}
2020-02-24 07:17:19 -06:00
fn cargo() -> Command {
2020-05-09 04:52:26 -05:00
Command::new(env::var_os("CARGO").unwrap_or_else(|| OsString::from("cargo")))
}
2020-03-02 15:10:48 -06:00
fn xargo_check() -> Command {
2020-05-09 04:52:26 -05:00
Command::new(env::var_os("XARGO_CHECK").unwrap_or_else(|| OsString::from("xargo-check")))
}
2018-12-15 08:08:03 -06:00
fn xargo_version() -> Option<(u32, u32, u32)> {
2020-03-02 15:10:48 -06:00
let out = xargo_check().arg("--version").output().ok()?;
2018-12-15 08:08:03 -06:00
if !out.status.success() {
return None;
}
// Parse output. The first line looks like "xargo 0.3.12 (b004f1c 2018-12-13)".
2019-12-23 05:56:23 -06:00
let line = out
.stderr
.lines()
.nth(0)
2018-12-15 08:08:03 -06:00
.expect("malformed `xargo --version` output: not at least one line")
.expect("malformed `xargo --version` output: error reading first line");
let (name, version) = {
let mut split = line.split(' ');
2019-12-23 05:56:23 -06:00
(
split.next().expect("malformed `xargo --version` output: empty"),
split.next().expect("malformed `xargo --version` output: not at least two words"),
)
};
2019-05-10 01:05:34 -05:00
if name != "xargo" {
// This is some fork of xargo
return None;
}
2018-12-15 08:08:03 -06:00
let mut version_pieces = version.split('.');
2019-12-23 05:56:23 -06:00
let major = version_pieces
.next()
2018-12-15 08:08:03 -06:00
.expect("malformed `xargo --version` output: not a major version piece")
.parse()
.expect("malformed `xargo --version` output: major version is not an integer");
2019-12-23 05:56:23 -06:00
let minor = version_pieces
.next()
2018-12-15 08:08:03 -06:00
.expect("malformed `xargo --version` output: not a minor version piece")
.parse()
.expect("malformed `xargo --version` output: minor version is not an integer");
2019-12-23 05:56:23 -06:00
let patch = version_pieces
.next()
2018-12-15 08:08:03 -06:00
.expect("malformed `xargo --version` output: not a patch version piece")
.parse()
.expect("malformed `xargo --version` output: patch version is not an integer");
if !version_pieces.next().is_none() {
panic!("malformed `xargo --version` output: more than three pieces in version");
}
Some((major, minor, patch))
}
fn ask_to_run(mut cmd: Command, ask: bool, text: &str) {
// Disable interactive prompts in CI (GitHub Actions, Travis, AppVeyor, etc).
2020-09-13 14:10:29 -05:00
// Azure doesn't set `CI` though (nothing to see here, just Microsoft being Microsoft),
// so we also check their `TF_BUILD`.
let is_ci = env::var_os("CI").is_some() || env::var_os("TF_BUILD").is_some();
if ask && !is_ci {
let mut buf = String::new();
print!("I will run `{:?}` to {}. Proceed? [Y/n] ", cmd, text);
io::stdout().flush().unwrap();
io::stdin().read_line(&mut buf).unwrap();
match buf.trim().to_lowercase().as_ref() {
// Proceed.
2019-12-23 05:56:23 -06:00
"" | "y" | "yes" => {}
"n" | "no" => show_error(format!("Aborting as per your request")),
2019-12-23 05:56:23 -06:00
a => show_error(format!("I do not understand `{}`", a)),
};
} else {
println!("Running `{:?}` to {}.", cmd, text);
}
2019-12-23 05:56:23 -06:00
if cmd.status().expect(&format!("failed to execute {:?}", cmd)).success().not() {
show_error(format!("Failed to {}", text));
}
}
2019-02-15 19:29:38 -06:00
/// Performs the setup required to make `cargo miri` work: Getting a custom-built libstd. Then sets
/// `MIRI_SYSROOT`. Skipped if `MIRI_SYSROOT` is already set, in which case we expect the user has
/// done all this already.
fn setup(subcommand: MiriCommand) {
if std::env::var_os("MIRI_SYSROOT").is_some() {
if subcommand == MiriCommand::Setup {
println!("WARNING: MIRI_SYSROOT already set, not doing anything.")
}
return;
}
// Subcommands other than `setup` will do a setup if necessary, but
// interactively confirm first.
let ask_user = subcommand != MiriCommand::Setup;
2019-02-15 19:29:38 -06:00
// First, we need xargo.
if xargo_version().map_or(true, |v| v < XARGO_MIN_VERSION) {
if std::env::var_os("XARGO_CHECK").is_some() {
// The user manually gave us a xargo binary; don't do anything automatically.
show_error(format!("Your xargo is too old; please upgrade to the latest version"))
}
let mut cmd = cargo();
cmd.args(&["install", "xargo", "-f"]);
ask_to_run(cmd, ask_user, "install a recent enough xargo");
}
// Determine where the rust sources are located. `XARGO_RUST_SRC` env var trumps everything.
let rust_src = match std::env::var_os("XARGO_RUST_SRC") {
Some(path) => {
2020-07-11 04:07:17 -05:00
let path = PathBuf::from(path);
// Make path absolute if possible.
path.canonicalize().unwrap_or(path)
2020-07-05 13:01:12 -05:00
}
None => {
// Check for `rust-src` rustup component.
let sysroot = miri()
2019-12-23 05:56:23 -06:00
.args(&["--print", "sysroot"])
.output()
.expect("failed to determine sysroot")
.stdout;
let sysroot = std::str::from_utf8(&sysroot).unwrap();
let sysroot = Path::new(sysroot.trim_end_matches('\n'));
2020-07-28 07:00:33 -05:00
// Check for `$SYSROOT/lib/rustlib/src/rust/library`; test if that contains `std/Cargo.toml`.
2020-05-21 04:32:12 -05:00
let rustup_src =
2020-07-28 07:00:33 -05:00
sysroot.join("lib").join("rustlib").join("src").join("rust").join("library");
if !rustup_src.join("std").join("Cargo.toml").exists() {
// Ask the user to install the `rust-src` component, and use that.
let mut cmd = Command::new("rustup");
cmd.args(&["component", "add", "rust-src"]);
ask_to_run(
cmd,
ask_user,
"install the `rust-src` component for the selected toolchain",
);
}
rustup_src
}
};
if !rust_src.exists() {
show_error(format!("Given Rust source directory `{}` does not exist.", rust_src.display()));
}
// Next, we need our own libstd. Prepare a xargo project for that purpose.
// We will do this work in whatever is a good cache dir for this platform.
2019-06-20 12:45:39 -05:00
let dirs = directories::ProjectDirs::from("org", "rust-lang", "miri").unwrap();
2018-11-27 04:26:53 -06:00
let dir = dirs.cache_dir();
if !dir.exists() {
2018-11-27 04:43:02 -06:00
fs::create_dir_all(&dir).unwrap();
}
// The interesting bit: Xargo.toml
2019-12-23 05:56:23 -06:00
File::create(dir.join("Xargo.toml"))
.unwrap()
.write_all(
br#"
[dependencies.std]
2018-12-02 07:03:29 -06:00
default_features = false
# We need the `panic_unwind` feature because we use the `unwind` panic strategy.
# Using `abort` works for libstd, but then libtest will not compile.
2019-09-16 15:22:54 -05:00
features = ["panic_unwind"]
[dependencies.test]
2020-03-01 04:53:34 -06:00
"#,
2019-12-23 05:56:23 -06:00
)
.unwrap();
2019-02-15 19:29:38 -06:00
// The boring bits: a dummy project for xargo.
2020-02-24 07:17:19 -06:00
// FIXME: With xargo-check, can we avoid doing this?
2019-12-23 05:56:23 -06:00
File::create(dir.join("Cargo.toml"))
.unwrap()
.write_all(
br#"
[package]
name = "miri-xargo"
description = "A dummy project for building libstd with xargo."
version = "0.0.0"
[lib]
path = "lib.rs"
2020-03-01 04:53:34 -06:00
"#,
2019-12-23 05:56:23 -06:00
)
.unwrap();
File::create(dir.join("lib.rs")).unwrap();
// Determine architectures.
// We always need to set a target so rustc bootstrap can tell apart host from target crates.
let host = version_info().host;
2018-12-10 02:32:54 -06:00
let target = get_arg_flag_value("--target");
let target = target.as_ref().unwrap_or(&host);
// Now invoke xargo.
2020-03-02 15:10:48 -06:00
let mut command = xargo_check();
command.arg("check").arg("-q");
command.arg("--target").arg(target);
command.current_dir(&dir);
command.env("XARGO_HOME", &dir);
command.env("XARGO_RUST_SRC", &rust_src);
// Use Miri as rustc to build a libstd compatible with us (and use the right flags).
// However, when we are running in bootstrap, we cannot just overwrite `RUSTC`,
// because we still need bootstrap to distinguish between host and target crates.
// In that case we overwrite `RUSTC_REAL` instead which determines the rustc used
// for target crates.
if env::var_os("RUSTC_STAGE").is_some() {
command.env("RUSTC_REAL", find_miri());
} else {
command.env("RUSTC", find_miri());
2018-12-10 02:32:54 -06:00
}
command.env("MIRI_BE_RUSTC", "1");
// Make sure there are no other wrappers or flags getting in our way
// (Cc https://github.com/rust-lang/miri/issues/1421).
// This is consistent with normal `cargo build` that does not apply `RUSTFLAGS`
// to the sysroot either.
command.env_remove("RUSTC_WRAPPER");
command.env_remove("RUSTFLAGS");
// Finally run it!
2019-12-23 05:56:23 -06:00
if command.status().expect("failed to run xargo").success().not() {
show_error(format!("Failed to run xargo"));
}
2019-02-15 19:29:38 -06:00
// That should be it! But we need to figure out where xargo built stuff.
2018-12-10 02:32:54 -06:00
// Unfortunately, it puts things into a different directory when the
// architecture matches the host.
let sysroot = if target == &host { dir.join("HOST") } else { PathBuf::from(dir) };
std::env::set_var("MIRI_SYSROOT", &sysroot); // pass the env var to the processes we spawn, which will turn it into "--sysroot" flags
// Figure out what to print.
2020-05-21 04:32:12 -05:00
let print_sysroot = subcommand == MiriCommand::Setup && has_arg_flag("--print-sysroot"); // whether we just print the sysroot path
2019-10-19 09:36:45 -05:00
if print_sysroot {
// Print just the sysroot and nothing else; this way we do not need any escaping.
println!("{}", sysroot.display());
} else if subcommand == MiriCommand::Setup {
println!("A libstd for Miri is now available in `{}`.", sysroot.display());
}
}
fn phase_cargo_miri(mut args: env::Args) {
// Require a subcommand before any flags.
// We cannot know which of those flags take arguments and which do not,
// so we cannot detect subcommands later.
let subcommand = match args.next().as_deref() {
Some("test") => MiriCommand::Test,
Some("run") => MiriCommand::Run,
Some("setup") => MiriCommand::Setup,
2019-02-15 19:29:38 -06:00
// Invalid command.
None => show_error(format!("`cargo miri` must be immediately followed by `test`, `run`, or `setup`.")),
Some(s) => show_error(format!("unknown command `{}`", s)),
};
let verbose = has_arg_flag("-v");
2019-02-15 19:29:38 -06:00
// We always setup.
setup(subcommand);
// Invoke actual cargo for the job, but with different flags.
let miri_path = std::env::current_exe().expect("current executable path invalid");
let cargo_cmd = match subcommand {
MiriCommand::Test => "test",
MiriCommand::Run => "run",
MiriCommand::Setup => return, // `cargo miri setup` stops here.
};
let mut cmd = cargo();
cmd.arg(cargo_cmd);
// Make sure we know the build target, and cargo does, too.
// This is needed to make the `CARGO_TARGET_*_RUNNER` env var do something,
// and it later helps us detect which crates are proc-macro/build-script
// (host crates) and which crates are needed for the program itself.
let target = if let Some(target) = get_arg_flag_value("--target") {
target
} else {
// No target given. Pick default and tell cargo about it.
let host = version_info().host;
cmd.arg("--target");
cmd.arg(&host);
host
};
// Forward all further arguments.
cmd.args(args);
// Set `RUSTC_WRAPPER` to ourselves. Cargo will prepend that binary to its usual invocation,
// i.e., the first argument is `rustc` -- which is what we use in `main` to distinguish
// the two codepaths. (That extra argument is why we prefer this over setting `RUSTC`.)
if env::var_os("RUSTC_WRAPPER").is_some() {
println!("WARNING: Ignoring `RUSTC_WRAPPER` environment variable, Miri does not support wrapping.");
}
cmd.env("RUSTC_WRAPPER", &miri_path);
if verbose {
eprintln!("+ RUSTC_WRAPPER={:?}", miri_path);
}
// Set the runner for the current target to us as well, so we can interpret the binaries.
let runner_env_name = format!("CARGO_TARGET_{}_RUNNER", target.to_uppercase().replace('-', "_"));
cmd.env(runner_env_name, &miri_path);
// Run cargo.
if verbose {
cmd.env("MIRI_VERBOSE", ""); // this makes `inside_cargo_rustc` verbose.
eprintln!("+ {:?}", cmd);
}
let exit_status =
cmd.spawn().expect("could not run cargo").wait().expect("failed to wait for cargo?");
std::process::exit(exit_status.code().unwrap_or(-1))
}
fn phase_cargo_rustc(mut args: env::Args) {
/// Determines if we are being invoked (as rustc) to build a crate for
/// the "target" architecture, in contrast to the "host" architecture.
/// Host crates are for build scripts and proc macros and still need to
/// be built like normal; target crates need to be built for or interpreted
/// by Miri.
2020-02-24 07:17:19 -06:00
///
/// Currently, we detect this by checking for "--target=", which is
/// never set for host crates. This matches what rustc bootstrap does,
2020-05-09 04:54:35 -05:00
/// which hopefully makes it "reliable enough". This relies on us always
/// invoking cargo itself with `--target`, which `in_cargo_miri` ensures.
2020-02-24 07:17:19 -06:00
fn is_target_crate() -> bool {
get_arg_flag_value("--target").is_some()
2020-02-24 07:17:19 -06:00
}
2020-02-24 07:17:19 -06:00
/// Returns whether or not Cargo invoked the wrapper (this binary) to compile
/// the final, binary crate (either a test for 'cargo test', or a binary for 'cargo run')
2020-02-24 07:17:19 -06:00
/// Cargo does not give us this information directly, so we need to check
/// various command-line flags.
fn is_runnable_crate() -> bool {
let is_bin = get_arg_flag_value("--crate-type").as_deref() == Some("bin");
let is_test = has_arg_flag("--test");
let print = get_arg_flag_value("--print").is_some();
(is_bin || is_test) && !print
2020-02-24 07:17:19 -06:00
}
let verbose = std::env::var_os("MIRI_VERBOSE").is_some();
2020-02-24 07:17:19 -06:00
let target_crate = is_target_crate();
if target_crate && is_runnable_crate() {
// This is the binary or test crate that we want to interpret under Miri.
// But we cannot run it here, as cargo invoked us as a compiler -- our stdin and stdout are not
// like we want them.
// Instead of compiling, we write JSON into the output file with all the relevant command-line flags
// and environment variables; this is sued alter when cargo calls us again in the CARGO_TARGET_RUNNER phase.
let info = CrateRunInfo { args: Vec::new(), env: HashMap::new() };
let mut path = PathBuf::from(get_arg_flag_value("--out-dir").unwrap());
path.push(format!(
"{}{}",
get_arg_flag_value("--crate-name").unwrap(),
// This is technically a `-C` flag but the prefix seems unique enough...
// (and cargo passes this before the filename so it should be unique)
get_arg_flag_value("extra-filename").unwrap_or(String::new()),
));
eprintln!("Miri is supposed to run {}", path.display());
let file = File::create(&path)
.unwrap_or_else(|_| show_error(format!("Cannot create {}", path.display())));
let file = BufWriter::new(file);
serde_json::ser::to_writer(file, &info)
.unwrap_or_else(|_| show_error(format!("Cannot write to {}", path.display())));
return;
}
let mut cmd = miri();
// Forward arguments.
cmd.args(args);
// FIXME: Make the build check-only!
// We make sure to only specify our custom Xargo sysroot for target crates - that is,
// crates which are needed for interpretation by Miri. proc-macros and build scripts
// should use the default sysroot.
2020-02-24 07:17:19 -06:00
if target_crate {
2020-03-01 04:54:19 -06:00
let sysroot =
env::var_os("MIRI_SYSROOT").expect("The wrapper should have set MIRI_SYSROOT");
cmd.arg("--sysroot");
cmd.arg(sysroot);
2020-02-24 07:17:19 -06:00
}
// We want to compile, not interpret. We still use Miri to make sure the compiler version etc
// are the exact same as what is used for interpretation.
cmd.env("MIRI_BE_RUSTC", "1");
2020-02-24 07:17:19 -06:00
// Run it.
if verbose {
eprintln!("+ {:?}", cmd);
}
match cmd.status() {
2019-12-23 05:56:23 -06:00
Ok(exit) =>
if !exit.success() {
std::process::exit(exit.code().unwrap_or(42));
2019-12-23 05:56:23 -06:00
},
Err(e) => panic!("error running {:?}:\n{:?}", cmd, e),
}
}
fn phase_cargo_runner(binary: &str, args: env::Args) {
eprintln!("Asked to execute {}, args: {:?}", binary, args.collect::<Vec<_>>());
let file = File::open(binary)
.unwrap_or_else(|_| show_error(format!("File {:?} not found, or cargo-miri invoked incorrectly", binary)));
let file = BufReader::new(file);
let info: CrateRunInfo = serde_json::from_reader(file)
.unwrap_or_else(|_| show_error(format!("File {:?} does not contain valid JSON", binary)));
// FIXME: remove the file.
}
fn main() {
// Check for version and help flags even when invoked as `cargo-miri`.
if has_arg_flag("--help") || has_arg_flag("-h") {
show_help();
return;
}
if has_arg_flag("--version") || has_arg_flag("-V") {
show_version();
return;
}
let mut args = std::env::args();
// Skip binary name.
args.next().unwrap();
// Dispatch to `cargo-miri` phase. There are three phases:
// - When we are called via `cargo miri`, we run as the frontend and invoke the underlying
// cargo. We set RUSTC_WRAPPER and CARGO_TARGET_RUNNER to ourselves.
// - When we are executed due to RUSTC_WRAPPER, we build crates or store the flags of
// binary crates for later interpretation.
// - When we are executed due to CARGO_TARGET_RUNNER, we start interpretation based on the
// flags that were stored earlier.
match &*args.next().unwrap() {
"miri" => phase_cargo_miri(args),
"rustc" => phase_cargo_rustc(args),
binary => phase_cargo_runner(binary, args),
}
}