mirror of
https://github.com/uutils/coreutils
synced 2024-11-16 17:58:06 +00:00
1669 lines
54 KiB
Rust
1669 lines
54 KiB
Rust
// * This file is part of the uutils coreutils package.
|
|
// *
|
|
// * For the full copyright and license information, please view the LICENSE
|
|
// * file that was distributed with this source code.
|
|
|
|
//spell-checker: ignore (linux) rlimit prlimit Rlim coreutil ggroups
|
|
|
|
#![allow(dead_code)]
|
|
|
|
use pretty_assertions::assert_eq;
|
|
#[cfg(target_os = "linux")]
|
|
use rlimit::{prlimit, rlim};
|
|
#[cfg(unix)]
|
|
use std::borrow::Cow;
|
|
use std::env;
|
|
#[cfg(not(windows))]
|
|
use std::ffi::CString;
|
|
use std::ffi::OsStr;
|
|
use std::fs::{self, hard_link, File, OpenOptions};
|
|
use std::io::{Read, Result, Write};
|
|
#[cfg(unix)]
|
|
use std::os::unix::fs::{symlink as symlink_dir, symlink as symlink_file};
|
|
#[cfg(windows)]
|
|
use std::os::windows::fs::{symlink_dir, symlink_file};
|
|
use std::path::{Path, PathBuf};
|
|
use std::process::{Child, Command, Stdio};
|
|
use std::rc::Rc;
|
|
use std::thread::sleep;
|
|
use std::time::Duration;
|
|
use tempfile::TempDir;
|
|
use uucore::{Args, InvalidEncodingHandling};
|
|
|
|
#[cfg(windows)]
|
|
static PROGNAME: &str = concat!(env!("CARGO_PKG_NAME"), ".exe");
|
|
#[cfg(not(windows))]
|
|
static PROGNAME: &str = env!("CARGO_PKG_NAME");
|
|
|
|
static TESTS_DIR: &str = "tests";
|
|
static FIXTURES_DIR: &str = "fixtures";
|
|
|
|
static ALREADY_RUN: &str = " you have already run this UCommand, if you want to run \
|
|
another command in the same test, use TestScenario::new instead of \
|
|
testing();";
|
|
static MULTIPLE_STDIN_MEANINGLESS: &str = "Ucommand is designed around a typical use case of: provide args and input stream -> spawn process -> block until completion -> return output streams. For verifying that a particular section of the input stream is what causes a particular behavior, use the Command type directly.";
|
|
|
|
static NO_STDIN_MEANINGLESS: &str = "Setting this flag has no effect if there is no stdin";
|
|
|
|
/// Test if the program is running under CI
|
|
pub fn is_ci() -> bool {
|
|
std::env::var("CI")
|
|
.unwrap_or_else(|_| String::from("false"))
|
|
.eq_ignore_ascii_case("true")
|
|
}
|
|
|
|
/// Read a test scenario fixture, returning its bytes
|
|
fn read_scenario_fixture<S: AsRef<OsStr>>(tmpd: &Option<Rc<TempDir>>, file_rel_path: S) -> Vec<u8> {
|
|
let tmpdir_path = tmpd.as_ref().unwrap().as_ref().path();
|
|
AtPath::new(tmpdir_path).read_bytes(file_rel_path.as_ref().to_str().unwrap())
|
|
}
|
|
|
|
/// A command result is the outputs of a command (streams and status code)
|
|
/// within a struct which has convenience assertion functions about those outputs
|
|
#[derive(Debug, Clone)]
|
|
pub struct CmdResult {
|
|
/// bin_path provided by `TestScenario` or `UCommand`
|
|
bin_path: String,
|
|
/// util_name provided by `TestScenario` or `UCommand`
|
|
util_name: Option<String>,
|
|
//tmpd is used for convenience functions for asserts against fixtures
|
|
tmpd: Option<Rc<TempDir>>,
|
|
/// exit status for command (if there is one)
|
|
code: Option<i32>,
|
|
/// zero-exit from running the Command?
|
|
/// see [`success`]
|
|
success: bool,
|
|
/// captured standard output after running the Command
|
|
stdout: Vec<u8>,
|
|
/// captured standard error after running the Command
|
|
stderr: Vec<u8>,
|
|
}
|
|
|
|
impl CmdResult {
|
|
pub fn new(
|
|
bin_path: String,
|
|
util_name: Option<String>,
|
|
tmpd: Option<Rc<TempDir>>,
|
|
code: Option<i32>,
|
|
success: bool,
|
|
stdout: &[u8],
|
|
stderr: &[u8],
|
|
) -> CmdResult {
|
|
CmdResult {
|
|
bin_path,
|
|
util_name,
|
|
tmpd,
|
|
code,
|
|
success,
|
|
stdout: stdout.to_vec(),
|
|
stderr: stderr.to_vec(),
|
|
}
|
|
}
|
|
|
|
/// Returns a reference to the program's standard output as a slice of bytes
|
|
pub fn stdout(&self) -> &[u8] {
|
|
&self.stdout
|
|
}
|
|
|
|
/// Returns the program's standard output as a string slice
|
|
pub fn stdout_str(&self) -> &str {
|
|
std::str::from_utf8(&self.stdout).unwrap()
|
|
}
|
|
|
|
/// Returns the program's standard output as a string
|
|
/// consumes self
|
|
pub fn stdout_move_str(self) -> String {
|
|
String::from_utf8(self.stdout).unwrap()
|
|
}
|
|
|
|
/// Returns the program's standard output as a vec of bytes
|
|
/// consumes self
|
|
pub fn stdout_move_bytes(self) -> Vec<u8> {
|
|
self.stdout
|
|
}
|
|
|
|
/// Returns a reference to the program's standard error as a slice of bytes
|
|
pub fn stderr(&self) -> &[u8] {
|
|
&self.stderr
|
|
}
|
|
|
|
/// Returns the program's standard error as a string slice
|
|
pub fn stderr_str(&self) -> &str {
|
|
std::str::from_utf8(&self.stderr).unwrap()
|
|
}
|
|
|
|
/// Returns the program's standard error as a string
|
|
/// consumes self
|
|
pub fn stderr_move_str(self) -> String {
|
|
String::from_utf8(self.stderr).unwrap()
|
|
}
|
|
|
|
/// Returns the program's standard error as a vec of bytes
|
|
/// consumes self
|
|
pub fn stderr_move_bytes(self) -> Vec<u8> {
|
|
self.stderr
|
|
}
|
|
|
|
/// Returns the program's exit code
|
|
/// Panics if not run
|
|
pub fn code(&self) -> i32 {
|
|
self.code.expect("Program must be run first")
|
|
}
|
|
|
|
pub fn code_is(&self, expected_code: i32) -> &CmdResult {
|
|
assert_eq!(self.code(), expected_code);
|
|
self
|
|
}
|
|
|
|
/// Returns the program's TempDir
|
|
/// Panics if not present
|
|
pub fn tmpd(&self) -> Rc<TempDir> {
|
|
match &self.tmpd {
|
|
Some(ptr) => ptr.clone(),
|
|
None => panic!("Command not associated with a TempDir"),
|
|
}
|
|
}
|
|
|
|
/// Returns whether the program succeeded
|
|
pub fn succeeded(&self) -> bool {
|
|
self.success
|
|
}
|
|
|
|
/// asserts that the command resulted in a success (zero) status code
|
|
pub fn success(&self) -> &CmdResult {
|
|
assert!(
|
|
self.success,
|
|
"Command was expected to succeed.\nstdout = {}\n stderr = {}",
|
|
self.stdout_str(),
|
|
self.stderr_str()
|
|
);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in a failure (non-zero) status code
|
|
pub fn failure(&self) -> &CmdResult {
|
|
assert!(
|
|
!self.success,
|
|
"Command was expected to fail.\nstdout = {}\n stderr = {}",
|
|
self.stdout_str(),
|
|
self.stderr_str()
|
|
);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command's exit code is the same as the given one
|
|
pub fn status_code(&self, code: i32) -> &CmdResult {
|
|
assert_eq!(self.code, Some(code));
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in empty (zero-length) stderr stream output
|
|
/// generally, it's better to use stdout_only() instead,
|
|
/// but you might find yourself using this function if
|
|
/// 1. you can not know exactly what stdout will be or
|
|
/// 2. you know that stdout will also be empty
|
|
pub fn no_stderr(&self) -> &CmdResult {
|
|
assert!(
|
|
self.stderr.is_empty(),
|
|
"Expected stderr to be empty, but it's:\n{}",
|
|
self.stderr_str()
|
|
);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in empty (zero-length) stderr stream output
|
|
/// unless asserting there was neither stdout or stderr, stderr_only is usually a better choice
|
|
/// generally, it's better to use stderr_only() instead,
|
|
/// but you might find yourself using this function if
|
|
/// 1. you can not know exactly what stderr will be or
|
|
/// 2. you know that stderr will also be empty
|
|
pub fn no_stdout(&self) -> &CmdResult {
|
|
assert!(
|
|
self.stdout.is_empty(),
|
|
"Expected stdout to be empty, but it's:\n{}",
|
|
self.stderr_str()
|
|
);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in stdout stream output that equals the
|
|
/// passed in value, trailing whitespace are kept to force strict comparison (#1235)
|
|
/// stdout_only is a better choice unless stderr may or will be non-empty
|
|
pub fn stdout_is<T: AsRef<str>>(&self, msg: T) -> &CmdResult {
|
|
assert_eq!(self.stdout_str(), String::from(msg.as_ref()));
|
|
self
|
|
}
|
|
|
|
/// like `stdout_is`, but succeeds if any elements of `expected` matches stdout.
|
|
pub fn stdout_is_any<T: AsRef<str> + std::fmt::Debug>(&self, expected: Vec<T>) -> &CmdResult {
|
|
if !expected.iter().any(|msg| self.stdout_str() == msg.as_ref()) {
|
|
panic!(
|
|
"stdout was {}\nExpected any of {:#?}",
|
|
self.stdout_str(),
|
|
expected
|
|
)
|
|
}
|
|
self
|
|
}
|
|
|
|
/// Like `stdout_is` but newlines are normalized to `\n`.
|
|
pub fn normalized_newlines_stdout_is<T: AsRef<str>>(&self, msg: T) -> &CmdResult {
|
|
let msg = msg.as_ref().replace("\r\n", "\n");
|
|
assert_eq!(self.stdout_str().replace("\r\n", "\n"), msg);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in stdout stream output,
|
|
/// whose bytes equal those of the passed in slice
|
|
pub fn stdout_is_bytes<T: AsRef<[u8]>>(&self, msg: T) -> &CmdResult {
|
|
assert_eq!(self.stdout, msg.as_ref());
|
|
self
|
|
}
|
|
|
|
/// like stdout_is(...), but expects the contents of the file at the provided relative path
|
|
pub fn stdout_is_fixture<T: AsRef<OsStr>>(&self, file_rel_path: T) -> &CmdResult {
|
|
let contents = read_scenario_fixture(&self.tmpd, file_rel_path);
|
|
self.stdout_is(String::from_utf8(contents).unwrap())
|
|
}
|
|
/// like stdout_is_fixture(...), but replaces the data in fixture file based on values provided in template_vars
|
|
/// command output
|
|
pub fn stdout_is_templated_fixture<T: AsRef<OsStr>>(
|
|
&self,
|
|
file_rel_path: T,
|
|
template_vars: &[(&str, &str)],
|
|
) -> &CmdResult {
|
|
let mut contents =
|
|
String::from_utf8(read_scenario_fixture(&self.tmpd, file_rel_path)).unwrap();
|
|
for kv in template_vars {
|
|
contents = contents.replace(kv.0, kv.1);
|
|
}
|
|
self.stdout_is(contents)
|
|
}
|
|
|
|
/// like `stdout_is_templated_fixture`, but succeeds if any replacement by `template_vars` results in the actual stdout.
|
|
pub fn stdout_is_templated_fixture_any<T: AsRef<OsStr>>(
|
|
&self,
|
|
file_rel_path: T,
|
|
template_vars: &[Vec<(String, String)>],
|
|
) {
|
|
let contents = String::from_utf8(read_scenario_fixture(&self.tmpd, file_rel_path)).unwrap();
|
|
let possible_values = template_vars.iter().map(|vars| {
|
|
let mut contents = contents.clone();
|
|
for kv in vars.iter() {
|
|
contents = contents.replace(&kv.0, &kv.1);
|
|
}
|
|
contents
|
|
});
|
|
self.stdout_is_any(possible_values.collect());
|
|
}
|
|
|
|
/// asserts that the command resulted in stderr stream output that equals the
|
|
/// passed in value, when both are trimmed of trailing whitespace
|
|
/// stderr_only is a better choice unless stdout may or will be non-empty
|
|
pub fn stderr_is<T: AsRef<str>>(&self, msg: T) -> &CmdResult {
|
|
assert_eq!(
|
|
self.stderr_str().trim_end(),
|
|
String::from(msg.as_ref()).trim_end()
|
|
);
|
|
self
|
|
}
|
|
|
|
/// asserts that the command resulted in stderr stream output,
|
|
/// whose bytes equal those of the passed in slice
|
|
pub fn stderr_is_bytes<T: AsRef<[u8]>>(&self, msg: T) -> &CmdResult {
|
|
assert_eq!(self.stderr, msg.as_ref());
|
|
self
|
|
}
|
|
|
|
/// Like stdout_is_fixture, but for stderr
|
|
pub fn stderr_is_fixture<T: AsRef<OsStr>>(&self, file_rel_path: T) -> &CmdResult {
|
|
let contents = read_scenario_fixture(&self.tmpd, file_rel_path);
|
|
self.stderr_is(String::from_utf8(contents).unwrap())
|
|
}
|
|
|
|
/// asserts that
|
|
/// 1. the command resulted in stdout stream output that equals the
|
|
/// passed in value
|
|
/// 2. the command resulted in empty (zero-length) stderr stream output
|
|
pub fn stdout_only<T: AsRef<str>>(&self, msg: T) -> &CmdResult {
|
|
self.no_stderr().stdout_is(msg)
|
|
}
|
|
|
|
/// asserts that
|
|
/// 1. the command resulted in a stdout stream whose bytes
|
|
/// equal those of the passed in value
|
|
/// 2. the command resulted in an empty stderr stream
|
|
pub fn stdout_only_bytes<T: AsRef<[u8]>>(&self, msg: T) -> &CmdResult {
|
|
self.no_stderr().stdout_is_bytes(msg)
|
|
}
|
|
|
|
/// like stdout_only(...), but expects the contents of the file at the provided relative path
|
|
pub fn stdout_only_fixture<T: AsRef<OsStr>>(&self, file_rel_path: T) -> &CmdResult {
|
|
let contents = read_scenario_fixture(&self.tmpd, file_rel_path);
|
|
self.stdout_only_bytes(contents)
|
|
}
|
|
|
|
/// asserts that
|
|
/// 1. the command resulted in stderr stream output that equals the
|
|
/// passed in value, when both are trimmed of trailing whitespace
|
|
/// 2. the command resulted in empty (zero-length) stdout stream output
|
|
pub fn stderr_only<T: AsRef<str>>(&self, msg: T) -> &CmdResult {
|
|
self.no_stdout().stderr_is(msg)
|
|
}
|
|
|
|
/// asserts that
|
|
/// 1. the command resulted in a stderr stream whose bytes equal the ones
|
|
/// of the passed value
|
|
/// 2. the command resulted in an empty stdout stream
|
|
pub fn stderr_only_bytes<T: AsRef<[u8]>>(&self, msg: T) -> &CmdResult {
|
|
self.no_stderr().stderr_is_bytes(msg)
|
|
}
|
|
|
|
pub fn fails_silently(&self) -> &CmdResult {
|
|
assert!(!self.success);
|
|
assert!(self.stderr.is_empty());
|
|
self
|
|
}
|
|
|
|
pub fn stdout_contains<T: AsRef<str>>(&self, cmp: T) -> &CmdResult {
|
|
assert!(
|
|
self.stdout_str().contains(cmp.as_ref()),
|
|
"'{}' does not contain '{}'",
|
|
self.stdout_str(),
|
|
cmp.as_ref()
|
|
);
|
|
self
|
|
}
|
|
|
|
pub fn stderr_contains<T: AsRef<str>>(&self, cmp: T) -> &CmdResult {
|
|
assert!(
|
|
self.stderr_str().contains(cmp.as_ref()),
|
|
"'{}' does not contain '{}'",
|
|
self.stderr_str(),
|
|
cmp.as_ref()
|
|
);
|
|
self
|
|
}
|
|
|
|
pub fn stdout_does_not_contain<T: AsRef<str>>(&self, cmp: T) -> &CmdResult {
|
|
assert!(
|
|
!self.stdout_str().contains(cmp.as_ref()),
|
|
"'{}' contains '{}' but should not",
|
|
self.stdout_str(),
|
|
cmp.as_ref(),
|
|
);
|
|
self
|
|
}
|
|
|
|
pub fn stderr_does_not_contain<T: AsRef<str>>(&self, cmp: T) -> &CmdResult {
|
|
assert!(!self.stderr_str().contains(cmp.as_ref()));
|
|
self
|
|
}
|
|
|
|
pub fn stdout_matches(&self, regex: ®ex::Regex) -> &CmdResult {
|
|
if !regex.is_match(self.stdout_str().trim()) {
|
|
panic!("Stdout does not match regex:\n{}", self.stdout_str())
|
|
}
|
|
self
|
|
}
|
|
|
|
pub fn stdout_does_not_match(&self, regex: ®ex::Regex) -> &CmdResult {
|
|
if regex.is_match(self.stdout_str().trim()) {
|
|
panic!("Stdout matches regex:\n{}", self.stdout_str())
|
|
}
|
|
self
|
|
}
|
|
}
|
|
|
|
pub fn log_info<T: AsRef<str>, U: AsRef<str>>(msg: T, par: U) {
|
|
println!("{}: {}", msg.as_ref(), par.as_ref());
|
|
}
|
|
|
|
pub fn recursive_copy(src: &Path, dest: &Path) -> Result<()> {
|
|
if fs::metadata(src)?.is_dir() {
|
|
for entry in fs::read_dir(src)? {
|
|
let entry = entry?;
|
|
let mut new_dest = PathBuf::from(dest);
|
|
new_dest.push(entry.file_name());
|
|
if fs::metadata(entry.path())?.is_dir() {
|
|
fs::create_dir(&new_dest)?;
|
|
recursive_copy(&entry.path(), &new_dest)?;
|
|
} else {
|
|
fs::copy(&entry.path(), new_dest)?;
|
|
}
|
|
}
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
pub fn get_root_path() -> &'static str {
|
|
if cfg!(windows) {
|
|
"C:\\"
|
|
} else {
|
|
"/"
|
|
}
|
|
}
|
|
|
|
/// Object-oriented path struct that represents and operates on
|
|
/// paths relative to the directory it was constructed for.
|
|
#[derive(Clone)]
|
|
pub struct AtPath {
|
|
pub subdir: PathBuf,
|
|
}
|
|
|
|
impl AtPath {
|
|
pub fn new(subdir: &Path) -> AtPath {
|
|
AtPath {
|
|
subdir: PathBuf::from(subdir),
|
|
}
|
|
}
|
|
|
|
pub fn as_string(&self) -> String {
|
|
self.subdir.to_str().unwrap().to_owned()
|
|
}
|
|
|
|
pub fn plus(&self, name: &str) -> PathBuf {
|
|
let mut pathbuf = self.subdir.clone();
|
|
pathbuf.push(name);
|
|
pathbuf
|
|
}
|
|
|
|
pub fn plus_as_string(&self, name: &str) -> String {
|
|
String::from(self.plus(name).to_str().unwrap())
|
|
}
|
|
|
|
fn minus(&self, name: &str) -> PathBuf {
|
|
let prefixed = PathBuf::from(name);
|
|
if prefixed.starts_with(&self.subdir) {
|
|
let mut unprefixed = PathBuf::new();
|
|
for component in prefixed.components().skip(self.subdir.components().count()) {
|
|
unprefixed.push(component.as_os_str().to_str().unwrap());
|
|
}
|
|
unprefixed
|
|
} else {
|
|
prefixed
|
|
}
|
|
}
|
|
|
|
pub fn minus_as_string(&self, name: &str) -> String {
|
|
String::from(self.minus(name).to_str().unwrap())
|
|
}
|
|
|
|
pub fn set_readonly(&self, name: &str) {
|
|
let metadata = fs::metadata(self.plus(name)).unwrap();
|
|
let mut permissions = metadata.permissions();
|
|
permissions.set_readonly(true);
|
|
fs::set_permissions(self.plus(name), permissions).unwrap();
|
|
}
|
|
|
|
pub fn open(&self, name: &str) -> File {
|
|
log_info("open", self.plus_as_string(name));
|
|
File::open(self.plus(name)).unwrap()
|
|
}
|
|
|
|
pub fn read(&self, name: &str) -> String {
|
|
let mut f = self.open(name);
|
|
let mut contents = String::new();
|
|
f.read_to_string(&mut contents)
|
|
.unwrap_or_else(|e| panic!("Couldn't read {}: {}", name, e));
|
|
contents
|
|
}
|
|
|
|
pub fn read_bytes(&self, name: &str) -> Vec<u8> {
|
|
let mut f = self.open(name);
|
|
let mut contents = Vec::new();
|
|
f.read_to_end(&mut contents)
|
|
.unwrap_or_else(|e| panic!("Couldn't read {}: {}", name, e));
|
|
contents
|
|
}
|
|
|
|
pub fn write(&self, name: &str, contents: &str) {
|
|
log_info("write(default)", self.plus_as_string(name));
|
|
std::fs::write(self.plus(name), contents)
|
|
.unwrap_or_else(|e| panic!("Couldn't write {}: {}", name, e));
|
|
}
|
|
|
|
pub fn write_bytes(&self, name: &str, contents: &[u8]) {
|
|
log_info("write(default)", self.plus_as_string(name));
|
|
std::fs::write(self.plus(name), contents)
|
|
.unwrap_or_else(|e| panic!("Couldn't write {}: {}", name, e));
|
|
}
|
|
|
|
pub fn append(&self, name: &str, contents: &str) {
|
|
log_info("write(append)", self.plus_as_string(name));
|
|
let mut f = OpenOptions::new()
|
|
.write(true)
|
|
.append(true)
|
|
.create(true)
|
|
.open(self.plus(name))
|
|
.unwrap();
|
|
f.write_all(contents.as_bytes())
|
|
.unwrap_or_else(|e| panic!("Couldn't write(append) {}: {}", name, e));
|
|
}
|
|
|
|
pub fn append_bytes(&self, name: &str, contents: &[u8]) {
|
|
log_info("write(append)", self.plus_as_string(name));
|
|
let mut f = OpenOptions::new()
|
|
.write(true)
|
|
.append(true)
|
|
.create(true)
|
|
.open(self.plus(name))
|
|
.unwrap();
|
|
f.write_all(contents)
|
|
.unwrap_or_else(|e| panic!("Couldn't write(append) to {}: {}", name, e));
|
|
}
|
|
|
|
pub fn truncate(&self, name: &str, contents: &str) {
|
|
log_info("write(truncate)", self.plus_as_string(name));
|
|
let mut f = OpenOptions::new()
|
|
.write(true)
|
|
.truncate(true)
|
|
.create(true)
|
|
.open(self.plus(name))
|
|
.unwrap();
|
|
f.write_all(contents.as_bytes())
|
|
.unwrap_or_else(|e| panic!("Couldn't write(truncate) {}: {}", name, e));
|
|
}
|
|
|
|
pub fn rename(&self, source: &str, target: &str) {
|
|
let source = self.plus(source);
|
|
let target = self.plus(target);
|
|
log_info("rename", format!("{:?} {:?}", source, target));
|
|
std::fs::rename(&source, &target)
|
|
.unwrap_or_else(|e| panic!("Couldn't rename {:?} -> {:?}: {}", source, target, e));
|
|
}
|
|
|
|
pub fn remove(&self, source: &str) {
|
|
let source = self.plus(source);
|
|
log_info("remove", format!("{:?}", source));
|
|
std::fs::remove_file(&source)
|
|
.unwrap_or_else(|e| panic!("Couldn't remove {:?}: {}", source, e));
|
|
}
|
|
|
|
pub fn copy(&self, source: &str, target: &str) {
|
|
let source = self.plus(source);
|
|
let target = self.plus(target);
|
|
log_info("copy", format!("{:?} {:?}", source, target));
|
|
std::fs::copy(&source, &target)
|
|
.unwrap_or_else(|e| panic!("Couldn't copy {:?} -> {:?}: {}", source, target, e));
|
|
}
|
|
|
|
pub fn rmdir(&self, dir: &str) {
|
|
log_info("rmdir", self.plus_as_string(dir));
|
|
fs::remove_dir(&self.plus(dir)).unwrap();
|
|
}
|
|
|
|
pub fn mkdir(&self, dir: &str) {
|
|
log_info("mkdir", self.plus_as_string(dir));
|
|
fs::create_dir(&self.plus(dir)).unwrap();
|
|
}
|
|
|
|
pub fn mkdir_all(&self, dir: &str) {
|
|
log_info("mkdir_all", self.plus_as_string(dir));
|
|
fs::create_dir_all(self.plus(dir)).unwrap();
|
|
}
|
|
|
|
pub fn make_file(&self, name: &str) -> File {
|
|
match File::create(&self.plus(name)) {
|
|
Ok(f) => f,
|
|
Err(e) => panic!("{}", e),
|
|
}
|
|
}
|
|
|
|
pub fn touch(&self, file: &str) {
|
|
log_info("touch", self.plus_as_string(file));
|
|
File::create(&self.plus(file)).unwrap();
|
|
}
|
|
|
|
#[cfg(not(windows))]
|
|
pub fn mkfifo(&self, fifo: &str) {
|
|
let full_path = self.plus_as_string(fifo);
|
|
log_info("mkfifo", &full_path);
|
|
unsafe {
|
|
let fifo_name: CString = CString::new(full_path).expect("CString creation failed.");
|
|
libc::mkfifo(fifo_name.as_ptr(), libc::S_IWUSR | libc::S_IRUSR);
|
|
}
|
|
}
|
|
|
|
#[cfg(not(windows))]
|
|
pub fn is_fifo(&self, fifo: &str) -> bool {
|
|
unsafe {
|
|
let name = CString::new(self.plus_as_string(fifo)).unwrap();
|
|
let mut stat: libc::stat = std::mem::zeroed();
|
|
if libc::stat(name.as_ptr(), &mut stat) >= 0 {
|
|
libc::S_IFIFO & stat.st_mode != 0
|
|
} else {
|
|
false
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn hard_link(&self, original: &str, link: &str) {
|
|
log_info(
|
|
"hard_link",
|
|
&format!(
|
|
"{},{}",
|
|
self.plus_as_string(original),
|
|
self.plus_as_string(link)
|
|
),
|
|
);
|
|
hard_link(&self.plus(original), &self.plus(link)).unwrap();
|
|
}
|
|
|
|
pub fn symlink_file(&self, original: &str, link: &str) {
|
|
log_info(
|
|
"symlink",
|
|
&format!(
|
|
"{},{}",
|
|
self.plus_as_string(original),
|
|
self.plus_as_string(link)
|
|
),
|
|
);
|
|
symlink_file(&self.plus(original), &self.plus(link)).unwrap();
|
|
}
|
|
|
|
pub fn symlink_dir(&self, original: &str, link: &str) {
|
|
log_info(
|
|
"symlink",
|
|
&format!(
|
|
"{},{}",
|
|
self.plus_as_string(original),
|
|
self.plus_as_string(link)
|
|
),
|
|
);
|
|
symlink_dir(&self.plus(original), &self.plus(link)).unwrap();
|
|
}
|
|
|
|
pub fn is_symlink(&self, path: &str) -> bool {
|
|
log_info("is_symlink", self.plus_as_string(path));
|
|
match fs::symlink_metadata(&self.plus(path)) {
|
|
Ok(m) => m.file_type().is_symlink(),
|
|
Err(_) => false,
|
|
}
|
|
}
|
|
|
|
pub fn resolve_link(&self, path: &str) -> String {
|
|
log_info("resolve_link", self.plus_as_string(path));
|
|
match fs::read_link(&self.plus(path)) {
|
|
Ok(p) => self.minus_as_string(p.to_str().unwrap()),
|
|
Err(_) => "".to_string(),
|
|
}
|
|
}
|
|
|
|
pub fn symlink_metadata(&self, path: &str) -> fs::Metadata {
|
|
match fs::symlink_metadata(&self.plus(path)) {
|
|
Ok(m) => m,
|
|
Err(e) => panic!("{}", e),
|
|
}
|
|
}
|
|
|
|
pub fn metadata(&self, path: &str) -> fs::Metadata {
|
|
match fs::metadata(&self.plus(path)) {
|
|
Ok(m) => m,
|
|
Err(e) => panic!("{}", e),
|
|
}
|
|
}
|
|
|
|
pub fn file_exists(&self, path: &str) -> bool {
|
|
match fs::metadata(&self.plus(path)) {
|
|
Ok(m) => m.is_file(),
|
|
Err(_) => false,
|
|
}
|
|
}
|
|
|
|
pub fn dir_exists(&self, path: &str) -> bool {
|
|
match fs::metadata(&self.plus(path)) {
|
|
Ok(m) => m.is_dir(),
|
|
Err(_) => false,
|
|
}
|
|
}
|
|
|
|
pub fn root_dir_resolved(&self) -> String {
|
|
log_info("current_directory_resolved", "");
|
|
let s = self
|
|
.subdir
|
|
.canonicalize()
|
|
.unwrap()
|
|
.to_str()
|
|
.unwrap()
|
|
.to_owned();
|
|
|
|
// Due to canonicalize()'s use of GetFinalPathNameByHandleW() on Windows, the resolved path
|
|
// starts with '\\?\' to extend the limit of a given path to 32,767 wide characters.
|
|
//
|
|
// To address this issue, we remove this prepended string if available.
|
|
//
|
|
// Source:
|
|
// http://stackoverflow.com/questions/31439011/getfinalpathnamebyhandle-without-prepended
|
|
let prefix = "\\\\?\\";
|
|
|
|
if let Some(stripped) = s.strip_prefix(prefix) {
|
|
String::from(stripped)
|
|
} else {
|
|
s
|
|
}
|
|
}
|
|
}
|
|
|
|
/// An environment for running a single uutils test case, serves three functions:
|
|
/// 1. centralizes logic for locating the uutils binary and calling the utility
|
|
/// 2. provides a unique temporary directory for the test case
|
|
/// 3. copies over fixtures for the utility to the temporary directory
|
|
///
|
|
/// Fixtures can be found under `tests/fixtures/$util_name/`
|
|
pub struct TestScenario {
|
|
pub bin_path: PathBuf,
|
|
pub util_name: String,
|
|
pub fixtures: AtPath,
|
|
tmpd: Rc<TempDir>,
|
|
}
|
|
|
|
impl TestScenario {
|
|
pub fn new(util_name: &str) -> TestScenario {
|
|
let tmpd = Rc::new(TempDir::new().unwrap());
|
|
let ts = TestScenario {
|
|
bin_path: {
|
|
// Instead of hard coding the path relative to the current
|
|
// directory, use Cargo's OUT_DIR to find path to executable.
|
|
// This allows tests to be run using profiles other than debug.
|
|
let target_dir = path_concat!(env!("OUT_DIR"), "..", "..", "..", PROGNAME);
|
|
PathBuf::from(AtPath::new(Path::new(&target_dir)).root_dir_resolved())
|
|
},
|
|
util_name: String::from(util_name),
|
|
fixtures: AtPath::new(tmpd.as_ref().path()),
|
|
tmpd,
|
|
};
|
|
let mut fixture_path_builder = env::current_dir().unwrap();
|
|
fixture_path_builder.push(TESTS_DIR);
|
|
fixture_path_builder.push(FIXTURES_DIR);
|
|
fixture_path_builder.push(util_name);
|
|
if let Ok(m) = fs::metadata(&fixture_path_builder) {
|
|
if m.is_dir() {
|
|
recursive_copy(&fixture_path_builder, &ts.fixtures.subdir).unwrap();
|
|
}
|
|
}
|
|
ts
|
|
}
|
|
|
|
/// Returns builder for invoking the target uutils binary. Paths given are
|
|
/// treated relative to the environment's unique temporary test directory.
|
|
pub fn ucmd(&self) -> UCommand {
|
|
self.composite_cmd(&self.bin_path, &self.util_name, true)
|
|
}
|
|
|
|
/// Returns builder for invoking the target uutils binary. Paths given are
|
|
/// treated relative to the environment's unique temporary test directory.
|
|
pub fn composite_cmd<S: AsRef<OsStr>, T: AsRef<OsStr>>(
|
|
&self,
|
|
bin: S,
|
|
util_name: T,
|
|
env_clear: bool,
|
|
) -> UCommand {
|
|
UCommand::new_from_tmp(bin, Some(util_name), self.tmpd.clone(), env_clear)
|
|
}
|
|
|
|
/// Returns builder for invoking any system command. Paths given are treated
|
|
/// relative to the environment's unique temporary test directory.
|
|
pub fn cmd<S: AsRef<OsStr>>(&self, bin: S) -> UCommand {
|
|
UCommand::new_from_tmp::<S, S>(bin, None, self.tmpd.clone(), true)
|
|
}
|
|
|
|
/// Returns builder for invoking any uutils command. Paths given are treated
|
|
/// relative to the environment's unique temporary test directory.
|
|
pub fn ccmd<S: AsRef<OsStr>>(&self, bin: S) -> UCommand {
|
|
self.composite_cmd(&self.bin_path, bin, true)
|
|
}
|
|
|
|
// different names are used rather than an argument
|
|
// because the need to keep the environment is exceedingly rare.
|
|
pub fn ucmd_keepenv(&self) -> UCommand {
|
|
self.composite_cmd(&self.bin_path, &self.util_name, false)
|
|
}
|
|
|
|
/// Returns builder for invoking any system command. Paths given are treated
|
|
/// relative to the environment's unique temporary test directory.
|
|
/// Differs from the builder returned by `cmd` in that `cmd_keepenv` does not call
|
|
/// `Command::env_clear` (Clears the entire environment map for the child process.)
|
|
pub fn cmd_keepenv<S: AsRef<OsStr>>(&self, bin: S) -> UCommand {
|
|
UCommand::new_from_tmp::<S, S>(bin, None, self.tmpd.clone(), false)
|
|
}
|
|
}
|
|
|
|
/// A `UCommand` is a wrapper around an individual Command that provides several additional features
|
|
/// 1. it has convenience functions that are more ergonomic to use for piping in stdin, spawning the command
|
|
/// and asserting on the results.
|
|
/// 2. it tracks arguments provided so that in test cases which may provide variations of an arg in loops
|
|
/// the test failure can display the exact call which preceded an assertion failure.
|
|
/// 3. it provides convenience construction arguments to set the Command working directory and/or clear its environment.
|
|
#[derive(Debug)]
|
|
pub struct UCommand {
|
|
pub raw: Command,
|
|
comm_string: String,
|
|
bin_path: String,
|
|
util_name: Option<String>,
|
|
tmpd: Option<Rc<TempDir>>,
|
|
has_run: bool,
|
|
ignore_stdin_write_error: bool,
|
|
stdin: Option<Stdio>,
|
|
stdout: Option<Stdio>,
|
|
stderr: Option<Stdio>,
|
|
bytes_into_stdin: Option<Vec<u8>>,
|
|
#[cfg(target_os = "linux")]
|
|
limits: Vec<(rlimit::Resource, rlim, rlim)>,
|
|
}
|
|
|
|
impl UCommand {
|
|
pub fn new<T: AsRef<OsStr>, S: AsRef<OsStr>, U: AsRef<OsStr>>(
|
|
bin_path: T,
|
|
util_name: Option<S>,
|
|
curdir: U,
|
|
env_clear: bool,
|
|
) -> UCommand {
|
|
let bin_path = bin_path.as_ref();
|
|
let util_name = util_name.as_ref().map(|un| un.as_ref());
|
|
|
|
let mut ucmd = UCommand {
|
|
tmpd: None,
|
|
has_run: false,
|
|
raw: {
|
|
let mut cmd = Command::new(bin_path);
|
|
cmd.current_dir(curdir.as_ref());
|
|
if env_clear {
|
|
if cfg!(windows) {
|
|
// spell-checker:ignore (dll) rsaenh
|
|
// %SYSTEMROOT% is required on Windows to initialize crypto provider
|
|
// ... and crypto provider is required for std::rand
|
|
// From `procmon`: RegQueryValue HKLM\SOFTWARE\Microsoft\Cryptography\Defaults\Provider\Microsoft Strong Cryptographic Provider\Image Path
|
|
// SUCCESS Type: REG_SZ, Length: 66, Data: %SystemRoot%\system32\rsaenh.dll"
|
|
for (key, _) in env::vars_os() {
|
|
if key.as_os_str() != "SYSTEMROOT" {
|
|
cmd.env_remove(key);
|
|
}
|
|
}
|
|
} else {
|
|
cmd.env_clear();
|
|
}
|
|
}
|
|
cmd
|
|
},
|
|
comm_string: String::from(bin_path.to_str().unwrap()),
|
|
bin_path: bin_path.to_str().unwrap().to_string(),
|
|
util_name: util_name.map(|un| un.to_str().unwrap().to_string()),
|
|
ignore_stdin_write_error: false,
|
|
bytes_into_stdin: None,
|
|
stdin: None,
|
|
stdout: None,
|
|
stderr: None,
|
|
#[cfg(target_os = "linux")]
|
|
limits: vec![],
|
|
};
|
|
|
|
if let Some(un) = util_name {
|
|
ucmd.arg(un);
|
|
}
|
|
|
|
ucmd
|
|
}
|
|
|
|
pub fn new_from_tmp<T: AsRef<OsStr>, S: AsRef<OsStr>>(
|
|
bin_path: T,
|
|
util_name: Option<S>,
|
|
tmpd: Rc<TempDir>,
|
|
env_clear: bool,
|
|
) -> UCommand {
|
|
let tmpd_path_buf = String::from(&(*tmpd.as_ref().path().to_str().unwrap()));
|
|
let mut ucmd: UCommand = UCommand::new(bin_path, util_name, tmpd_path_buf, env_clear);
|
|
ucmd.tmpd = Some(tmpd);
|
|
ucmd
|
|
}
|
|
|
|
pub fn set_stdin<T: Into<Stdio>>(&mut self, stdin: T) -> &mut UCommand {
|
|
self.stdin = Some(stdin.into());
|
|
self
|
|
}
|
|
|
|
pub fn set_stdout<T: Into<Stdio>>(&mut self, stdout: T) -> &mut UCommand {
|
|
self.stdout = Some(stdout.into());
|
|
self
|
|
}
|
|
|
|
pub fn set_stderr<T: Into<Stdio>>(&mut self, stderr: T) -> &mut UCommand {
|
|
self.stderr = Some(stderr.into());
|
|
self
|
|
}
|
|
|
|
/// Add a parameter to the invocation. Path arguments are treated relative
|
|
/// to the test environment directory.
|
|
pub fn arg<S: AsRef<OsStr>>(&mut self, arg: S) -> &mut UCommand {
|
|
assert!(!self.has_run, "{}", ALREADY_RUN);
|
|
self.comm_string.push(' ');
|
|
self.comm_string
|
|
.push_str(arg.as_ref().to_str().unwrap_or_default());
|
|
self.raw.arg(arg.as_ref());
|
|
self
|
|
}
|
|
|
|
/// Add multiple parameters to the invocation. Path arguments are treated relative
|
|
/// to the test environment directory.
|
|
pub fn args<S: AsRef<OsStr>>(&mut self, args: &[S]) -> &mut UCommand {
|
|
assert!(!self.has_run, "{}", MULTIPLE_STDIN_MEANINGLESS);
|
|
let strings = args
|
|
.iter()
|
|
.map(|s| s.as_ref().to_os_string())
|
|
.collect_str(InvalidEncodingHandling::Ignore)
|
|
.accept_any();
|
|
|
|
for s in strings {
|
|
self.comm_string.push(' ');
|
|
self.comm_string.push_str(&s);
|
|
}
|
|
|
|
self.raw.args(args.as_ref());
|
|
self
|
|
}
|
|
|
|
/// provides standard input to feed in to the command when spawned
|
|
pub fn pipe_in<T: Into<Vec<u8>>>(&mut self, input: T) -> &mut UCommand {
|
|
assert!(
|
|
!self.bytes_into_stdin.is_some(),
|
|
"{}",
|
|
MULTIPLE_STDIN_MEANINGLESS
|
|
);
|
|
self.bytes_into_stdin = Some(input.into());
|
|
self
|
|
}
|
|
|
|
/// like pipe_in(...), but uses the contents of the file at the provided relative path as the piped in data
|
|
pub fn pipe_in_fixture<S: AsRef<OsStr>>(&mut self, file_rel_path: S) -> &mut UCommand {
|
|
let contents = read_scenario_fixture(&self.tmpd, file_rel_path);
|
|
self.pipe_in(contents)
|
|
}
|
|
|
|
/// Ignores error caused by feeding stdin to the command.
|
|
/// This is typically useful to test non-standard workflows
|
|
/// like feeding something to a command that does not read it
|
|
pub fn ignore_stdin_write_error(&mut self) -> &mut UCommand {
|
|
assert!(!self.bytes_into_stdin.is_none(), "{}", NO_STDIN_MEANINGLESS);
|
|
self.ignore_stdin_write_error = true;
|
|
self
|
|
}
|
|
|
|
pub fn env<K, V>(&mut self, key: K, val: V) -> &mut UCommand
|
|
where
|
|
K: AsRef<OsStr>,
|
|
V: AsRef<OsStr>,
|
|
{
|
|
assert!(!self.has_run, "{}", ALREADY_RUN);
|
|
self.raw.env(key, val);
|
|
self
|
|
}
|
|
|
|
#[cfg(target_os = "linux")]
|
|
pub fn with_limit(
|
|
&mut self,
|
|
resource: rlimit::Resource,
|
|
soft_limit: rlim,
|
|
hard_limit: rlim,
|
|
) -> &mut Self {
|
|
self.limits.push((resource, soft_limit, hard_limit));
|
|
self
|
|
}
|
|
|
|
/// Spawns the command, feeds the stdin if any, and returns the
|
|
/// child process immediately.
|
|
pub fn run_no_wait(&mut self) -> Child {
|
|
assert!(!self.has_run, "{}", ALREADY_RUN);
|
|
self.has_run = true;
|
|
log_info("run", &self.comm_string);
|
|
let mut child = self
|
|
.raw
|
|
.stdin(self.stdin.take().unwrap_or_else(Stdio::piped))
|
|
.stdout(self.stdout.take().unwrap_or_else(Stdio::piped))
|
|
.stderr(self.stderr.take().unwrap_or_else(Stdio::piped))
|
|
.spawn()
|
|
.unwrap();
|
|
|
|
#[cfg(target_os = "linux")]
|
|
for &(resource, soft_limit, hard_limit) in &self.limits {
|
|
prlimit(
|
|
child.id() as i32,
|
|
resource,
|
|
Some((soft_limit, hard_limit)),
|
|
None,
|
|
)
|
|
.unwrap();
|
|
}
|
|
|
|
if let Some(ref input) = self.bytes_into_stdin {
|
|
let write_result = child
|
|
.stdin
|
|
.take()
|
|
.unwrap_or_else(|| panic!("Could not take child process stdin"))
|
|
.write_all(input);
|
|
if !self.ignore_stdin_write_error {
|
|
if let Err(e) = write_result {
|
|
panic!("failed to write to stdin of child: {}", e)
|
|
}
|
|
}
|
|
}
|
|
|
|
child
|
|
}
|
|
|
|
/// Spawns the command, feeds the stdin if any, waits for the result
|
|
/// and returns a command result.
|
|
/// It is recommended that you instead use succeeds() or fails()
|
|
pub fn run(&mut self) -> CmdResult {
|
|
let prog = self.run_no_wait().wait_with_output().unwrap();
|
|
|
|
CmdResult {
|
|
bin_path: self.bin_path.clone(),
|
|
util_name: self.util_name.clone(),
|
|
tmpd: self.tmpd.clone(),
|
|
code: prog.status.code(),
|
|
success: prog.status.success(),
|
|
stdout: prog.stdout,
|
|
stderr: prog.stderr,
|
|
}
|
|
}
|
|
|
|
/// Spawns the command, feeding the passed in stdin, waits for the result
|
|
/// and returns a command result.
|
|
/// It is recommended that, instead of this, you use a combination of pipe_in()
|
|
/// with succeeds() or fails()
|
|
pub fn run_piped_stdin<T: Into<Vec<u8>>>(&mut self, input: T) -> CmdResult {
|
|
self.pipe_in(input).run()
|
|
}
|
|
|
|
/// Spawns the command, feeds the stdin if any, waits for the result,
|
|
/// asserts success, and returns a command result.
|
|
pub fn succeeds(&mut self) -> CmdResult {
|
|
let cmd_result = self.run();
|
|
cmd_result.success();
|
|
cmd_result
|
|
}
|
|
|
|
/// Spawns the command, feeds the stdin if any, waits for the result,
|
|
/// asserts failure, and returns a command result.
|
|
pub fn fails(&mut self) -> CmdResult {
|
|
let cmd_result = self.run();
|
|
cmd_result.failure();
|
|
cmd_result
|
|
}
|
|
|
|
pub fn get_full_fixture_path(&self, file_rel_path: &str) -> String {
|
|
let tmpdir_path = self.tmpd.as_ref().unwrap().path();
|
|
format!("{}/{}", tmpdir_path.to_str().unwrap(), file_rel_path)
|
|
}
|
|
}
|
|
|
|
/// Wrapper for `child.stdout.read_exact()`.
|
|
/// Careful, this blocks indefinitely if `size` bytes is never reached.
|
|
pub fn read_size(child: &mut Child, size: usize) -> String {
|
|
let mut output = Vec::new();
|
|
output.resize(size, 0);
|
|
sleep(Duration::from_secs(1));
|
|
child
|
|
.stdout
|
|
.as_mut()
|
|
.unwrap()
|
|
.read_exact(output.as_mut_slice())
|
|
.unwrap();
|
|
String::from_utf8(output).unwrap()
|
|
}
|
|
|
|
pub fn vec_of_size(n: usize) -> Vec<u8> {
|
|
let result = vec![b'a'; n];
|
|
assert_eq!(result.len(), n);
|
|
result
|
|
}
|
|
|
|
pub fn whoami() -> String {
|
|
// Apparently some CI environments have configuration issues, e.g. with 'whoami' and 'id'.
|
|
//
|
|
// From the Logs: "Build (ubuntu-18.04, x86_64-unknown-linux-gnu, feat_os_unix, use-cross)"
|
|
// whoami: cannot find name for user ID 1001
|
|
// id --name: cannot find name for user ID 1001
|
|
// id --name: cannot find name for group ID 116
|
|
//
|
|
// However, when running "id" from within "/bin/bash" it looks fine:
|
|
// id: "uid=1001(runner) gid=118(docker) groups=118(docker),4(adm),101(systemd-journal)"
|
|
// whoami: "runner"
|
|
|
|
// Use environment variable to get current user instead of
|
|
// invoking `whoami` and fall back to user "nobody" on error.
|
|
std::env::var("USER")
|
|
.or_else(|_| std::env::var("USERNAME"))
|
|
.unwrap_or_else(|e| {
|
|
println!("{}: {}, using \"nobody\" instead", UUTILS_WARNING, e);
|
|
"nobody".to_string()
|
|
})
|
|
}
|
|
|
|
/// Add prefix 'g' for `util_name` if not on linux
|
|
#[cfg(unix)]
|
|
pub fn host_name_for(util_name: &str) -> Cow<str> {
|
|
// In some environments, e.g. macOS/freebsd, the GNU coreutils are prefixed with "g"
|
|
// to not interfere with the BSD counterparts already in `$PATH`.
|
|
#[cfg(not(target_os = "linux"))]
|
|
{
|
|
// make call to `host_name_for` idempotent
|
|
if util_name.starts_with('g') && util_name != "groups" {
|
|
return util_name.into();
|
|
} else {
|
|
return format!("g{}", util_name).into();
|
|
}
|
|
}
|
|
#[cfg(target_os = "linux")]
|
|
return util_name.into();
|
|
}
|
|
|
|
// GNU coreutils version 8.32 is the reference version since it is the latest version and the
|
|
// GNU test suite in "coreutils/.github/workflows/GnuTests.yml" runs against it.
|
|
// However, here 8.30 was chosen because right now there's no ubuntu image for the github actions
|
|
// CICD available with a higher version than 8.30.
|
|
// GNU coreutils versions from the CICD images for comparison:
|
|
// ubuntu-2004: 8.30 (latest)
|
|
// ubuntu-1804: 8.28
|
|
// macos-latest: 8.32
|
|
const VERSION_MIN: &str = "8.30"; // minimum Version for the reference `coreutil` in `$PATH`
|
|
|
|
const UUTILS_WARNING: &str = "uutils-tests-warning";
|
|
const UUTILS_INFO: &str = "uutils-tests-info";
|
|
|
|
/// Run `util_name --version` and return Ok if the version is >= `version_expected`.
|
|
/// Returns an error if
|
|
/// * `util_name` cannot run
|
|
/// * the version cannot be parsed
|
|
/// * the version is too low
|
|
///
|
|
/// This is used by `expected_result` to check if the coreutils version is >= `VERSION_MIN`.
|
|
/// It makes sense to use this manually in a test if a feature
|
|
/// is tested that was introduced after `VERSION_MIN`
|
|
///
|
|
/// Example:
|
|
///
|
|
/// ```no_run
|
|
/// use crate::common::util::*;
|
|
/// const VERSION_MIN_MULTIPLE_USERS: &str = "8.31";
|
|
///
|
|
/// #[test]
|
|
/// fn test_xyz() {
|
|
/// unwrap_or_return!(check_coreutil_version(
|
|
/// util_name!(),
|
|
/// VERSION_MIN_MULTIPLE_USERS
|
|
/// ));
|
|
/// // proceed with the test...
|
|
/// }
|
|
/// ```
|
|
#[cfg(unix)]
|
|
pub fn check_coreutil_version(
|
|
util_name: &str,
|
|
version_expected: &str,
|
|
) -> std::result::Result<String, String> {
|
|
// example:
|
|
// $ id --version | head -n 1
|
|
// id (GNU coreutils) 8.32.162-4eda
|
|
|
|
let util_name = &host_name_for(util_name);
|
|
log_info("run", format!("{} --version", util_name));
|
|
let version_check = match Command::new(util_name.as_ref())
|
|
.env("LC_ALL", "C")
|
|
.arg("--version")
|
|
.output()
|
|
{
|
|
Ok(s) => s,
|
|
Err(e) => {
|
|
return Err(format!(
|
|
"{}: '{}' {}",
|
|
UUTILS_WARNING,
|
|
util_name,
|
|
e.to_string()
|
|
))
|
|
}
|
|
};
|
|
std::str::from_utf8(&version_check.stdout).unwrap()
|
|
.split('\n')
|
|
.collect::<Vec<_>>()
|
|
.get(0)
|
|
.map_or_else(
|
|
|| Err(format!("{}: unexpected output format for reference coreutil: '{} --version'", UUTILS_WARNING, util_name)),
|
|
|s| {
|
|
if s.contains(&format!("(GNU coreutils) {}", version_expected)) {
|
|
Ok(format!("{}: {}", UUTILS_INFO, s.to_string()))
|
|
} else if s.contains("(GNU coreutils)") {
|
|
let version_found = parse_coreutil_version(s);
|
|
let version_expected = version_expected.parse::<f32>().unwrap_or_default();
|
|
if version_found > version_expected {
|
|
Ok(format!("{}: version for the reference coreutil '{}' is higher than expected; expected: {}, found: {}", UUTILS_INFO, util_name, version_expected, version_found))
|
|
} else {
|
|
Err(format!("{}: version for the reference coreutil '{}' does not match; expected: {}, found: {}", UUTILS_WARNING, util_name, version_expected, version_found)) }
|
|
} else {
|
|
Err(format!("{}: no coreutils version string found for reference coreutils '{} --version'", UUTILS_WARNING, util_name))
|
|
}
|
|
},
|
|
)
|
|
}
|
|
|
|
// simple heuristic to parse the coreutils SemVer string, e.g. "id (GNU coreutils) 8.32.263-0475"
|
|
fn parse_coreutil_version(version_string: &str) -> f32 {
|
|
version_string
|
|
.split_whitespace()
|
|
.last()
|
|
.unwrap()
|
|
.split('.')
|
|
.take(2)
|
|
.collect::<Vec<_>>()
|
|
.join(".")
|
|
.parse::<f32>()
|
|
.unwrap_or_default()
|
|
}
|
|
|
|
/// This runs the GNU coreutils `util_name` binary in `$PATH` in order to
|
|
/// dynamically gather reference values on the system.
|
|
/// If the `util_name` in `$PATH` doesn't include a coreutils version string,
|
|
/// or the version is too low, this returns an error and the test should be skipped.
|
|
///
|
|
/// Example:
|
|
///
|
|
/// ```no_run
|
|
/// use crate::common::util::*;
|
|
/// #[test]
|
|
/// fn test_xyz() {
|
|
/// let ts = TestScenario::new(util_name!());
|
|
/// let result = ts.ucmd().run();
|
|
/// let exp_result = unwrap_or_return!(expected_result(&ts, &[]));
|
|
/// result
|
|
/// .stdout_is(exp_result.stdout_str())
|
|
/// .stderr_is(exp_result.stderr_str())
|
|
/// .code_is(exp_result.code());
|
|
/// }
|
|
///```
|
|
#[cfg(unix)]
|
|
pub fn expected_result(ts: &TestScenario, args: &[&str]) -> std::result::Result<CmdResult, String> {
|
|
println!("{}", check_coreutil_version(&ts.util_name, VERSION_MIN)?);
|
|
let util_name = &host_name_for(&ts.util_name);
|
|
|
|
let result = ts
|
|
.cmd_keepenv(util_name.as_ref())
|
|
.env("LC_ALL", "C")
|
|
.args(args)
|
|
.run();
|
|
|
|
let (stdout, stderr): (String, String) = if cfg!(target_os = "linux") {
|
|
(
|
|
result.stdout_str().to_string(),
|
|
result.stderr_str().to_string(),
|
|
)
|
|
} else {
|
|
// `host_name_for` added prefix, strip 'g' prefix from results:
|
|
let from = util_name.to_string() + ":";
|
|
let to = &from[1..];
|
|
(
|
|
result.stdout_str().replace(&from, to),
|
|
result.stderr_str().replace(&from, to),
|
|
)
|
|
};
|
|
|
|
Ok(CmdResult::new(
|
|
ts.bin_path.as_os_str().to_str().unwrap().to_string(),
|
|
Some(ts.util_name.clone()),
|
|
Some(result.tmpd()),
|
|
Some(result.code()),
|
|
result.succeeded(),
|
|
stdout.as_bytes(),
|
|
stderr.as_bytes(),
|
|
))
|
|
}
|
|
|
|
/// Sanity checks for test utils
|
|
#[cfg(test)]
|
|
mod tests {
|
|
// spell-checker:ignore (tests) asdfsadfa
|
|
use super::*;
|
|
|
|
#[test]
|
|
fn test_code_is() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: Some(32),
|
|
success: false,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.code_is(32);
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_code_is_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: Some(32),
|
|
success: false,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.code_is(1);
|
|
}
|
|
|
|
#[test]
|
|
fn test_failure() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: false,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.failure();
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_failure_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.failure();
|
|
}
|
|
|
|
#[test]
|
|
fn test_success() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.success();
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_success_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: false,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.success();
|
|
}
|
|
|
|
#[test]
|
|
fn test_no_stderr_output() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "".into(),
|
|
stderr: "".into(),
|
|
};
|
|
res.no_stderr();
|
|
res.no_stdout();
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_no_stderr_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "".into(),
|
|
stderr: "asdfsadfa".into(),
|
|
};
|
|
|
|
res.no_stderr();
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_no_stdout_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "asdfsadfa".into(),
|
|
stderr: "".into(),
|
|
};
|
|
|
|
res.no_stdout();
|
|
}
|
|
|
|
#[test]
|
|
fn test_std_does_not_contain() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "This is a likely error message\n".into(),
|
|
stderr: "This is a likely error message\n".into(),
|
|
};
|
|
res.stdout_does_not_contain("unlikely");
|
|
res.stderr_does_not_contain("unlikely");
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_stdout_does_not_contain_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "This is a likely error message\n".into(),
|
|
stderr: "".into(),
|
|
};
|
|
|
|
res.stdout_does_not_contain("likely");
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_stderr_does_not_contain_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "".into(),
|
|
stderr: "This is a likely error message\n".into(),
|
|
};
|
|
|
|
res.stderr_does_not_contain("likely");
|
|
}
|
|
|
|
#[test]
|
|
fn test_stdout_matches() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "This is a likely error message\n".into(),
|
|
stderr: "This is a likely error message\n".into(),
|
|
};
|
|
let positive = regex::Regex::new(".*likely.*").unwrap();
|
|
let negative = regex::Regex::new(".*unlikely.*").unwrap();
|
|
res.stdout_matches(&positive);
|
|
res.stdout_does_not_match(&negative);
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_stdout_matches_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "This is a likely error message\n".into(),
|
|
stderr: "This is a likely error message\n".into(),
|
|
};
|
|
let negative = regex::Regex::new(".*unlikely.*").unwrap();
|
|
|
|
res.stdout_matches(&negative);
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_stdout_not_matches_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "This is a likely error message\n".into(),
|
|
stderr: "This is a likely error message\n".into(),
|
|
};
|
|
let positive = regex::Regex::new(".*likely.*").unwrap();
|
|
|
|
res.stdout_does_not_match(&positive);
|
|
}
|
|
|
|
#[test]
|
|
fn test_normalized_newlines_stdout_is() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "A\r\nB\nC".into(),
|
|
stderr: "".into(),
|
|
};
|
|
|
|
res.normalized_newlines_stdout_is("A\r\nB\nC");
|
|
res.normalized_newlines_stdout_is("A\nB\nC");
|
|
res.normalized_newlines_stdout_is("A\nB\r\nC");
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic]
|
|
fn test_normalized_newlines_stdout_is_fail() {
|
|
let res = CmdResult {
|
|
bin_path: "".into(),
|
|
util_name: None,
|
|
tmpd: None,
|
|
code: None,
|
|
success: true,
|
|
stdout: "A\r\nB\nC".into(),
|
|
stderr: "".into(),
|
|
};
|
|
|
|
res.normalized_newlines_stdout_is("A\r\nB\nC\n");
|
|
}
|
|
|
|
#[test]
|
|
#[cfg(unix)]
|
|
fn test_parse_coreutil_version() {
|
|
use std::assert_eq;
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 9.0.123-0123").to_string(),
|
|
"9"
|
|
);
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 8.32.263-0475").to_string(),
|
|
"8.32"
|
|
);
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 8.25.123-0123").to_string(),
|
|
"8.25"
|
|
);
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 9.0").to_string(),
|
|
"9"
|
|
);
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 8.32").to_string(),
|
|
"8.32"
|
|
);
|
|
assert_eq!(
|
|
parse_coreutil_version("id (GNU coreutils) 8.25").to_string(),
|
|
"8.25"
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg(unix)]
|
|
fn test_check_coreutil_version() {
|
|
match check_coreutil_version("id", VERSION_MIN) {
|
|
Ok(s) => assert!(s.starts_with("uutils-tests-")),
|
|
Err(s) => assert!(s.starts_with("uutils-tests-warning")),
|
|
};
|
|
#[cfg(target_os = "linux")]
|
|
std::assert_eq!(
|
|
check_coreutil_version("no test name", VERSION_MIN),
|
|
Err("uutils-tests-warning: 'no test name' \
|
|
No such file or directory (os error 2)"
|
|
.to_string())
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
#[cfg(unix)]
|
|
fn test_expected_result() {
|
|
let ts = TestScenario::new("id");
|
|
// assert!(expected_result(&ts, &[]).is_ok());
|
|
match expected_result(&ts, &[]) {
|
|
Ok(r) => assert!(r.succeeded()),
|
|
Err(s) => assert!(s.starts_with("uutils-tests-warning")),
|
|
}
|
|
let ts = TestScenario::new("no test name");
|
|
assert!(expected_result(&ts, &[]).is_err());
|
|
}
|
|
|
|
#[test]
|
|
#[cfg(unix)]
|
|
fn test_host_name_for() {
|
|
#[cfg(target_os = "linux")]
|
|
{
|
|
std::assert_eq!(host_name_for("id"), "id");
|
|
std::assert_eq!(host_name_for("groups"), "groups");
|
|
std::assert_eq!(host_name_for("who"), "who");
|
|
}
|
|
#[cfg(not(target_os = "linux"))]
|
|
{
|
|
// spell-checker:ignore (strings) ggroups gwho
|
|
std::assert_eq!(host_name_for("id"), "gid");
|
|
std::assert_eq!(host_name_for("groups"), "ggroups");
|
|
std::assert_eq!(host_name_for("who"), "gwho");
|
|
std::assert_eq!(host_name_for("gid"), "gid");
|
|
std::assert_eq!(host_name_for("ggroups"), "ggroups");
|
|
std::assert_eq!(host_name_for("gwho"), "gwho");
|
|
}
|
|
}
|
|
}
|