mirror of
https://github.com/fish-shell/fish-shell.git
synced 2024-12-20 06:13:47 +08:00
8c62f733b3
This updates is_windows_subsystem_for_linux() to take a WSL version to test for (any, v1, or v2) and returns the boolean result depending on the system. I've benchmarked and when running on regular Linux, this is still just as fast as the previous binary check; it's only when it's WSL that this takes about 20ns longer to figure out which variant. Note that older WSLv2 kernels had a `-microsoft-standard` suffix while newer ones appear to have a `-microsoft-standard-WSL2` suffix, so we make sure to test for the least common denominator. (It doesn't matter to us, but note that newer WSLv2 kernels have four dots in the version string!) WSL workarounds pertaining to the default Windows terminal or executable behavior of win32 binaries under a WSL shell are extended to WSLv2 while those specific to oddities in kernel behavior are confined to WSLv1 only. (It technically wouldn't hurt to extend them to WSLv2 but there's no good reason to do so, either.)
620 lines
20 KiB
Rust
620 lines
20 KiB
Rust
use crate::common::{is_windows_subsystem_for_linux, str2wcstring, wcs2osstring, wcs2string, WSL};
|
|
use crate::env::{EnvMode, EnvStack};
|
|
use crate::history::{self, History, HistoryItem, HistorySearch, PathList, SearchDirection};
|
|
use crate::path::path_get_data;
|
|
use crate::tests::prelude::*;
|
|
use crate::tests::string_escape::ESCAPE_TEST_CHAR;
|
|
use crate::wchar::prelude::*;
|
|
use crate::wcstringutil::{string_prefixes_string, string_prefixes_string_case_insensitive};
|
|
use rand::random;
|
|
use std::collections::VecDeque;
|
|
use std::ffi::CString;
|
|
use std::io::BufReader;
|
|
use std::time::SystemTime;
|
|
use std::time::UNIX_EPOCH;
|
|
|
|
fn history_contains(history: &History, txt: &wstr) -> bool {
|
|
for i in 1.. {
|
|
let Some(item) = history.item_at_index(i) else {
|
|
break;
|
|
};
|
|
|
|
if item.str() == txt {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
false
|
|
}
|
|
|
|
fn random_string() -> WString {
|
|
let mut result = WString::new();
|
|
let max = 1 + random::<usize>() % 32;
|
|
for _ in 0..max {
|
|
let c = char::from_u32(u32::try_from(1 + random::<usize>() % ESCAPE_TEST_CHAR).unwrap())
|
|
.unwrap();
|
|
result.push(c);
|
|
}
|
|
result
|
|
}
|
|
|
|
#[test]
|
|
#[serial]
|
|
fn test_history() {
|
|
let _cleanup = test_init();
|
|
macro_rules! test_history_matches {
|
|
($search:expr, $expected:expr) => {
|
|
let expected: Vec<&wstr> = $expected;
|
|
let mut found = vec![];
|
|
while $search.go_to_next_match(SearchDirection::Backward) {
|
|
found.push($search.current_string().to_owned());
|
|
}
|
|
assert_eq!(expected, found);
|
|
};
|
|
}
|
|
|
|
let items = [
|
|
L!("Gamma"),
|
|
L!("beta"),
|
|
L!("BetA"),
|
|
L!("Beta"),
|
|
L!("alpha"),
|
|
L!("AlphA"),
|
|
L!("Alpha"),
|
|
L!("alph"),
|
|
L!("ALPH"),
|
|
L!("ZZZ"),
|
|
];
|
|
let nocase = history::SearchFlags::IGNORE_CASE;
|
|
|
|
// Populate a history.
|
|
let history = History::with_name(L!("test_history"));
|
|
history.clear();
|
|
for s in items {
|
|
history.add_commandline(s.to_owned());
|
|
}
|
|
|
|
// Helper to set expected items to those matching a predicate, in reverse order.
|
|
let set_expected = |filt: fn(&wstr) -> bool| {
|
|
let mut expected = vec![];
|
|
for s in items {
|
|
if filt(s) {
|
|
expected.push(s);
|
|
}
|
|
}
|
|
expected.reverse();
|
|
expected
|
|
};
|
|
|
|
// Items matching "a", case-sensitive.
|
|
let mut searcher = HistorySearch::new(history.clone(), L!("a").to_owned());
|
|
let expected = set_expected(|s| s.contains('a'));
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items matching "alpha", case-insensitive.
|
|
let mut searcher =
|
|
HistorySearch::new_with_flags(history.clone(), L!("AlPhA").to_owned(), nocase);
|
|
let expected = set_expected(|s| s.to_lowercase().find(L!("alpha")).is_some());
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items matching "et", case-sensitive.
|
|
let mut searcher = HistorySearch::new(history.clone(), L!("et").to_owned());
|
|
let expected = set_expected(|s| s.find(L!("et")).is_some());
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items starting with "be", case-sensitive.
|
|
let mut searcher = HistorySearch::new_with_type(
|
|
history.clone(),
|
|
L!("be").to_owned(),
|
|
history::SearchType::Prefix,
|
|
);
|
|
let expected = set_expected(|s| string_prefixes_string(L!("be"), s));
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items starting with "be", case-insensitive.
|
|
let mut searcher = HistorySearch::new_with(
|
|
history.clone(),
|
|
L!("be").to_owned(),
|
|
history::SearchType::Prefix,
|
|
nocase,
|
|
0,
|
|
);
|
|
let expected = set_expected(|s| string_prefixes_string_case_insensitive(L!("be"), s));
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items exactly matching "alph", case-sensitive.
|
|
let mut searcher = HistorySearch::new_with_type(
|
|
history.clone(),
|
|
L!("alph").to_owned(),
|
|
history::SearchType::Exact,
|
|
);
|
|
let expected = set_expected(|s| s == "alph");
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Items exactly matching "alph", case-insensitive.
|
|
let mut searcher = HistorySearch::new_with(
|
|
history.clone(),
|
|
L!("alph").to_owned(),
|
|
history::SearchType::Exact,
|
|
nocase,
|
|
0,
|
|
);
|
|
let expected = set_expected(|s| s.to_lowercase() == "alph");
|
|
test_history_matches!(searcher, expected);
|
|
|
|
// Test item removal case-sensitive.
|
|
let mut searcher = HistorySearch::new(history.clone(), L!("Alpha").to_owned());
|
|
test_history_matches!(searcher, vec![L!("Alpha")]);
|
|
history.remove(L!("Alpha"));
|
|
let mut searcher = HistorySearch::new(history.clone(), L!("Alpha").to_owned());
|
|
test_history_matches!(searcher, vec![]);
|
|
|
|
// Test history escaping and unescaping, yaml, etc.
|
|
let mut before: VecDeque<HistoryItem> = VecDeque::new();
|
|
let mut after: VecDeque<HistoryItem> = VecDeque::new();
|
|
history.clear();
|
|
let max = 100;
|
|
for i in 1..=max {
|
|
// Generate a value.
|
|
let mut value = WString::from_str("test item ") + &i.to_wstring()[..];
|
|
|
|
// Maybe add some backslashes.
|
|
if i % 3 == 0 {
|
|
value += L!("(slashies \\\\\\ slashies)");
|
|
}
|
|
|
|
// Generate some paths.
|
|
let mut paths = PathList::new();
|
|
for _ in 0..random::<usize>() % 6 {
|
|
paths.push(random_string());
|
|
}
|
|
|
|
// Record this item.
|
|
let mut item =
|
|
HistoryItem::new(value, SystemTime::now(), 0, history::PersistenceMode::Disk);
|
|
item.set_required_paths(paths);
|
|
before.push_back(item.clone());
|
|
history.add(item, false);
|
|
}
|
|
history.save();
|
|
|
|
// Empty items should just be dropped (#6032).
|
|
history.add_commandline(L!("").into());
|
|
assert!(!history.item_at_index(1).unwrap().is_empty());
|
|
|
|
// Read items back in reverse order and ensure they're the same.
|
|
for i in (1..=100).rev() {
|
|
after.push_back(history.item_at_index(i).unwrap());
|
|
}
|
|
assert_eq!(before.len(), after.len());
|
|
for i in 0..before.len() {
|
|
let bef = &before[i];
|
|
let aft = &after[i];
|
|
assert_eq!(bef.str(), aft.str());
|
|
assert_eq!(bef.timestamp(), aft.timestamp());
|
|
assert_eq!(bef.get_required_paths(), aft.get_required_paths());
|
|
}
|
|
|
|
// Items should be explicitly added to the history.
|
|
history.add_commandline(L!("test-command").into());
|
|
assert!(history_contains(&history, L!("test-command")));
|
|
|
|
// Clean up after our tests.
|
|
history.clear();
|
|
}
|
|
|
|
// Wait until the next second.
|
|
fn time_barrier() {
|
|
let start = SystemTime::now();
|
|
loop {
|
|
std::thread::sleep(std::time::Duration::from_millis(1));
|
|
if SystemTime::now()
|
|
.duration_since(UNIX_EPOCH)
|
|
.unwrap()
|
|
.as_secs()
|
|
!= start.duration_since(UNIX_EPOCH).unwrap().as_secs()
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
fn generate_history_lines(item_count: usize, idx: usize) -> Vec<WString> {
|
|
let mut result = Vec::with_capacity(item_count);
|
|
for i in 0..item_count {
|
|
result.push(sprintf!("%lu %lu", idx, i));
|
|
}
|
|
result
|
|
}
|
|
|
|
fn test_history_races_pound_on_history(item_count: usize, idx: usize) {
|
|
let _cleanup = test_init();
|
|
// Called in child thread to modify history.
|
|
let hist = History::new(L!("race_test"));
|
|
let hist_lines = generate_history_lines(item_count, idx);
|
|
for line in hist_lines {
|
|
hist.add_commandline(line);
|
|
hist.save();
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
#[serial]
|
|
fn test_history_races() {
|
|
let _cleanup = test_init();
|
|
// This always fails under WSL
|
|
if is_windows_subsystem_for_linux(WSL::V1) {
|
|
return;
|
|
}
|
|
|
|
// This fails too often on Github Actions,
|
|
// leading to a bunch of spurious test failures on unrelated PRs.
|
|
// For now it's better to disable it.
|
|
// TODO: Figure out *why* it does that and fix it.
|
|
if std::env::var_os("CI").is_some() {
|
|
return;
|
|
}
|
|
|
|
// Testing history race conditions
|
|
|
|
// Test concurrent history writing.
|
|
// How many concurrent writers we have
|
|
const RACE_COUNT: usize = 4;
|
|
|
|
// How many items each writer makes
|
|
const ITEM_COUNT: usize = 256;
|
|
|
|
// Ensure history is clear.
|
|
History::new(L!("race_test")).clear();
|
|
|
|
// history::CHAOS_MODE.store(true);
|
|
|
|
let mut children = Vec::with_capacity(RACE_COUNT);
|
|
for i in 0..RACE_COUNT {
|
|
children.push(std::thread::spawn(move || {
|
|
test_history_races_pound_on_history(ITEM_COUNT, i);
|
|
}));
|
|
}
|
|
|
|
// Wait for all children.
|
|
for child in children {
|
|
child.join().unwrap();
|
|
}
|
|
|
|
// Compute the expected lines.
|
|
let mut expected_lines: [Vec<WString>; RACE_COUNT] =
|
|
std::array::from_fn(|i| generate_history_lines(ITEM_COUNT, i));
|
|
|
|
// Ensure we consider the lines that have been outputted as part of our history.
|
|
time_barrier();
|
|
|
|
// Ensure that we got sane, sorted results.
|
|
let hist = History::new(L!("race_test"));
|
|
history::CHAOS_MODE.store(false);
|
|
|
|
// History is enumerated from most recent to least
|
|
// Every item should be the last item in some array
|
|
let mut hist_idx = 0;
|
|
loop {
|
|
hist_idx += 1;
|
|
let Some(item) = hist.item_at_index(hist_idx) else {
|
|
break;
|
|
};
|
|
|
|
let mut found = false;
|
|
for list in &mut expected_lines {
|
|
let Some(position) = list.iter().position(|elem| *elem == item.str()) else {
|
|
continue;
|
|
};
|
|
|
|
// Remove everything from this item on
|
|
let removed = list.splice(position.., []);
|
|
for line in removed.into_iter().rev() {
|
|
println!("Item dropped from history: {line}");
|
|
}
|
|
|
|
found = true;
|
|
break;
|
|
}
|
|
if !found {
|
|
println!(
|
|
"Line '{}' found in history, but not found in some array",
|
|
item.str()
|
|
);
|
|
for list in &expected_lines {
|
|
if !list.is_empty() {
|
|
printf!("\tRemaining: %ls\n", list.last().unwrap())
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// +1 to account for history's 1-based offset
|
|
let expected_idx = RACE_COUNT * ITEM_COUNT + 1;
|
|
assert_eq!(hist_idx, expected_idx);
|
|
|
|
for list in expected_lines {
|
|
assert_eq!(list, Vec::<WString>::new(), "Lines still left in the array");
|
|
}
|
|
hist.clear();
|
|
}
|
|
|
|
#[test]
|
|
#[serial]
|
|
fn test_history_merge() {
|
|
let _cleanup = test_init();
|
|
// In a single fish process, only one history is allowed to exist with the given name But it's
|
|
// common to have multiple history instances with the same name active in different processes,
|
|
// e.g. when you have multiple shells open. We try to get that right and merge all their history
|
|
// together. Test that case.
|
|
const COUNT: usize = 3;
|
|
let name = L!("merge_test");
|
|
let hists = [History::new(name), History::new(name), History::new(name)];
|
|
let texts = [L!("History 1"), L!("History 2"), L!("History 3")];
|
|
let alt_texts = [
|
|
L!("History Alt 1"),
|
|
L!("History Alt 2"),
|
|
L!("History Alt 3"),
|
|
];
|
|
|
|
// Make sure history is clear.
|
|
for hist in &hists {
|
|
hist.clear();
|
|
}
|
|
|
|
// Make sure we don't add an item in the same second as we created the history.
|
|
time_barrier();
|
|
|
|
// Add a different item to each.
|
|
for i in 0..COUNT {
|
|
hists[i].add_commandline(texts[i].to_owned());
|
|
}
|
|
|
|
// Save them.
|
|
for hist in &hists {
|
|
hist.save()
|
|
}
|
|
|
|
// Make sure each history contains what it ought to, but they have not leaked into each other.
|
|
#[allow(clippy::needless_range_loop)]
|
|
for i in 0..COUNT {
|
|
for j in 0..COUNT {
|
|
let does_contain = history_contains(&hists[i], texts[j]);
|
|
let should_contain = i == j;
|
|
assert_eq!(should_contain, does_contain);
|
|
}
|
|
}
|
|
|
|
// Make a new history. It should contain everything. The time_barrier() is so that the timestamp
|
|
// is newer, since we only pick up items whose timestamp is before the birth stamp.
|
|
time_barrier();
|
|
let everything = History::new(name);
|
|
for text in texts {
|
|
assert!(history_contains(&everything, text));
|
|
}
|
|
|
|
// Tell all histories to merge. Now everybody should have everything.
|
|
for hist in &hists {
|
|
hist.incorporate_external_changes();
|
|
}
|
|
|
|
// Everyone should also have items in the same order (#2312)
|
|
let hist_vals1 = hists[0].get_history();
|
|
for hist in &hists {
|
|
assert_eq!(hist_vals1, hist.get_history());
|
|
}
|
|
|
|
// Add some more per-history items.
|
|
for i in 0..COUNT {
|
|
hists[i].add_commandline(alt_texts[i].to_owned());
|
|
}
|
|
// Everybody should have old items, but only one history should have each new item.
|
|
#[allow(clippy::needless_range_loop)]
|
|
for i in 0..COUNT {
|
|
for j in 0..COUNT {
|
|
// Old item.
|
|
assert!(history_contains(&hists[i], texts[j]));
|
|
|
|
// New item.
|
|
let does_contain = history_contains(&hists[i], alt_texts[j]);
|
|
let should_contain = i == j;
|
|
assert_eq!(should_contain, does_contain);
|
|
}
|
|
}
|
|
|
|
// Make sure incorporate_external_changes doesn't drop items! (#3496)
|
|
let writer = &hists[0];
|
|
let reader = &hists[1];
|
|
let more_texts = [
|
|
L!("Item_#3496_1"),
|
|
L!("Item_#3496_2"),
|
|
L!("Item_#3496_3"),
|
|
L!("Item_#3496_4"),
|
|
L!("Item_#3496_5"),
|
|
L!("Item_#3496_6"),
|
|
];
|
|
for i in 0..more_texts.len() {
|
|
// time_barrier because merging will ignore items that may be newer
|
|
if i > 0 {
|
|
time_barrier();
|
|
}
|
|
writer.add_commandline(more_texts[i].to_owned());
|
|
writer.incorporate_external_changes();
|
|
reader.incorporate_external_changes();
|
|
for text in more_texts.iter().take(i) {
|
|
assert!(history_contains(reader, text));
|
|
}
|
|
}
|
|
everything.clear();
|
|
}
|
|
|
|
#[test]
|
|
#[serial]
|
|
fn test_history_path_detection() {
|
|
let _cleanup = test_init();
|
|
// Regression test for #7582.
|
|
let tmpdirbuff = CString::new("/tmp/fish_test_history.XXXXXX").unwrap();
|
|
let tmpdir = unsafe { libc::mkdtemp(tmpdirbuff.into_raw()) };
|
|
let tmpdir = unsafe { CString::from_raw(tmpdir) };
|
|
let mut tmpdir = str2wcstring(tmpdir.to_bytes());
|
|
if !tmpdir.ends_with('/') {
|
|
tmpdir.push('/');
|
|
}
|
|
|
|
// Place one valid file in the directory.
|
|
let filename = L!("testfile");
|
|
std::fs::write(wcs2osstring(&(tmpdir.clone() + &filename[..])), []).unwrap();
|
|
|
|
let test_vars = EnvStack::new();
|
|
test_vars.set_one(L!("PWD"), EnvMode::GLOBAL, tmpdir.clone());
|
|
test_vars.set_one(L!("HOME"), EnvMode::GLOBAL, tmpdir.clone());
|
|
|
|
let history = History::with_name(L!("path_detection"));
|
|
history.clear();
|
|
assert_eq!(history.size(), 0);
|
|
history.clone().add_pending_with_file_detection(
|
|
L!("cmd0 not/a/valid/path"),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
&(L!("cmd1 ").to_owned() + filename),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
&(L!("cmd2 ").to_owned() + &tmpdir[..] + L!("/") + filename),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
&(L!("cmd3 $HOME/").to_owned() + filename),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
L!("cmd4 $HOME/notafile"),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
&(L!("cmd5 ~/").to_owned() + filename),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
L!("cmd6 ~/notafile"),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
L!("cmd7 ~/*f*"),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.clone().add_pending_with_file_detection(
|
|
L!("cmd8 ~/*zzz*"),
|
|
&test_vars,
|
|
history::PersistenceMode::Disk,
|
|
);
|
|
history.resolve_pending();
|
|
|
|
const hist_size: usize = 9;
|
|
assert_eq!(history.size(), 9);
|
|
|
|
// Expected sets of paths.
|
|
let expected_paths = [
|
|
vec![], // cmd0
|
|
vec![filename.to_owned()], // cmd1
|
|
vec![tmpdir + L!("/") + filename], // cmd2
|
|
vec![L!("$HOME/").to_owned() + filename], // cmd3
|
|
vec![], // cmd4
|
|
vec![L!("~/").to_owned() + filename], // cmd5
|
|
vec![], // cmd6
|
|
vec![], // cmd7 - we do not expand globs
|
|
vec![], // cmd8
|
|
];
|
|
|
|
let maxlap = 128;
|
|
for _lap in 0..maxlap {
|
|
let mut failures = 0;
|
|
for i in 1..=hist_size {
|
|
if history.item_at_index(i).unwrap().get_required_paths()
|
|
!= expected_paths[hist_size - i]
|
|
{
|
|
failures += 1;
|
|
}
|
|
}
|
|
if failures == 0 {
|
|
break;
|
|
}
|
|
// The file detection takes a little time since it occurs in the background.
|
|
// Loop until the test passes.
|
|
std::thread::sleep(std::time::Duration::from_millis(2));
|
|
}
|
|
history.clear();
|
|
}
|
|
|
|
fn install_sample_history(name: &wstr) {
|
|
let path = path_get_data().expect("Failed to get data directory");
|
|
std::fs::copy(
|
|
env!("CARGO_MANIFEST_DIR").to_owned()
|
|
+ "/tests/"
|
|
+ std::str::from_utf8(&wcs2string(&name[..])).unwrap(),
|
|
wcs2osstring(&(path + L!("/") + &name[..] + L!("_history"))),
|
|
)
|
|
.unwrap();
|
|
}
|
|
|
|
#[test]
|
|
#[serial]
|
|
fn test_history_formats() {
|
|
let _cleanup = test_init();
|
|
// Test inferring and reading legacy and bash history formats.
|
|
let name = L!("history_sample_fish_2_0");
|
|
install_sample_history(name);
|
|
let expected: Vec<WString> = vec![
|
|
"echo this has\\\nbackslashes".into(),
|
|
"function foo\necho bar\nend".into(),
|
|
"echo alpha".into(),
|
|
];
|
|
let test_history_imported = History::with_name(name);
|
|
assert_eq!(test_history_imported.get_history(), expected);
|
|
test_history_imported.clear();
|
|
|
|
// Test bash import
|
|
// The results are in the reverse order that they appear in the bash history file.
|
|
// We don't expect whitespace to be elided (#4908: except for leading/trailing whitespace)
|
|
let expected: Vec<WString> = vec![
|
|
"EOF".into(),
|
|
"sleep 123".into(),
|
|
"posix_cmd_sub $(is supported but only splits on newlines)".into(),
|
|
"posix_cmd_sub \"$(is supported)\"".into(),
|
|
"a && echo valid construct".into(),
|
|
"final line".into(),
|
|
"echo supsup".into(),
|
|
"export XVAR='exported'".into(),
|
|
"history --help".into(),
|
|
"echo foo".into(),
|
|
];
|
|
let test_history_imported_from_bash = History::with_name(L!("bash_import"));
|
|
let file =
|
|
std::fs::File::open(env!("CARGO_MANIFEST_DIR").to_owned() + "/tests/history_sample_bash")
|
|
.unwrap();
|
|
test_history_imported_from_bash.populate_from_bash(BufReader::new(file));
|
|
assert_eq!(test_history_imported_from_bash.get_history(), expected);
|
|
test_history_imported_from_bash.clear();
|
|
|
|
let name = L!("history_sample_corrupt1");
|
|
install_sample_history(name);
|
|
// We simply invoke get_string_representation. If we don't die, the test is a success.
|
|
let test_history_imported_from_corrupted = History::with_name(name);
|
|
let expected: Vec<WString> = vec![
|
|
"no_newline_at_end_of_file".into(),
|
|
"corrupt_prefix".into(),
|
|
"this_command_is_ok".into(),
|
|
];
|
|
assert_eq!(test_history_imported_from_corrupted.get_history(), expected);
|
|
test_history_imported_from_corrupted.clear();
|
|
}
|