Open mkpro118 opened 1 month ago
Attempting to fix this in PR #46, but at least on windows, the std::fs::remove_dir_all
fails to unlink the temporary directories.
The call to remove_dir_all
fails with the error
Err(
Os {
code: 32,
kind: Uncategorized,
message: "The process cannot access the file because it is being used by another process."
}
)
This clearly suggests another process has an open handle to the temporary directory. Unfortunately I have no idea which process could possibly have an open handle. Is it cargo that has an open handle?
With some debug logging, as of commit 4dbc65a
, running cargo test --tests test_cat_file -- --nocapture
produces (obviously prone to race conditions) the following output ($TEMPDIR is the path to the system's tempdir)
running 3 tests
incr: 1
incr: 2
decr: 1
incr: 2
test core::test_cat_file::tests::test_cmd_cat_file_testfile ... ok
decr: 1
test core::test_cat_file::tests::test_cmd_cat_file_readme ... ok
decr: 0
Dropping MUTEX!!!!
"$TEMPDIR\\cmd_cat_file1721716735083089700"
CLEANING UP "$TEMPDIR\\cmd_cat_file1721716735083089700"
RESULT Err(Os { code: 32, kind: Uncategorized, message: "The process cannot access the file because it is being used by another process." })
Ok(/* directory where the test command was executed */)
"$TEMPDIR\\cmd_cat_file1721716735083089700" exists? true
test core::test_cat_file::tests::test_cmd_cat_file_bad_file ... ok
However the "Dropping MUTEX" is always after all the test are complete, so there really should not be any open handles to the files in the temporary directory. Additionally, all FS operations are done with mutexes, so it should not be possible for any handles to not be closed due to race conditions.
Okay this is super weird, but the deletion works when run with many other tests.
I reduced some output verbosity in tests in commit b554361
, and only print in TempDir::drop if remove_dirs_all
returns Err
.
Running cargo test --tests core::test_ -- --nocapture
, I get the following output,
running 19 tests
test core::test_init::tests::test_cmd_init_extra_args ... ok
Deleting "$TEMPDIR\\cmd_init_no_args1721719759369800200"
test core::test_hash_object::tests::test_cmd_hash_object_testfile ... ok
incr: 1
incr: 2
test core::test_init::tests::test_cmd_init_path ... ok
test core::test_hash_object::tests::test_cmd_hash_object_readme_write ... ok
test core::test_ls_tree::tests::test_dir1_no_recursive ... ok
decr: 1
incr: 2
test core::test_cat_file::tests::test_cmd_cat_file_testfile ... ok
Deleting "$TEMPDIR\\cmd_init_no_args1721719759370316300"
Deleting "$TEMPDIR\\cmd_init_no_args1721719759370517600"
decr: 1
test core::test_ls_tree::tests::test_dir2_no_recursive ... ok
test core::test_hash_object::tests::test_cmd_hash_object_testfile_write ... ok
test core::test_hash_object::tests::test_cmd_hash_object_readme ... ok
decr: 0
Dropping MUTEX!!!!
Deleting "$TEMPDIR\\cmd_cat_file1721719759369280000" /* Deletion somehow worked */
// A bunch of other test results omitted
However the output is unchanged on running tests with cargo test --tests test_cat_file -- --nocapture
.
Running the full test suite with cargo test --tests -- --nocapture
also fails to unlink the temporary directory. Very weird.
The Problem
The core command tests do not clean up the temporary directories they create in setup after tests are done. In particular, the following tests fail to remove the tempdirs,
tests::core::test_cat_file
tests::core::test_hash_object
tests::core::test_init
tests::core::test_ls_tree
tests::utils::test_fnmatch
mini_git::utils::path
Anticipated Cause
I believe the error is with
TempDir
s that are being used with staticMutex
es, asdrop
is not called on them, no cleanup happens with the interior data.Proposed Solution
So far the only solution I believe is guaranteed to work is:
cleanup
function after each test.cleanup
would decrement the count of testsdrop
on theTempDir
when the count reaches zero.I am not a big fan of this since it requires manually setting the number of tests, which is harder to maintain and prone to error.
Essentially, the trivial fix would be something along the lines of