fix(tar): append slash to top-level directory mtree entries

[sin-ack]

bsdtar's mtree format has a quirk wherein entries without "/" in their first word are treated as "relative" entries, and "relative" directories will cause tar to "change directory" into the declared directory entry. If such a directory is followed by a "relative" entry, then the file will be created within the directory, instead of at top-level as expected. To mitigate, we append a slash to top-level directory entries.

Fixes #851.

---

Changes are visible to end-users: no

Test plan

• New test cases added: //:tar_test13, //lib/tests/tar:test14

[CLAassistant]

All committers have signed the CLA.

[alexeagle]

Thanks - you should be able to find somewhere in https://github.com/aspect-build/bazel-lib/blob/main/lib/tests/tar/BUILD.bazel that tickles this case, and assert that the new mtree matches some string

[sin-ack]

I tried to introduce a test, however the main problem is that any file I create with file_write is created inside lib/tests/tar/, and mtree_spec doesn't hit the bug. That also made me realize that we need to add this logic to mtree_mutate (because of strip_prefix), but that doesn't solve the problem that we can't test mtree_spec by itself. What should I do here?

[alexeagle]

Simple answer would be to add the test case to the root package in this repo with a comment about why we're doing something unusual

[sin-ack]

Changes since last push:

• mtree_mutate now also correctly handles directory entries with a single component when strip_prefix is given
• Added tests for mtree_spec and mtree_mutate

[sin-ack]

Could I get another look at this please? :)

[alexeagle]

Sorry the CI doesn't trigger for you automatically. Looks like there are some errors, can you reproduce them?

[sin-ack]

Interesting, it seems like the output order of mtree_spec is either random or differs between my machine and the CI. Perhaps I could write a genrule that orders the output of mtree in such a way to trigger the bug.

[sin-ack]

Sorry for the delay, I was on vacation. I've made the tests run the sort command on the output, which should hopefully make the order consistent between systems now.

@alexeagle Could I get a re-run of the tests please?

[sin-ack]

Thanks!