Jim Meyering <address@hidden> writes:
> please do at least simulate it once, e.g., on Linux by tweaking
> lib/openat.c so that its open always fails and it falls through to
> the save_cwd/restore_cwd code. Then, run rm from a write-only
> directory (forcing save_cwd to use getcwd) with two relative-named
> non-empty hierarchies. When rm is in the first openat call, and has
> changed to the top of the first hierarchy, rename the initial
> working directory so that the first restore_cwd call fails. That
> should provoke the new diagnostic.
Yes, it did:
/home/eggert/src/gnu/cu-tune/src/rm: failed to return to initial working
directory: No such file or directory
Thanks for the scenario. Too bad we can't easily make tests for this
stuff....
I checked in the change.