pjdfstest rename/21.t

https://github.com/pjd/pjdfstest/blob/c711b5f6b666579846afba399a998f74f60c488b/tests/rename/21.t#L26

expect 0 mkdir ${n2}/${n0} 0700                                                  
expect "0|EACCES" -u 65534 -g 65534 rename ${n2}/${n0} ${n2}/${n1}               
expect "0|EACCES" -u 65534 -g 65534 rename ${n2}/${n1} ${n2}/${n0}

rename/21.t renames ${n2}/${n0} to ${n2}/${n1} while it doesn't have write permission to ${n2}/${n0}. The result can be 0 or EACCES. However, if the filesystem returns EACCES, it will return ENOENT to the next test because ${n2}/${n1} doesn't exist. The test needs to remove and recreate the directory or allow ENOENT in the next test.

Jan 23 '24 10:01 zhanyang

Are you saying that the rest of the test contained spurious failures? What OS and FS are you using?

Jan 23 '24 14:01 asomers

I was writing a filesystem on Linux. There is a bug that messed up on the permission bits and failed a lot of tests. For rename/21.t, it passed the 1st rename with EACCES and failed the 2nd rename with ENOENT. I looked at the test and felt that the filesystem is doing things correctly. Later, I fixed the bug and passed the test.

My question is that if failure (EACCES) is allowed in the first rename, why the second rename cannot return ENOENT?

Jan 24 '24 04:01 zhanyang

You're correct. The second rename on line 26 should allow ENOENT, if the first one failed. Would you care to submit a patch?

Jan 24 '24 14:01 asomers