Data loss with --replay and hard-linked files due to missing checksums.

[misieck]

Under certain conditions, rmlint does not calculate checksums of encountered files. All entries without checksums are seemingly interpreted by --replay as one and the same file, irrelevant of their content. This results in deletion of all but one of such files.
One such condition that I found is when encountering hard links. Consider the following test setup:

\testdir
    - file_one 
    - file_one_link
    - file_two 
    - file_two_link

Sizes of file_one and file_two must differ. Running rmlint produces a rmlint.json file with no checksums in it, but the actions are correct.

# Duplicate(s):
    ls '/home/muser/rmlint/test/two'
    rm '/home/muser/rmlint/test/two_link'
    ls '/home/muser/rmlint/test/one'
    rm '/home/muser/rmlint/test/one_link'

Running rmlint --replay rmlint.json afterwards produces a script attempting to delete all but one of the listed files:

# Duplicate(s):
    ls '/home/muser/rmlint/test/one'
    rm '/home/muser/rmlint/test/one_link'
    rm '/home/muser/rmlint/test/two'
    rm '/home/muser/rmlint/test/two_link'

This would delete both two and two_link.

[vassilit]

I reproduce with version 2.10.2 compiled: Jan 3 2025 at [14:24:47] "Ludicrous Lemur" (rev f0ca57ec)
(HEAD of current master branch)

[lovesynth]

It would be great if anybody with permissions would like to put a big red warning into the documentation about this. I just very nearly lost a lot of files, I just out of paranoia checked the issue list before executing the commands, thinking to myself "ah, you are a goddamn misanthropic pessimist, I bet there is no deadly problem here"... just to find that there actually is one. I think it is also not too much expectation to put at least some warning into the docs, or on top of the readme of the repo to warn people. Also disabling the function immediately would be totally OK in this situation. Thank you very much.