Test $6$rounds=10$roundstoolow fails with libxcrypt

[mweinelt]

Hi,

we are replacing glibc's crypt functionality with that of libxcrypt in nixpkgs and we are seeing one test failure.

Would you mind looking into that?

crypt-sha> Running 1 test suites...
crypt-sha> Test suite example: RUNNING...
crypt-sha> Tests
crypt-sha>   cryptSHA512
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$rounds=10000$saltstringsaltstring: OK
crypt-sha>     $6$rounds=5000$toolongsaltstring:     OK
crypt-sha>     $6$rounds=1400$anotherlongsaltstring: OK
crypt-sha>     $6$rounds=77777$short:                OK (0.05s)
crypt-sha>     $6$rounds=123456$asaltof16chars..:    OK (0.05s)
crypt-sha>     $6$rounds=10$roundstoolow:            OK
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$saltstring:                        OK
crypt-sha>   cryptSHA512Raw
crypt-sha>     saltstring Nothing:                   OK
crypt-sha>     saltstringsaltstring Just 10000:      OK
crypt-sha>     toolongsaltstring Just 5000:          OK
crypt-sha>     anotherlongsaltstring Just 1400:      OK
crypt-sha>     short Just 77777:                     OK (0.05s)
crypt-sha>     asaltof16chars.. Just 123456:         OK (0.05s)
crypt-sha>     roundstoolow Just 10:                 OK
crypt-sha>   crypt
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$rounds=10000$saltstringsaltstring: OK
crypt-sha>     $6$rounds=5000$toolongsaltstring:     OK
crypt-sha>     $6$rounds=1400$anotherlongsaltstring: OK
crypt-sha>     $6$rounds=77777$short:                OK (0.05s)
crypt-sha>     $6$rounds=123456$asaltof16chars..:    OK (0.04s)
crypt-sha>     $6$rounds=10$roundstoolow:            FAIL
crypt-sha>       test/Example.hs:96:
crypt-sha>       expected: Just "$6$rounds=1000$roundstoolow$kUMsbe306n21p9R.FRkW3IGn.S9NPN0x50YhH1xhLsPuWGsUSklZt58jaTfF4ZEQpyUNGc0dqbpBYYBaHHrsX."
crypt-sha>        but got: Nothing
crypt-sha>       Use -p '/crypt.$6$rounds=10$roundstoolow/' to rerun this test only.
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$saltstring:                        OK
crypt-sha>     $6$saltstring:                        OK
crypt-sha>   glibc and native impl comparison:       OK (0.51s)
crypt-sha>     +++ OK, passed 100 tests.
crypt-sha> 
crypt-sha> 1 out of 28 tests failed (0.86s)
crypt-sha> Test suite example: FAIL
crypt-sha> Test suite logged to: dist/test/crypt-sha512-0-example.log
crypt-sha> 0 of 1 test suites (0 of 1 test cases) passed.
error: builder for '/nix/store/43qcy1vr31zsknf2j5lg3sq2nwgpvyz6-crypt-sha512-0.drv' failed with exit code 1;

[phadej]

This looks like a difference between libxcrypt and glibc's crypt implementation, and libxcrypt implementation doesn't mention anything about round too low behavior differences.

I consider this a bug in libxcrypt until there is a reference showing it's intended behavior and why so.