Daniel Markstedt

Discarding this. This code can probably be refactored to have a more consistent logical separation between return codes and DSI offsets, but I won't mess with it now.

The earlier test results were from `ea = none` actually. Or rather, leave ea undefined and let it fall back to *none*. Here are the results with `ea = ad`...

Out of curiosity, I got a FreeBSD environment set up. The tier-1 tests all pass (as on OpenBSD) while 14 of the tier-2 tests are failing. Yet another pattern! ```...

Discarding this. It's the most important that the Tier 1 spectests as passing, since they're the ones that cover the AFP commands proper. The Tier 2 tests are more hackish...

Turns out this happens only when exactly len 8 passwords are used.

The failure occurs when comparing the password string with the input buffer: ``` p = crypt(password, pwd->pw_passwd); if (strcmp(p, pwd->pw_passwd) != 0) ``` ...so I'm suspecting either garbage values or...

~~It's universal: All BSDs, Linux. Both PAM and passwd auth.~~

Pardon, the universal bug seems to be with the afpfs-ng client (afpcmd). Reverted the description of the bug back to impacting OpenBSD specifically. Neither NetBSD nor FreeBSD is impacted. Note...

This no longer reproduces in the bleeding edge main code today, against the same OpenBSD VM. Not sure if a subsequent code change had an impact, or if there was...

Good stuff! The python linter reports a minor issue. You can fix it automatically by running `black`. Please see https://github.com/PiSCSI/piscsi/blob/develop/python/README.md#static-analysis-and-formatting