Support AIX operating system

[ecnelises]

This change makes this crate support detecting timezone on AIX. Since AIX support hasn't been merged yet, this is in draft status.

[Kijewski]

Thank you for working on this!

According to https://www.ibm.com/docs/en/aix/7.1?topic=files-environment-file and https://unixhealthcheck.com/blog?id=291 AIX has its own time zone naming scheme, incompatible with the names in the Time Zone Database. So I guess there needs to be some translation step, that translates (at least) the standard time zones to one of the cities, e.g. "CET-1CEST" → "Europe/Berlin".

The other implementations don't evaluate the environment variables by design. I would rather scan /etc/environment for TZ=… like in the Solaris/Illumos implementation.

[ecnelises]

Hi @Kijewski , do you mean that the (timezone, offset, daylight timezone, offset) form should be transformed into Region/City form?

If so, mapping from (timezone, offset) to city is a one-to-many relation. For example, CST-8 means either Asia/Chongqing Asia/Shanghai or Asia/Harbin. So arbitrary will be chosen?

[ecnelises]

Gentle ping.. any further comments?

[Kijewski]

I would scan /etc/environment for TZ=…, same as in illumos. This way the logic is the same as in our other platform implementations. Maybe you don't even need to add a new file, and you can re-use tz_illumos.rs.

On AIX you can use Olson and POSIX time zones (https://developer.ibm.com/articles/au-aix-posix/). For no obvious reason the American POSIX time zones like PST8PDT are also valid IANA time zone identifiers, but e.g. "CET-1CEST" is not.

I guess we can only hope that the system time zone is an IANA identifier. I would not like to add a white list. We could filter out results like CET-1CEST-2,M3.5.0/02:00:00,M10.5.0/03:00:00 (anything that contains a comma), strip off everything after the comma, or let the caller decide how to use the information.

@astraw, what do you think?

[astraw]

Why is this still a draft?

I'm far from an AIX expert nor IANA database expert, but this all seems OK to me.

Although new to me, as far as I understand https://web.archive.org/web/20220525113023/https://www.ietf.org/timezones/tzdb-2018g/theory.html, POSIX time zones are valid Olson tz names, so I don't think we need to do any filtering or parsing.

[astraw]

@Kijewski do think the logic in my comment above is acceptable and we can merge this?

[Kijewski]

As far as I understand it, every name in the Olson database is a valid value for a POSIX tz string, but the reverse is not true.

I am only mildly hesitant if we should allow POSIX tz strings, because a downstream consumer might assume that we are only returning valid names, and try to put it in a path name. E.g. in windows a colon can be used to for file stream selection, but with AIX being a Unix implementation, I guess that a colon does not pose a security risk. In Unix colons are sometimes used to separate paths, e.g. in the $PATH variable, but I don't think anyone would/should paste their time zone in such a variable anyway.

And we aren't checking if the result is sane for any other OS, so I guess this PR is at least good enough. If the implementation can be made better in the future, then we can still update it.

[astraw]

These are valid concerns and perhaps we should have a sanity check on the output shared across all implementations. I don't think that's a specific issue with this AIX implementation. Therefore I'll go ahead and merge this now and we can consider sanity checks in a platform independent way.

[ecnelises]

Thanks!