Building phobos with wine fails

[Jonathan M Davis via Digitalmars-d]

On Monday, July 03, 2017 15:43:57 Shachar Shemesh via Digitalmars-d wrote:
> On 03/07/17 15:35, Jonathan M Davis via Digitalmars-d wrote:
> > Well, that looks bad, and it really should be working better than that,
> > but I would point out that (at least, the last time I checked), wine
> > does not behave correctly with regards to some of the date/time stuff,
> > so some of the std.datetime tests will fail with wine, and fixing that
> > requires that wine start behaving like Windows like it's supposed to.
>
> Do you have specific examples? My Wine hacking days are long behind me,
> but I'm fairly sure such easy wins would be simple enough fix to get
> into Wine.
>
> The Wine charter is to be "bug for bug compatible with Windows", so such
> changes should not have any trouble to make it in, assuming we can
> supply a test that proves that's the case.

Two come to mind:

1. The functions for converting to and from a timezone take a struct
containing information about what that timezone is - including the DST
switches. The way that MS did that orginally involved specifying something
like the nth instance of the xth day of the week in a particular month (e.g.
the first Sunday in April), and that was for all time, as if time zones
never changed when they have DST switches - which of course, doesn't work
well for many timezones (there's even a wikipedia article on how Windows
kept having the time wrong in Israel because of this). MS didn't bother to
fix this until the US and Canada changed when DST was going to be starting
in 2007.

To fix this for Vista, MS added a new version of these functions which
assume that the DST rules are good for a particular year (which still isn't
entirely correct but at least is _much_ closer), but they couldn't
retroactively add that to existing programs. Anything using the old
functions would still be broken. Their solution seems to have been to make
it so that the old versions of those functions look at the struct that
they're given, figure out which timezone they're for, and then ignore what's
in the struct and use whatever the rules are supposed to be for that
particular date. Without that change, you end up using the old DST rules,
and last time I checked, wine had not made that change. So, if you used the
old functions (which IIRC, std.datetime does, because at the time it was
written, we were trying to support Windows XP), and you use wine, then DST
will be wrong for some dates which would be correct on Windows. I had
actually originally developed std.datetime's Windows stuff using wine, so I
had buggy code because of this, which was caught as soon as someone (I
forget whether it was me or someone else) actually ran the tests on an
actual Windows box.

2. The time zone conversion stuff in Windows is unfortunately defined in the
registry, and AFAIK, the only way to look up an arbitrary time zone instead
of using local time or UTC is to read in the information from the registry
and fill in the aforementioned struct with time zone information with that
information. The time zones are listed in the registry based on the English
names that MS uses - e.g. "Pacific Standard Time" is what they use for the
Pacific time zone (though it has both the standard time and DST information
in there). Pretty much every other OS uses the time zone database for their
time zone information, and that uses major cities to distinguish time zones
(e.g. the Pacific time zone would be "America/Los_Angeles"), not something
like "Pacific Standard Time". wine chose to use the time zone database names
like "America/Los_Angeles" in their registry instead of the names that
Windows uses. So, any Windows program that tries to look up a time zone with
wine will fail, because wine is not using the names that Windows uses even
though they're supposedly mimicking the Windows registry. To work around
that, when I wrote std.datetime, I was actually looking at each time zone
registry entry to look at the name it had for standard time (since it was
the same name as the registry key on Windows), so it _was_ compatible with
wine. However, as it turns out, the values in the registry entry are
locale-dependent (unlike the registry key itself, which is always in
Engish). So, the tests failed on systems for folks outside the US (e.g.
Kenji, being in Japan, ran into the issue and was actually the one to fix it
so that std.datetime used the registry key rather than looking into the
entry for the time zone name).

I have not actually tested any of that with wine recently, so it's possible
that they've fixed one or both of those issues or that it's fixed if you
tell it that you're using a newer version of Windows, but the last time I
checked, both of those were still broken on wine. And I really should have
created issues in the wine bug tracker for them, but I've just never gotten
around to it.

But it's the sort of thing that you probably don't catch unless you're
dealign with specific time zone stuff in a way that most programs don't do.

- Jonathan M Davis