The equals signs are included in the day-of-transition capture sign.

Not all time specs need to have a time type associated with them - only one particular kind. This commit mandates that, by removing the type from the struct.

Also, rename RulesSave to Saving, which is what it's being called everywhere else.

This code uses an algorithm cribbed from the C examples in the tzinfo database itself, so they're actually more correct than whatever I was using!

- Rename GMT to UTC wherever it's used. It means pretty much the same, but is more modern (I hope this commit message isn't used to make fun of me in the future)
- Use the Time Type enum in datetime - there's no point having two of them.
- Add a lookup function in the data crate.

This introduces a Format enum, which parses the formats beforehand, and figures out which to use based on that. It still panicks, though.

This makes it easier to test whether things have changed between crate compilations.

Add a test for the transition-optimisation function, which uses some of the Antarctica/Macquarie rules. However, fix all the other tests to get it to compile at all!

- I had to replace some of the end times (123456 and 234567) with their ZoneTime equivalent.
- This required some of the fields for the spec structs in line.rs to be made public.

This makes the code more Rustful, and has no changes to the resulting output.

This makes it compile without warnings again!

The error in question appears to have gone away! Current version:

rustc 1.5.0-nightly (95fb8d1c8 2015-10-27)

Commit 2551f9224a1c188cc7f963f504b27c8cea4d92ae to datetime changes the Transition struct to only have one offset field, rather than two. This commit changes the data crate builder to use the new version of the struct.

Also, work around a linker error that I'm not even sure why I'm getting.

This makes it easier to see how the consolidated offset was produced.

This marks a change to the way transition data was calculated and stored.

Previously, each timespan—a period of consistent time zone name, GMT offset, and DST offset—was stored as a transition with an optional timestamp at which the transition occurs. This doesn’t solve the problem of what the time details should be *before* the first transition has occurred, which is why the timestamp was optional: the first period would simply have a `None` timestamp.

This approach had problems, namely:

1. It was possible to have multiple `None` timestamps, which would confuse the datetime calculator;
2. It was not only possible but actually manifest to have time zones with *zero* `None` timestamps, which would also confuse the datetime calculator.

To remove the possibility of having invalid data, each time zone now has one fixed “ante” period, and at least zero transitions, each of which now has a non-Optional timestamp.

It’s a rather simple change, but most of the terminology has been updated to reflect this—Transition to ZoneDetails, among other structs—which is where most of the lines in the diff come from.

The only time zones that change in zoneinfo-data are the “fixed” ones like CET or EST5EDT.

This uses the PartialEq implementation on ZoneDetails to simplify some of the code, which was getting rather repetitive.

build-data-crate.rs now uses the new ZoneSet instead of Transition, which is great because Transition doesn't exist anymore.

See 74f69f6 for more information on why this is necessary.

ZoneDetails to FixedTimespan, and ZoneSet to FixedTimespanSet. Also, add the transition instant time for clarity.

Many types have been moved to the root of the crate, with others (such as TimeType) accessible from a different module now.

See datetime@b86a5752678d2cf410b37c909309138895a61f8a for the commit that necessitated. this.

The ‘get’ function (that *will* be in the data crate’s root) was confused over the elided lifetimes of the ‘lookup’ function (in the data module). Adding an explicit 'static fixes this.

- Update nomenclature in test code (FixedTimespan, FixedTimespanSet, no more Transition)
- Add local mean times to the rules tests to make them stop failing (it’ll never be valid to have a time zone that *begins* with a rule—it should always have a local mean time first!)
- Removed the only feature flag to remove dependence on Rust Nightly

Although I’d usually be for encapsulating behaviour like this, there’s a lot of surrounding code in this module, and they just kind of get lost in the noise. Plus, this makes the structure of the Table type a bit more clear.

The timezone data crate now contains *all* the timezones, rather than just the ones listed in the zoneinfo files.

This completes the timezone data. All that’s left to do is the geolocation stuff.

See datetime@edb41e214d784cb59dca683029fe3867ebe11d50 for the change that broke the old code.

Commit datetime@e6012ba003a8f5949510dea0f9ff334234202b07 caused this. All the time zones we want to build are static (available throughout the program).

- to_string -> to_owned
- Eliminate redundant closures and bindings
- More easily clone a vector

...and print them out all in one go, rather than as they occur. This is work towards making the library more error-tolerant.

This separates the table module, which deals with table *construction*, and the transitions module, which deals with table *examination*.

Casualties include the missing_docs lint, which had to go as I had to make lots of stuff public, and a few re-exports in lib.

Attack of the killer apostrophes!

The explanatory timestamps in the comments had the wrong number months, and this way we can use the new ISO formatting that the datetime crate provides.

This moves most of the 'convert this time zone into a list of timespans' code to a new Builder type.

The refactoring isn't yet complete, hopefully obviously, but it's at a stable point when it compiles and the data crate output is not modified.

It highlights how many of the variables being changed on each iteration are only important for one or the other Saving variant. It also combines the code for the NoSaving and Fixed variants, as they differed only by one variable.

Next steps include fixing some of the warnings and getting rid of more per-iteration variables.

Trying to get a non-existent timezone out of a table used to panic... which worked well until I needed to write a test showing that it would fail! So this commit changes it to returning None instead.

Fix invalid lifetime name

Handle negative offsets, "Alternative" parsing error and comment lines