Curtis Vogt

Related: https://github.com/JuliaLang/julia/issues/29909

> This is still a problem Your particular example will always be an issue. If you are passing in a `DateFormat` explicitly where you are trying to parse milliseconds the...

> Though an alternative might be to all astimezone on all of them, so they all become Variable, or they all become Fixed. I'm definitely against converting the time zones...

Just a thought: what if we introduced a new alternative type to `ZonedDateTime`? That should allow us to experiment with more breaking changes in a more opt-in manner.

> My current preferred solution is to introduce `UTCZonedDateTime`. Relevant: https://github.com/invenia/Intervals.jl/issues/178#issuecomment-901347323

I'll note the benchmarks above are all using https://github.com/JuliaTime/TimeZones.jl/pull/281

Looks like we don't always get better performance (Julia 1.5.1) Current master (70df06e) ```julia julia> using BenchmarkTools, TimeZones, Dates [ Info: Precompiling BenchmarkTools [6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf] [ Info: Precompiling TimeZones [f269a46b-ccf7-5d73-abea-4c690281aa53] julia>...

Nice find. We can probably address that by making `FixedTimeZone` a isbits type and having that directly accessible by the `ZonedDateTime` struct

> It doesn't seem to make vectors of `ZonedDateTimes` isbits Isn't that true for all isbits types? ```julia julia> xs = [1,2,3] 3-element Array{Int64,1}: 1 2 3 julia> isbits(xs[1]) true...

> Do we want to add some of these to the packages benchmark suite? Yes, we will. I do need to get the package benchmarks up to date but posting...