Rafael Schouten

I figured something like that might have been the case. If we can get precompilation working nearly everywhere and functions are broken up enough that they are decoupled, we should...

Precompile in Parsers.jl helps a lot here: https://github.com/JuliaData/Parsers.jl/pull/108 TTFX with both of these changes is now about 5 seconds. (but there are quite a few bugs in this PR, and...

There another 30-40% TTFX improvement in some of this code, but some changes are unnecessary now - this branch had larger gains without the Parsers.jl fix. The changes here that...

Probably restructuring Context and the preloaded of that would be the most obvious next step.

Type instability pretty much everywhere in very large functions. There are easy improvements everywhere just by adding function barriers and splitting things up. For example, moving the `ctx.threaded` code into...

@jishnub I have to disagree that an error not throwing is not serious type piracy. When OffsetArrays.jl is just a secondary dependency not loaded by the user, it can still...

The most obvious one from my other issue: I typed something like `zeros(10:20)` instead of `zero(10:20)`. Hurrah, `OffsetArray`. Then get an error somewhere totally different, with a bounds error. I...

Absolutely, apologies for not being clearer. Part of the vagueness is how I experience the problems with type piracy on errors is quite abstract. "`zeros` should error" sounds like a...

Great. A wrapper type is how to do this. DimensionalData.jl does similar things without any piracy.

Yes thats exactly how DimensionalData.jl does it. `X/Y/Z` wrappers can be used in base methods like `zero`, `fill`, `rand` to get arrays with named axes and lookup keys. `rand(X('a':'z'), Y(5))`...