Extending Unitful
Making your own units package
New units or dimensions can be defined from the Julia REPL or from within other packages. To avoid duplication of code and effort, it is advised to put new unit definitions into a Julia package that is then published for others to use. For an example of how to do this, examine the code in UnitfulUS.jl
, which defines U.S. customary units. It's actually very easy! Just make sure you read all of the cautionary notes on this page. If you make a units package for Unitful, please submit a pull request so that I can provide a link from Unitful's README!
Some limitations
Precompilation
When creating new units in a precompiled package that need to persist into run-time (usually true), it is important that the following or something very similar make it into your code:
const localunits = Unitful.basefactors const localpromotion = Unitful.promotion # only if you've used @dimension function __init__() merge!(Unitful.basefactors, localunits) merge!(Unitful.promotion, localpromotion) # only if you've used @dimension end
The definition of localunits
(localpromotion
) must happen after all new units (dimensions) have been defined.
The problem is that the @unit
macro needs to add some information to a dictionary defined in Unitful, regardless of where the macro is executed (the use of this dictionary does not lead to run-time penalties, if you were wondering). However, because Unitful is precompiled, changes made to it from another module at compile-time will not persist.
The const localunits = Unitful.basefactors
line makes a copy of the compile-time-modified dictionary, which can be precompiled into the module where this code appears, and then the dictionary is merged into Unitful's dictionary at runtime.
If you'd like, you can also consider adding a call to Unitful.register
in your __init__
function, which will make your units accessible using Unitful's @u_str
macro. Your unit symbols should ideally be distinctive to avoid colliding with symbols defined in other packages or in Unitful. If there is a collision, the @u_str
macro will still work, but it will use the unit found in whichever package was registered most recently, and it will omit a warning every time.
Type uniqueness
Currently, when the @dimension
, @derived_dimension
, @refunit
, or @unit
macros are used, some unique symbols must be provided which are used to differentiate types in dispatch. These are typically the names of dimensions or units (e.g. Length
, Meter
, etc.) One problem that could occur is that if multiple units or dimensions are defined with the same name, then they will be indistinguishable in dispatch and errors will result.
I don't expect a flood of units packages to come out, so probably the likelihood of name collision is pretty small. When defining units yourself, do take care to use unique symbols, perhaps with the aid of Base.gensym()
if creating units at runtime. When making packages, look and see what symbols are used by existing units packages to avoid trouble.
Archaic or fictitious unit systems
In the rare event that you want to define physical units which are not convertible to SI units, you need to do a bit of extra work. To be clear, such a conversion should always exist, in principle. One can imagine, however, archaic or fictitious unit systems for which a precise conversion to SI units is unknown. For example, a cullishigay is one-third of a mudi, but only approximately 1.25 imperial bushels. There may be cases where you don't even have an approximate conversion to imperial bushels. At such a time, you may feel uncomfortable specifying the "base unit" of this hypothetical unit system in terms of an SI quantity, and may want to explicitly forbid any attempt to convert to SI units.
One can achieve this by defining new dimensions with the @dimension
or @derived_dimension
macros. The trick is to define dimensions that display suggestively like physical dimensions, like 𝐋, 𝐓 etc., but are distinct as far as Julia's type system is concerned. Then, you can use @refunit
to base units for these new dimensions without reference to SI. The result will be that attempted conversion between the hypothetical unit system and SI will fail with a DimensionError
, so be sure you provide some hints in how your new dimensions are displayed to avoid confusing users. It would be confusing to throw a DimensionError
when attempting to convert between lengths which are incompatible in the sense of the previous paragraph, when both lengths display their dimension as 𝐋.