4 releases
0.1.3 | Jun 18, 2020 |
---|---|
0.1.2 | Apr 20, 2020 |
0.1.1 | Apr 19, 2020 |
0.1.0 | Apr 19, 2020 |
#22 in #sugar
165 downloads per month
35KB
296 lines
The num_traits
API, but in prefix notation
What is this
If you're trying to do scientific computing in Rust, and you can't get used
to mathematical functions like sin()
or cos()
being postfix methods, this
crate may be for you!
It provides free function versions of the trait methods of the num
traits,
so that you can easily do things like sin(x) + 3*ln(y)
.
Each trait's methods are exposed as an module of free functions, named after a snake_case version of the trait's name, and it only takes a couple of use clauses to go from there to using the above syntax in your math expressions.
API coverage
This crate generally opts for maximal coverage of the num traits, except in the following circumstances:
- The trait represents an operator whose standard notation is closer to a
postfix method than to a prefix function, as is the case for most binary
operators. For this reason,
AsPrimitive
,CheckedAdd
,CheckedDiv
,CheckedMul
,CheckedRem
,CheckedShl
,CheckedShr
,CheckedSub
,MulAdd
,MulAddAssign
,Saturating
,WrappingAdd
,WrappingMul
,WrappingShl
,WrappingShr
andWrappingSub
are not covered. - The num_trait crate already provides a set of free functions that cover
90% of a trait's functionality, and we re-export them. Thus,
One
,Signed
andZero
are not covered. - A specific trait or its methods would require very significant supporting
infrastructure to be exposed as a free function by this crate, and the
extent of its real-world usage does not seem to justify the effort.
To be more specific...
FloatConst::TAU()
would require adding Self trait bound support to the underlying macro infrastructure, whileTAU
is arguably a math expert joke that most normal persons would spell out as2.0 * PI
.MulAdd
andMulAddAssign
would require adding generic trait support to the underlying macro infrastructure, while it is debatable whether a multiply-add should be considered a prefix or postfix operator.NumCast
would require adding generic trait method support, when it is dubious whetherfrom::<T, _>(n)
is actually a readability improvement over theT::from(n)
that it replaces.i128
-based casts would require extending this crate's conditional compilation setup quite a bit through use of theautocfg
crate, which seems to be a bit much considering how obscure that type is.
If you find a num trait functionality which is neither exposed by this crate nor covered by the above list, this is likely an oversight from my part, please ping me about it.
I am also willing to reconsider any point of the above policy if someone manages to make a good argument against it. Issues are welcome!
Limitations
Documentation
Only a one-line summary of each method's documentation is provided. Please
refer to the corresponding trait method's documentation in num_traits
for
the full details of each function's API contract.
Namespace collisions
One advantage of using a trait-based approach like num_traits
instead of
free functions like this crate is that trait methods gracefully handle
namespace collisions.
With this crate, you will instead be the one responsible for only use
-ing
one function with a given name at a time.
For what it's worth, this is why programming languages with prefix numerical methods usually also support method overloading. But Rust could not have that language feature, as it would break the kind of advanced type inference that all Rustaceans are used to enjoy today...
License
This crate is distributed under the terms of the MPLv2 license. See the LICENSE file for details.
More relaxed licensing (Apache, MIT, BSD...) may also be negociated, in exchange of a financial contribution. Contact me for details at knights_of_ni AT gmx DOTCOM.
Dependencies
~210KB