Missing Missing methods for angle, hypot and 2-arg atan

[yha]

Should be fixed with

Base.atan(::Union{Real,Missing}, ::Union{Real,Missing}) = missing
Base.hypot(::Union{Number,Missing}...) = missing
Base.angle(::Missing) = missing

I would open a PR if I could get Julia to build on my Windows machine...

[musm]

I would open a PR if I could get Julia to build on my Windows machine...

https://github.com/JuliaLang/julia/blob/master/doc/build/windows.md#cygwin-to-mingw-cross-compiling

the packages can be easily installed

make,python2,libiconv,curl,time,p7zip,ccache,mingw64-x86_64-gcc-g++,mingw64-x86_64-gcc-fortran

[JeffBezanson]

I'm curious --- did you actually encounter a need for these or do you just think they should be defined for general reasons?

[yha]

@musm Trying to build, I get
Generating precompile statements...ERROR: LoadError: IOError: stream is closed or unusable
make cleanall doesn't help.

@JeffBezanson I did encounter a need: I was calculating speed (hypot) and heading (atan) from vectors of coordinates which include missings, and was using angle in the calculation of circular means of angles (circmean(x) = angle(mean(exp.(x.*im))))

[musm]

@yha usually that error is fixed by rerunning make

[nalimilan]

You could try making a PR directly from GitHub. In general I wouldn't recommend that, but in this case it's quite simple (just add the function names to the existing loops in base/missing.jl and test/missing.jl), so it should be doable.

[deszoeke]

All math functions need methods for missing or people will fall back on reliable IEEE NaN practices.
A broadcasting cosd.( ) calls Base.deg2rad, which doesn't handle missing.

Using Revise and adding the method
Base.deg2rad(::Missing) = missing
inside my module results in a seg fault and hang.

It's a doozy of a seg fault. My other shell has lost its user ID, and I can't ssh from a new window anymore. Needless to say, I haven't been able to do intensive testing.

Julia 1.5.3 nalimilan Copr port on CentOS.

[KristofferC]

All math functions need methods for missing or people will fall back on reliable IEEE NaN practices.

It is hard to define "math functions" and people have also claimed that every "string function" should also have an extra method for Missing.

Instead, you can use explicit tools like skipmissing or coalesce. Then you compose missing handling and normal functions instead of having a design where you need to "color" certain functions to have this extra Missing method.

[deszoeke]

In this case, the error is that the basic safe behavior, that an operation (multiplying by pi/180) on a missing returns missing, did not accept the missing type.

Since my data are Floats, I will use a cringeworthy function to preemptively convert all missing to NaN.

[pdeffebach]

OP if you aren't aware of passmissing in Missings.jl that will fix your problem.

Find the highest-level function that accepts a scalar and wrap it in passmissing. Then it will propagate missings for you. No need for NaN.

[deszoeke]

Thanks. I was not aware of ‘passmissing’, but will find it useful.

The type error here was actually from trying to broadcast on an array defined in the lexical scope of the function, rather than an argument passed into the function.