Use zero in istriu, istril methods

[ajkeller34]

This PR replaces 0 with zero(T) for appropriate T in methods of istriu and istril. I'm not sure what tests to add if any since the behavior is unchanged for subtypes of Number found in Base, and there are already some tests for these methods. I'm happy to add some if there are suggestions. I was unable to detect any regressions from these changes using BaseBenchmarks and BenchmarkTools.

For context, I noticed this when playing around with matrices of Unitful numbers, where comparisons with 0 are problematic.

[andreasnoack]

This is the right direction. We have wrongly been relying on 0 for some time. I have a few comments in the code.

[andreasnoack]

I'd make this one

all(t -> t == zero(t), M.ev)

because then you use the instance instead of the type to create the zero. Not all types support zero, e.g. Matrix.

[andreasnoack]

Use A[i,j] != zeros(A[i,j]) for the same reason as before and since the element type could potentially change over the array. This would also fail if the array was a Vector{Any}.

[Jutho]

probably a typo but I guess zero(A[i,j]) instead of zeros.

[andreasnoack]

Indeed

[tkelman]

See the issues discussed in #9325. They all still apply, so I don't think this is so simple.

[Jutho]

After reading #9325, +1 for an iszero function. I have wanted this even before reading that issue, because

• I think iszero(x) reads nicer than x == 0 or x == zero(x)
• I am always doubting which of the latter two to use
• it is perfectly complementary with isinf(and to a lesser extent isnan) and fits nicely with the rest of the is... functions; if there are iseven and isodd functions, than surely iszero is as fundamental

[ajkeller34]

@tkelman Thanks for the info, I wasn't aware of some of the subtleties. I'll close this for now.