Symbolic sqrt(2) and sin(pi)

[ChrisRackauckas]

@antoine-levitt

Wouldn’t help since 2pi is a float64 (which is pretty annoying when writing generic code). We’d need a “taking irrationals seriously” thing where n*pi would also be an irrational, but it would probably lead to madness pretty soon.

Yeah, the bigger thing to do would probably be to handle it in the switch to SymbolicTracing.jl.

@DNF

If I may ask: What is the purpose of wrapping the inputs in [ ] ? Why not just Term(sqrt, 2) , which seems to give the same results, but is dramatically faster?

I’m actually surprised that works. It should be very simpler to the Expr object which has an op and arguments. I guess the analysis code is now generic enough that it just iterates through the latter argument, and numbers are iterable so it works? @shashi

@oscarbenjamin

What I expected to happen is that expr1 and expr3 would be the same but instead expr3 does not seem to be a symbolic object. You describe it as exact but the printed representation looks more like a high-precision approximation. In an exact representation I would expect to be able to query the object in some way to discover that it represents sqrt(2) symbolically but I don’t see a way to do that with expr3 .

We should probably handle that in SymbolicTracing.jl as mentioned above.