@jasper Thanks a lot. The following worked (I had to add the `auto` return type
as you mentioned):
template `+`[L, R](lhs: L, rhs: R): auto =
DualBinaryExpr[AddOp, L, R](l: lhs, r: rhs)
Run
@mratsim Your tensor lib Lux seems quite impressive. My idea is to have an
autodiff library in Nim similar to this one I developed in C++:
[autodiff.github.io](https://forum.nim-lang.org/autodiff.github.io) . It relies
heavily on template metaprogramming, `if constexpr`, SFINAE, etc., to optimize
the expression tree (whenever possible) and avoid temporaries.
There are forward and reverse mode automatic differentiation algorithms in
there. My research needs, however, relies mainly (if not only) in forward mode;
so, forward mode in autodiff lib is what I have spent most of my available time
for this dev.
I'm quite new to Nim; just started exploring it yesterday and I found it quite
neat. Right now, just want to experiment with it and check if it could be used
for advanced scientific computing (for problems that involve numerical
optimization, solving ODEs, PDEs, etc.).
