code_gen julia bug

[jarlebring]

Bug

julia> gen_code("/tmp/bla.jl",graph,lang=LangJulia(false));
julia> include("/tmp/bla.jl")
julia> A=[3 4.0; 5.5 0.1];
julia> dummy(A)
ERROR: UndefRefError: access to undefined reference
Stacktrace:
 [1] getindex
   @ ./array.jl:835 [inlined]
 [2] dummy(A::Matrix{Float64})
   @ Main /tmp/bla.jl:22
 [3] top-level scope
   @ REPL[10]:1

It has something to do with precomputed nodes.

[eringh]

It seems as if the copy!(memslots[1],A) is accessing the undef that is in memslots[1].
I suggest we rewrite it to memslots[1]=copy(A), i.e., change

https://github.com/jarlebring/GraphMatFun.jl/blob/6da3279287f7c9452241171ef8758b39db23f2f7/src/code_gen/gen_julia_code.jl#L138

to

push_code!(code,"$Ak_slot_name=copy($n)")

[jarlebring]

👍

[eringh]

It also think these lines needs to be looked over
https://github.com/jarlebring/GraphMatFun.jl/blob/6da3279287f7c9452241171ef8758b39db23f2f7/src/code_gen/gen_julia_code.jl#L142-L144

Even if we do not write over the input, the pre-computed nodes are still in the memslot and accessed from there, right?
So that should not be overwritten in
https://github.com/jarlebring/GraphMatFun.jl/blob/6da3279287f7c9452241171ef8758b39db23f2f7/src/code_gen/gen_julia_code.jl#L159-L162

Do I understand the code generation correctly?
Does this affect the number of memslots that we need to allocate? It doesn't right? The only difference is whether the original inputs are still in memory or overwritten on the fly, right?

[jarlebring]

Better solution: Move the copy! to after the for-loop.