IssueTracker

Pop quiz! What value does the following Reactor code produce:

1:  Array<Int, 2> a;
2:
3:  a[0] = 1;
4:  a[1] = 2;
5:
6:  auto x = a[0];
7:  a[0] = a[1];
8:  a[1] = x;
9:  
10: Return(a[0] + a[1]);

The answer is 4.

While this code may appear to implement swapping two array elements, auto x evaluates to Reference<Int> x. It merely holds the address of a[0] (i.e. it's like writing int &x in C++). So when we assign it to a[1] at line 8, it copies the value 2 which was previously assigned to a[0] at line 7.

This can be made to behave more like C code by having Reference<T> also store the dereferenced value on creation. This value can then be returned on implicit casts to T, instead of doing the load at that time. One caveat might be that when memory is writeable but not readable, an expression like a[0] = x could produce a read violation. This could be addressed by eliminating the load when a store to the reference has happened instead.

By deriving Referece<T> from T we could also support things like a[0].x where we're indexing into an array of vectors and and accessing a component of a vector.

It might also be worth looking into whether we can discern between left-hand-side and right-hand-side usage by using rvalue references in constructors and/or operators.