char* tensor_to_string function doesn't look right

[pl-csci]

I think you should remove the conditional at the top of the tensor_to_string function in tensor1d.c. Tell me if I'm wrong.
Here's the current code:

char* tensor_to_string(Tensor* t) {
    // if we already have a string representation, return it
    if (t->repr != NULL) { return t->repr; }
    // otherwise create a new string representation
    ... 
    }

However,

Tensor* tensor_setitem(Tensor* T, int ix, float val) {...}

is an in-place operation that doesn't update the string representation (t->repr) of the tensor.
Or did I miss the line somewhere else?
Adding a dirty flag at the tensor level that flips to true when t->repr would help, but not fully. Because if we have two tensors, t1 and t2, where t1->storage == t2->storage, and if we then perform an in-place operation on t1, t2 wouldn't get the memo, so to speak. We'd need a dirty flag at the storage level as well.

In a micro-implementation like this, it's probably better to nix the conditional in the tensor_to_string function, allow it to regenerate the string on the fly, since we only need it for text output, which is an infrequent task in the context of tensor ops.