Fix issue #604: Using the .item function on a Complex Tensor crashes (#…

…605)

Also adds a test case for the issue.

Co-authored-by: Angel Ezquerra <[email protected]>

src/arraymancer/laser/tensor/initialization.nim

@@ -285,7 +285,19 @@ func item*[T_IN, T_OUT](t: Tensor[T_IN], _: typedesc[T_OUT]): T_OUT =
   ## This only works for Tensors (of any rank) that contain one single element.
   ## If the tensor has more than one element IndexDefect is raised.
   if likely(t.size == 1):
-    when T_OUT is Complex64:
+    when T_IN is Complex and T_OUT is Complex:
+      # When the input and the output types are Complex, we need to find
+      # the "base" type of the output type (e.g. float32 or float64),
+      # and then convert the real and imaginary parts of the input value
+      # into the output base type before creating the output complex type 
+      type TT = typeof(
+        block:
+          var tmp: T_OUT
+          tmp.re
+      )
+      let val = t.squeeze[0]
+      result = complex(TT(val.re), TT(val.im))
+    elif T_OUT is Complex64:
       result = complex(float64(t.squeeze[0]))
     elif T_OUT is Complex32:
       result = complex(float32(t.squeeze[0]))

tests/tensor/test_shapeshifting.nim

@@ -157,6 +157,10 @@ proc main() =
         let a = [[[[1]]]].toTensor
         let value = a.item(Complex64)
         check value == complex(1.0, 0)
+      block:
+        let a = [[[[complex[float64](1.0, 1.1)]]]].toTensor
+        let value = a.item(Complex32)
+        check value == complex[float32](1.0, 1.1)
 
     test "Unsqueeze":
       block: