make arraymancer compatible with ORC and latest devel

.github/workflows/ci.yml

@@ -14,7 +14,7 @@ jobs:
       max-parallel: 20
       matrix:
         # branch: [version-1-0, version-1-2, version-1-4, version-1-6] # [devel]
-        nim: ['version-1-4', 'version-1-6'] # 'devel'
+        nim: ['version-1-4', 'version-1-6', 'devel'] # 'devel'
         target:
           - os: linux
             cpu: amd64
@@ -216,6 +216,14 @@ jobs:
           # Run the tests.
           nimble test
 
+      - name: Run Arraymancer tests with ORC
+        working-directory: arraymancer
+        run: |
+          export ARRAYMANCER_TEST_LANG=${{ matrix.test_lang }}
+          nimble install -y --depsOnly
+          # Run the tests.
+          nimble test_orc_release
+
       - name: Build docs
         if: >
           github.event_name == 'push' && github.ref == 'refs/heads/master' &&

src/arraymancer/io/io_csv.nim

@@ -84,7 +84,11 @@ proc read_csv*[T: SomeNumber|bool|string](
   elif T is bool:
     parser = parseBool
   elif T is string:
-    parser = proc(x: string): string = shallowCopy(result, x) # no-op
+    parser = proc(x: string): string =
+      when defined(gcArc) or defined(gcOrc):
+        result = x
+      else:
+        shallowCopy(result, x) # no-op
 
   # 1. count number of lines and columns using memfile interface
   let (numRows, numCols) = countLinesAndCols(csvPath, separator, quote, skipHeader)

src/arraymancer/linear_algebra/helpers/decomposition_lapack.nim

@@ -69,7 +69,7 @@ proc syevr*[T: SomeFloat](a: var Tensor[T], uplo: static char, return_eigenvecto
     lwork: int32 = -1  # dimension of a workspace array
     work_size: T
     liwork: int32 = -1 # dimension of a second workspace array
-    iwork: seq[cint]
+    iwork: seq[int32]
     iwork_size: int32
     info: int32
 
@@ -84,7 +84,7 @@ proc syevr*[T: SomeFloat](a: var Tensor[T], uplo: static char, return_eigenvecto
 
   # Setting up output
   var
-    isuppz: seq[cint] # unused
+    isuppz: seq[int32] # unused
     isuppz_ptr: ptr int32
 
   eigenval = newTensorUninit[T](a.shape[0]) # Even if less eigenval are selected Lapack requires this much workspace
@@ -246,7 +246,7 @@ proc gesdd*[T: SomeFloat](a: var Tensor[T], U, S, Vh: var Tensor[T], scratchspac
     work_size: T
     lwork = -1'i32 # size query
     info: int32
-    iwork = newSeqUninit[cint](8 * k)
+    iwork = newSeqUninit[int32](8 * k)
 
   U.newMatrixUninitColMajor(ldu, ucol)
   S = newTensorUninit[T](k.int)

src/arraymancer/linear_algebra/helpers/least_squares_lapack.nim

@@ -77,7 +77,7 @@ proc gelsd*[T: SomeFloat](
   singular_values = newTensorUninit[T](minmn) # will hold the singular values of A
 
   var # Temporary parameter values
-    iwork = newSeqUninit[cint](liwork)
+    iwork = newSeqUninit[int32](liwork)
     info, rank: int32
     lwork: int32 = -1
 

src/arraymancer/tensor/higher_order_foldreduce.nim

@@ -125,7 +125,10 @@ proc reduce*[T](t: Tensor[T],
   ##     a.reduce(max) ## This returns the maximum value in the Tensor.
 
   t.reduce_inline():
-    shallowCopy(x, f(x,y))
+    when defined(gcArc) or defined(gcOrc):
+      x = f(x,y) # hopefully nvro will work
+    else:
+      shallowCopy(x, f(x,y))
 
 proc reduce*[T](t: Tensor[T],
                 f: (Tensor[T], Tensor[T]) -> Tensor[T],
@@ -142,4 +145,7 @@ proc reduce*[T](t: Tensor[T],
   ##     - A tensor aggregate of the function called all elements of the tensor
 
   t.reduce_axis_inline(axis):
-    shallowCopy(x, f(x,y))
+    when defined(gcArc) or defined(gcOrc):
+      x = f(x,y) # hopefully nvro will work
+    else:
+      shallowCopy(x, f(x,y))

tests/tensor/test_init.nim

@@ -183,8 +183,8 @@ proc main() =
         var t: TestObject
         t.x = @[1.0, 2, 3, 4].toTensor()
         t.x = t.x
-        doAssert t.x.storage.isNil, "If you see this message, bug Nim #16185 is fixed." &
-          " Remove this test or set it to `not t.x.isNil`!"
+        when (NimMajor, NimMinor) >= (1, 6): # works since 1.6; see https://github.com/nim-lang/Nim/issues/16185
+          doAssert not t.x.storage.isNil
 
     test "Init tensor from raw buffer":
       let size = 100