lidavidm commented on a change in pull request #11080:
URL: https://github.com/apache/arrow/pull/11080#discussion_r710466286



##########
File path: cpp/src/arrow/compute/kernels/scalar_if_else.cc
##########
@@ -1988,6 +1988,65 @@ Status ExecBinaryCoalesce(KernelContext* ctx, Datum 
left, Datum right, int64_t l
   return Status::OK();
 }
 
+template <typename AppendScalar>
+static Status ExecVarWidthCoalesceImpl(KernelContext* ctx, const ExecBatch& 
batch,
+                                       Datum* out,
+                                       std::function<Status(ArrayBuilder*)> 
reserve_data,
+                                       AppendScalar append_scalar) {
+  // Special case: grab any leading non-null scalar or array arguments
+  for (const auto& datum : batch.values) {
+    if (datum.is_scalar()) {
+      if (!datum.scalar()->is_valid) continue;
+      ARROW_ASSIGN_OR_RAISE(
+          *out, MakeArrayFromScalar(*datum.scalar(), batch.length, 
ctx->memory_pool()));
+      return Status::OK();
+    } else if (datum.is_array() && !datum.array()->MayHaveNulls()) {
+      *out = datum;
+      return Status::OK();
+    }
+    break;
+  }
+  ArrayData* output = out->mutable_array();
+  std::unique_ptr<ArrayBuilder> raw_builder;
+  RETURN_NOT_OK(MakeBuilder(ctx->memory_pool(), out->type(), &raw_builder));
+  RETURN_NOT_OK(raw_builder->Reserve(batch.length));
+  RETURN_NOT_OK(reserve_data(raw_builder.get()));
+
+  for (int64_t i = 0; i < batch.length; i++) {
+    bool set = false;
+    for (const auto& datum : batch.values) {
+      if (datum.is_scalar()) {
+        if (datum.scalar()->is_valid) {
+          RETURN_NOT_OK(append_scalar(raw_builder.get(), *datum.scalar()));
+          set = true;
+          break;
+        }
+      } else {
+        const ArrayData& source = *datum.array();
+        if (!source.MayHaveNulls() ||
+            BitUtil::GetBit(source.buffers[0]->data(), source.offset + i)) {
+          RETURN_NOT_OK(raw_builder->AppendArraySlice(source, i, 
/*length=*/1));

Review comment:
       Right, we'd have the overhead of a virtual call and some other 
bookkeeping vs just a direct append call. 




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to