Fix COPY TO does not produce an output file for the empty set (comments added)

Author: Bert Vermeiren

datafusion/datasource/src/write/demux.rs

@@ -40,9 +40,9 @@ use datafusion_common::cast::{
 };
 use datafusion_common::{exec_datafusion_err, internal_datafusion_err, not_impl_err};
 use datafusion_common_runtime::SpawnedTask;
-use datafusion_execution::TaskContext;
 
 use chrono::NaiveDate;
+use datafusion_execution::TaskContext;
 use futures::StreamExt;
 use object_store::path::Path;
 use rand::distr::SampleString;
@@ -68,6 +68,11 @@ pub type DemuxedStreamReceiver = UnboundedReceiver<(Path, RecordBatchReceiver)>;
 /// be written with the extension from the path. Otherwise the default extension
 /// will be used and the output will be split into multiple files.
 ///
+/// Output file guarantees:
+///  - Partitioned files: Files are created only for non-empty partitions.
+///  - Single-file output: 1 file is always written, even when the stream is empty.
+///  - Multi-file output: Depending on the number of record batches, 0 or more files are written.
+///
 /// Examples of `base_output_path`
 ///  * `tmp/dataset/` -> is a folder since it ends in `/`
 ///  * `tmp/dataset` -> is still a folder since it does not end in `/` but has no valid file extension
@@ -171,10 +176,8 @@ async fn row_count_demuxer(
         max_rows_per_file
     };
 
-    // Single-file output requires creating at least one file stream in advance.
-    // If no record batches are present in the input stream,
-    // the file stream must still be created to produce a valid output file.
     if single_file_output {
+        // ensure we have one file open, even when the input stream is empty
         open_file_streams.push(create_new_file_stream(
             &base_output_path,
             &write_id,