This sample runs a benchmark, illustrating some unexpected results. The benchmark explores 3 approaches for parsing a text file. Each approach is run twice, once parsing a single row of 252 bytes from the file, and then again parsing 50 rows of 252 bytes.
Parses the file in a traditional fashion by using a StreamReader, and Substring for each field.
Parses the file using Span<byte> and then Slice and encoding.GetString(ReadOnlySpan<byte>) for each field.
Parses the file by first converting the byte array to a string, and then referencing it using a Span<char> and then Slice for each field.
From what I understand from Span, I would expect the approaches using Span to exhibit a significantly lower amount of allocated memory and possibly a considerable lower mean execution time.
Although when parsing 1 row from the text file, the results confirm the expectation (less allocated memory and lower mean exeuction time), the second run parsing 50 rows from the same text file shows a far smaller difference, making it almost negligible.
| Method | RowCount | Mean | Gen 0 | Gen 1 | Gen 2 | Allocated |
|---|---|---|---|---|---|---|
| WithSpan | 1 | 638.2 ns | 0.2298 | 0.0010 | - | 1.41 KB |
| WithSpan_StringFirst | 1 | 521.7 ns | 0.3147 | 0.0010 | - | 1.94 KB |
| WithoutSpan | 1 | 1,260.6 ns | 0.9270 | 0.0095 | - | 5.7 KB |
| WithSpan | 50 | 31,935.7 ns | 11.4746 | - | - | 70.7 KB |
| WithSpan_StringFirst | 50 | 24,503.3 ns | 15.5029 | 1.1902 | - | 95.34 KB |
| WithoutSpan | 50 | 26,990.0 ns | 16.3574 | 0.2747 | - | 100.64 KB |
Set build configuration to Release and run the solution. Let the benchmark run and observe the results in the console.