Question

I’ve got a large set of data for which computing the sort key is fairly expensive. What I’d like to do is use the DSU pattern where I take the rows and compute a sort key. An example:

        Qty   Name      Supplier   
Row 1:   50   Widgets   IBM
Row 2:   48   Thingies  Dell
Row 3:   99   Googaws   IBM

To sort by Quantity and Supplier I could have the sort keys: 0050 IBM, 0048 Dell, 0099 IBM. The numbers are right-aligned and the text is left-aligned, everything is padded as needed.

If I need to sort by the Quanty in descending order I can just subtract the value from a constant (say, 10000) to build the sort keys: 9950 IBM, 9952 Dell, 9901 IBM.

How do I quickly/cheaply build a descending key for the alphabetic fields in C#?

[My data is all 8-bit ASCII w/ISO 8859 extension characters.]

Note: In Perl, this could be done by bit-complementing the strings:

 $subkey = $string ^ ( "\xFF" x length $string );

Porting this solution straight into C# doesn’t work:

 subkey = encoding.GetString(encoding.GetBytes(stringval).
                      Select(x => (byte)(x ^ 0xff)).ToArray());

I suspect because of the differences in the way that strings are handled in C#/Perl. Maybe Perl is sorting in ASCII order and C# is trying to be smart?

Here’s a sample piece of code that tries to accomplish this:

        System.Text.ASCIIEncoding encoding = new System.Text.ASCIIEncoding();
        List<List<string>> sample = new List<List<string>>() {
            new List<string>() { "", "apple", "table" },
            new List<string>() { "", "apple", "chair" },
            new List<string>() { "", "apple", "davenport" },
            new List<string>() { "", "orange", "sofa" },
            new List<string>() { "", "peach", "bed" },
        };
        foreach(List<string> line in sample)
        {
            StringBuilder sb = new StringBuilder();

            string key1 = line[1].PadRight(10, ' ');
            string key2 = line[2].PadRight(10, ' ');

            // Comment the next line to sort desc, desc
            key2 = encoding.GetString(encoding.GetBytes(key2).
                  Select(x => (byte)(x ^ 0xff)).ToArray());

            sb.Append(key2);
            sb.Append(key1);
            line[0] = sb.ToString();
        }

        List<List<string>> output = sample.OrderBy(p => p[0]).ToList();

        return;

Answer 1

You can get to where you want, although I’ll admit I don’t know whether there’s a better overall way.

The problem you have with the straight translation of the Perl method is that .NET simply will not allow you to be so laissez-faire with encoding. However, if as you say your data is all printable ASCII (ie consists of characters with Unicode codepoints in the range 32..127) – note that there is no such thing as ‘8-bit ASCII’ – then you can do this:

            key2 = encoding.GetString(encoding.GetBytes(key2).
                Select(x => (byte)(32+95-(x-32))).ToArray());

In this expression I have been explicit about what I’m doing:

• Take x (which I assume to be in 32..127)
• Map the range to 0..95 to make it zero-based
• Reverse by subtracting from 95
• Add 32 to map back to the printable range

It’s not very nice but it does work.