String.comparison performance (with trim)

Question

I need to do alot of high-performance case-insensitive string comparisons and realized that my way of doing it .ToLower().Trim() was really stupid due do all the new strings being allocated

So I digged around a little and this way seems preferable:

String.Compare(txt1,txt2, StringComparison.OrdinalIgnoreCase)

The only problem here is that I want to ignore leading or trailing spaces, ie Trim() but if I use Trim I have the same problem with string allocations. I guess I could check each string and see if it StartsWith(" ") or EndsWith(" ") and only then Trim. Either that or figure out the index,length for each string and pass to string.Compare override

public static int Compare
(
    string strA,
    int indexA,
    string strB,
    int indexB,
    int length,
    StringComparison comparisonType
) 

but that seems rather messy and I probably have to to use some integers if I dont make a really big if-else statement for every combination of trailing and leading blanks on both strings... so any ideas of an elegant solution?

Here's my current proposal:

public bool IsEqual(string a, string b)
    {
        return (string.Compare(a, b, StringComparison.OrdinalIgnoreCase) == 0);
    }

    public bool IsTrimEqual(string a, string b)
    {
        if (Math.Abs(a.Length- b.Length) > 2 ) // if length differs by more than 2, cant be equal
        {
            return  false;
        }
        else if (IsEqual(a,b))
        {
            return true;
        }
        else 
        {
            return (string.Compare(a.Trim(), b.Trim(), StringComparison.OrdinalIgnoreCase) == 0);
        }
    }

Answer 1

I notice that your first suggestion only compares for equality rather than sorting, that allows some further efficiency savings.

public static bool TrimmedOrdinalIgnoreCaseEquals(string x, string y)
{
    //Always check for identity (same reference) first for
    //any comparison (equality or otherwise) that could take some time.
    //Identity always entails equality, and equality always entails
    //equivalence.
    if(ReferenceEquals(x, y))
        return true;
    //We already know they aren't both null as ReferenceEquals(null, null)
    //returns true.
    if(x == null || y == null)
        return false;
    int startX = 0;
    //note we keep this one further than the last char we care about.
    int endX = x.Length;
    int startY = 0;
    //likewise, one further than we care about.
    int endY = y.Length;
    while(startX != endX && char.IsWhiteSpace(x[startX]))
        ++startX;
    while(startY != endY && char.IsWhiteSpace(y[startY]))
        ++startY;
    if(startX == endX)      //Empty when trimmed.
        return startY == endY;
    if(startY == endY)
        return false;
    //lack of bounds checking is safe as we would have returned
    //already in cases where endX and endY can fall below zero.
    while(char.IsWhiteSpace(x[endX - 1]))
        --endX;
    while(char.IsWhiteSpace(y[endY - 1]))
        --endY;
    //From this point on I am assuming you do not care about
    //the complications of case-folding, based on your example
    //referencing the ordinal version of string comparison
    if(endX - startX != endY - startY)
        return false;
    while(startX != endX)
    {
        //trade-off: with some data a case-sensitive
        //comparison first
        //could be more efficient.
        if(
            char.ToLowerInvariant(x[startX++])
            != char.ToLowerInvariant(y[startY++])
        )
            return false;
    }
    return true;
}

Of course, what is an equality checker without a matching hashcode producer:

public static int TrimmedOrdinalIgnoreCaseHashCode(string str)
{
    //Higher CMP_NUM (or get rid of it altogether) gives
    //better hash, at cost of taking longer to compute.
    const int CMP_NUM = 12;
    if(str == null)
        return 0;
    int start = 0;
    int end = str.Length;
    while(start != end && char.IsWhiteSpace(str[start]))
        ++start;
    if(start != end)
        while(char.IsWhiteSpace(str[end - 1]))
            --end;

    int skipOn = (end - start) / CMP_NUM + 1;
    int ret = 757602046; // no harm matching native .NET with empty string.
    while(start < end)
    {
            //prime numbers are our friends.
        ret = unchecked(ret * 251 + (int)(char.ToLowerInvariant(str[start])));
        start += skipOn;
    }
    return ret;
}

Answer 2

Something like this should do it:

public static int TrimCompareIgnoreCase(string a, string b) {
   int indexA = 0;
   int indexB = 0;
   while (indexA < a.Length && Char.IsWhiteSpace(a[indexA])) indexA++;
   while (indexB < b.Length && Char.IsWhiteSpace(b[indexB])) indexB++;
   int lenA = a.Length - indexA;
   int lenB = b.Length - indexB;
   while (lenA > 0 && Char.IsWhiteSpace(a[indexA + lenA - 1])) lenA--;
   while (lenB > 0 && Char.IsWhiteSpace(b[indexB + lenB - 1])) lenB--;
   if (lenA == 0 && lenB == 0) return 0;
   if (lenA == 0) return 1;
   if (lenB == 0) return -1;
   int result = String.Compare(a, indexA, b, indexB, Math.Min(lenA, lenB), true);
   if (result == 0) {
      if (lenA < lenB) result--;
      if (lenA > lenB) result++;
   }
   return result;
}

Example:

string a = "  asdf ";
string b = " ASDF \t   ";

Console.WriteLine(TrimCompareIgnoreCase(a, b));

Output:

0

You should profile it against a simple Trim and Compare with some real data, to see if there really is any difference for what you are going to use it for.