views:

586

answers:

9

I have a string that may have whitespace characters around it and I want to check to see whether it is essentially empty.

There are quite a few ways to do this:

1  if (myString.Trim().Length == 0)
2  if (myString.Trim() == "")
3  if (myString.Trim().Equals(""))
4  if (myString.Trim() == String.Empty)
5  if (myString.Trim().Equals(String.Empty))

I'm aware that this would usually be a clear case of premature optimization, but I'm curious and there's a chance that this will be done enough to have a performance impact.

So which of these is the most efficient method?

Are there any better methods I haven't thought of?


Edit: Notes for visitors to this question:

  1. There have been some amazingly detailed investigations into this question - particularly from Andy and Jon Skeet.

  2. If you've stumbled across the question while searching for something, it's well worth your while reading at least Andy's and Jon's posts in their entirety.

It seems that there are a few very efficient methods and the most efficient depends on the contents of the strings I need to deal with.

If I can't predict the strings (which I can't in my case), Jon's IsEmptyOrWhiteSpace methods seem to be faster generally.

Thanks all for your input. I'm going to select Andy's answer as the "correct" one simply because he deserves the reputation boost for the effort he put in and Jon has like eleventy-billion reputation already.

+4  A: 

I really don't know which is faster; although my gut feeling says number one. But here's another method:

if (String.IsNullOrEmpty(myString.Trim()))
Cloud
Didn't think of that one, and it's probably safer too...
Damovisa
except if it is null it will throw a NullPointerException
Chad Grant
That's true, however none of the other methods pass the "myString = null"-situation. So in relation to this question, it's no biggy.
Cloud
I think it's confusing, becasue myString must not be null to begin with.
Danko Durbić
It might be confusing, I can agree on that. But he is asking for the fastest way. And apparently, this suggestion is one of the faster :)
Cloud
+2  A: 

Checking the length of a string for being zero is the most efficient way to test for an empty string, so I would say number 1:

if (myString.Trim().Length == 0)

The only way to optimize this further might be to avoid trimming by using a compiled regular expression (Edit: this is actually much slower than using Trim().Length).

Edit: The suggestion to use Length came from a FxCop guideline. I've also just tested it: it's 2-3 times faster than comparing to an empty string. However both approaches are still extremely fast (we're talking nanoseconds) - so it hardly matters which one you use. Trimming is so much more of a bottleneck it's hundreds of times slower than the actual comparison at the end.

Joe Albahari
Do you have any arguments to back that statement up?
Cloud
Can you explain why? Don't get me wrong, I believe you, but I'd be interested.
Damovisa
Strings are stored with a header that indicates string length, therefore a string.Length check simply returns that value.
Richard Ev
+12  A: 

(EDIT: See bottom of post for benchmarks on different micro-optimizations of the method)

Don't trim it - that might create a new string which you don't actually need. Instead, look through the string for any characters that aren't whitespace (for whatever definition you want). For example:

public static bool IsEmptyOrWhitespace(string text)
{
    // Avoid creating iterator for trivial case
    if (text.Length == 0)
    {
        return true;
    }
    foreach (char c in text)
    {
        // Could use Char.IsWhiteSpace(c) instead
        if (c==' ' || c=='\t' || c=='\r' || c=='\n')
        {
            continue;
        }
        return false;
    }
    return true;
}

You might also consider what you want the method to do if text is null.

Possible further micro-optimizations to experiment with:

  • Is foreach faster or slower than using a for loop like the one below? Note that with the for loop you can remove the "if (text.Length==0)" test at the start.

    for (int i = 0; i < text.Length; i++)
    {
        char c = text[i];
        // ...
    
  • Same as above, but hoisting the Length call. Note that this isn't good for normal arrays, but might be useful for strings. I haven't tested it.

    int length = text.Length;
    for (int i = 0; i < length; i++)
    {
        char c = text[i];
    
  • In the body of the loop, is there any difference (in speed) between what we've got and:

    if (c != ' ' && c != '\t' && c != '\r' && c != '\n')
    {
        return false;
    }
    
  • Would a switch/case be faster?

    switch (c)
    {
        case ' ': case '\r': case '\n': case '\t':
            return false;               
    }
    

Update on Trim behaviour

I've just been looking into how Trim can be as efficient as this. It seems that Trim will only create a new string if it needs to. If it can return this or "" it will:

using System;

class Test
{
    static void Main()
    {
        CheckTrim(string.Copy(""));
        CheckTrim("  ");
        CheckTrim(" x ");
        CheckTrim("xx");
    }

    static void CheckTrim(string text)
    {
        string trimmed = text.Trim();
        Console.WriteLine ("Text: '{0}'", text);
        Console.WriteLine ("Trimmed ref == text? {0}",
                          object.ReferenceEquals(text, trimmed));
        Console.WriteLine ("Trimmed ref == \"\"? {0}",
                          object.ReferenceEquals("", trimmed));
        Console.WriteLine();
    }
}

This means it's really important that any benchmarks in this question should use a mixture of data:

  • Empty string
  • Whitespace
  • Whitespace surrounding text
  • Text without whitespace

Of course, the "real world" balance between these four is impossible to predict...

Benchmarks I've run some benchmarks of the original suggestions vs mine, and mine appears to win in everything I throw at it, which surprises me given the results in other answers. However, I've also benchmarked the difference between foreach, for using text.Length, for using text.Length once and then reversing the iteration order, and for with a hoisted length.

Basically the for loop is very slightly faster, but hoisting the length check makes it slower than foreach. Reversing the for loop direction is very slightly slower than foreach too. I strongly suspect that the JIT is doing interesting things here, in terms of removing duplicate bounds checks etc.

Code: (see my benchmarking blog entry for the framework this is written against)

using System;
using BenchmarkHelper;

public class TrimStrings
{
    static void Main()
    {
        Test("");
        Test(" ");
        Test(" x ");
        Test("x");
        Test(new string('x', 1000));
        Test(" " + new string('x', 1000) + " ");
        Test(new string(' ', 1000));
    }

    static void Test(string text)
    {
        bool expectedResult = text.Trim().Length == 0;
        string title = string.Format("Length={0}, result={1}", text.Length, 
                                     expectedResult);

        var results = TestSuite.Create(title, text, expectedResult)
/*            .Add(x => x.Trim().Length == 0, "Trim().Length == 0")
            .Add(x => x.Trim() == "", "Trim() == \"\"")
            .Add(x => x.Trim().Equals(""), "Trim().Equals(\"\")")
            .Add(x => x.Trim() == string.Empty, "Trim() == string.Empty")
            .Add(x => x.Trim().Equals(string.Empty), "Trim().Equals(string.Empty)")
*/
            .Add(OriginalIsEmptyOrWhitespace)
            .Add(IsEmptyOrWhitespaceForLoop)
            .Add(IsEmptyOrWhitespaceForLoopReversed)
            .Add(IsEmptyOrWhitespaceForLoopHoistedLength)
            .RunTests()                          
            .ScaleByBest(ScalingMode.VaryDuration);

        results.Display(ResultColumns.NameAndDuration | ResultColumns.Score,
                        results.FindBest());
    }

    public static bool OriginalIsEmptyOrWhitespace(string text)
    {
        if (text.Length == 0)
        {
            return true;
        }
        foreach (char c in text)
        {
            if (c==' ' || c=='\t' || c=='\r' || c=='\n')
            {
                continue;
            }
            return false;
        }
        return true;
    }

    public static bool IsEmptyOrWhitespaceForLoop(string text)
    {
        for (int i=0; i < text.Length; i++)
        {
            char c = text[i];
            if (c==' ' || c=='\t' || c=='\r' || c=='\n')
            {
                continue;
            }
            return false;
        }
        return true;
    }

    public static bool IsEmptyOrWhitespaceForLoopReversed(string text)
    {
        for (int i=text.Length-1; i >= 0; i--)
        {
            char c = text[i];
            if (c==' ' || c=='\t' || c=='\r' || c=='\n')
            {
                continue;
            }
            return false;
        }
        return true;
    }

    public static bool IsEmptyOrWhitespaceForLoopHoistedLength(string text)
    {
        int length = text.Length;
        for (int i=0; i < length; i++)
        {
            char c = text[i];
            if (c==' ' || c=='\t' || c=='\r' || c=='\n')
            {
                continue;
            }
            return false;
        }
        return true;
    }
}

Results:

============ Length=0, result=True ============
OriginalIsEmptyOrWhitespace             30.012 1.00
IsEmptyOrWhitespaceForLoop              30.802 1.03
IsEmptyOrWhitespaceForLoopReversed      32.944 1.10
IsEmptyOrWhitespaceForLoopHoistedLength 35.113 1.17

============ Length=1, result=True ============
OriginalIsEmptyOrWhitespace             31.150 1.04
IsEmptyOrWhitespaceForLoop              30.051 1.00
IsEmptyOrWhitespaceForLoopReversed      31.602 1.05
IsEmptyOrWhitespaceForLoopHoistedLength 33.383 1.11

============ Length=3, result=False ============
OriginalIsEmptyOrWhitespace             30.221 1.00
IsEmptyOrWhitespaceForLoop              30.131 1.00
IsEmptyOrWhitespaceForLoopReversed      34.502 1.15
IsEmptyOrWhitespaceForLoopHoistedLength 35.690 1.18

============ Length=1, result=False ============
OriginalIsEmptyOrWhitespace             31.626 1.05
IsEmptyOrWhitespaceForLoop              30.005 1.00
IsEmptyOrWhitespaceForLoopReversed      32.383 1.08
IsEmptyOrWhitespaceForLoopHoistedLength 33.666 1.12

============ Length=1000, result=False ============
OriginalIsEmptyOrWhitespace             30.177 1.00
IsEmptyOrWhitespaceForLoop              33.207 1.10
IsEmptyOrWhitespaceForLoopReversed      30.867 1.02
IsEmptyOrWhitespaceForLoopHoistedLength 31.837 1.06

============ Length=1002, result=False ============
OriginalIsEmptyOrWhitespace             30.217 1.01
IsEmptyOrWhitespaceForLoop              30.026 1.00
IsEmptyOrWhitespaceForLoopReversed      34.162 1.14
IsEmptyOrWhitespaceForLoopHoistedLength 34.860 1.16

============ Length=1000, result=True ============
OriginalIsEmptyOrWhitespace             30.303 1.01
IsEmptyOrWhitespaceForLoop              30.018 1.00
IsEmptyOrWhitespaceForLoopReversed      35.475 1.18
IsEmptyOrWhitespaceForLoopHoistedLength 40.927 1.36
Jon Skeet
Will the foreach(char c in text) actually do an inline search or will it create a new array of chars?
Damovisa
It will create an appropriate IEnumerator<char> associated with the string. It doesn't copy the string.
Jon Skeet
I've just tested this method with a million iterations - it's slightly slower than using Trim(). I guess you might save a bit down the line, though, through avoiding GC.
Joe Albahari
P.S. It might also depend on how long the (untrimmed) string was.
Joe Albahari
@Jon Thanks for that - very good to know. I've been blindly using .ToCharArray()
Damovisa
@albahari (are you one of Joe or Ben, btw?) That's completely bizarre. I can't see how it *could* be slower than Trim, unless it's the iterator killing it (see alternative options in edit). What is your test data like, out of interest?
Jon Skeet
Looping backwards through the string may also improve matters. I haven't tested this though.
Pondidum
Andy: Why? The whitespace could be at the start or the end. Why would it be better to go from the end?
Jon Skeet
I've updated the answer with more info about the behaviour of Trim, which is quite interesting and important...
Jon Skeet
@Jon: Trim uses substring internally to extract the copy of the "trimmed" string. Substring does not make a copy when asked to duplicate a string, it returns the source string instead.That said, I can't see a reason that "doing it yourself" could be slower that Trim - as long as you replace the foreach with a plain for.
Dan C.
RnR
Also - suggesting the usage of text[i] as an optimization is strange - to get to the i'th element of a table you have to multiply i by the size of the elements and then add that to the begining of your table to get the address - even if the multiplication will be a left shift (the size is a power of 2) it's always slower then using an iterator approach and just incrementing the current address that foreach surely gives you.
RnR
@RnR: he just needs a for loop instead of the foreach, as suggested in the later edit.
Dan C.
@RnR: Using foreach involves creating an iterator. Why do that if we know the result already? This optimizes for the "empty" case of course, at the slight cost of the non-empty case. I don't know how much that affects things.
Jon Skeet
@RnR: Do you know what the string iterator and string indexer look like internally? Are you absolutely sure they differ significantly? :)
Jon Skeet
@DanC: Have you tested a for loop to ensure it's definitely faster? I haven't yet, although I will if I get the time...
Jon Skeet
Comparing with a compiled Regex might be interesting too. :)
leppie
@Jon: I haven't, it's common knowledge :) (just being ironic, I know one should test before making assumptions). I strongly expect the for to be faster; the foreach involves creating an enumerator object and the MoveNext call is just an indexer access - so in the end, it's just a complicated way of doing char c = myString[i];
Dan C.
sad that you are missing an unsafe version :(
Chad Grant
@DanC: The difference is that the iterator version *could* potentially avoid doing the bounds check in both the loop and the indexer. There's enough room for doubt that I wouldn't like to say without testing.
Jon Skeet
@Deviant: True. I don't tend to think in terms of unsafe code. That may well squeeze out a bit more speed.
Jon Skeet
@Jon: it *could*, but it doesn't. MoveNext checks the string length on every call before getting the character from the string, just as a for would on every loop. But as you say, it should be tested.
Dan C.
I would love to be able to use all the tools i have at home, but i am somewhat restricted at work - hence my thrown-together-testing.
Pondidum
+17  A: 

Edit: New tests:

Test orders:
x. Test name
Ticks: xxxxx //Empty String
Ticks: xxxxx //two space
Ticks: xxxxx //single letter
Ticks: xxxxx //single letter with space
Ticks: xxxxx //long string
Ticks: xxxxx //long string  with space

1. if (myString.Trim().Length == 0)
ticks: 4121800
ticks: 7523992
ticks: 17655496
ticks: 29312608
ticks: 17302880
ticks: 38160224

2.  if (myString.Trim() == "")
ticks: 4862312
ticks: 8436560
ticks: 21833776
ticks: 32822200
ticks: 21655224
ticks: 42358016


3.  if (myString.Trim().Equals(""))
ticks: 5358744
ticks: 9336728
ticks: 18807512
ticks: 30340392
ticks: 18598608
ticks: 39978008


4.  if (myString.Trim() == String.Empty)
ticks: 4848368
ticks: 8306312
ticks: 21552736
ticks: 32081168
ticks: 21486048
ticks: 41667608


5.  if (myString.Trim().Equals(String.Empty))
ticks: 5372720
ticks: 9263696
ticks: 18677728
ticks: 29634320
ticks: 18551904
ticks: 40183768


6.  if (IsEmptyOrWhitespace(myString))  //See John Skeet's Post for algorithm
ticks: 6597776
ticks: 9988304
ticks: 7855664
ticks: 7826296
ticks: 7885200
ticks: 7872776

7. is (string.IsNullOrEmpty(myString.Trim())  //Cloud's suggestion
ticks: 4302232
ticks: 10200344
ticks: 18425416
ticks: 29490544
ticks: 17800136
ticks: 38161368

And the code used:

public void Main()
{

    string res = string.Empty;

    for (int j = 0; j <= 5; j++) {

        string myString = "";

        switch (j) {

            case 0:
                myString = "";
                break;
            case 1:
                myString = "  ";
                break;
            case 2:
                myString = "x";
                break;
            case 3:
                myString = "x ";
                break;
            case 4:
                myString = "this is a long string for testing triming empty things.";
                break;
            case 5:
                myString = "this is a long string for testing triming empty things. ";

                break;
        }

        bool result = false;
        Stopwatch sw = new Stopwatch();

        sw.Start();
        for (int i = 0; i <= 100000; i++) {


            result = myString.Trim().Length == 0;
        }
        sw.Stop();


        res += "ticks: " + sw.ElapsedTicks + Environment.NewLine;
    }


    Console.ReadKey();  //break point here to get the results
}
Pondidum
Could you benchmark my answer too please?
Jon Skeet
+1 for actually testing the difference
Cloud
Could you benchmark my answer (IsNullOrEmpty) too? Then we have all the methods in one bench situation and on one system.
Cloud
Concur - IsNullOrEmpty() would be good too :)
Damovisa
Try other test cases too - including "" and " x ". Otherwise you're only testing one path.
Jon Skeet
For your benchmark to be fair, shouldn't you also measure the time taken if the test is false? The fastest way in the average case might depend on how often you expect the test to succeed.
Simon Nickerson
Oh, and "x" (i.e. no trimming required). And long strings :)
Jon Skeet
ok, give me a few minutes...
Pondidum
Add a long string with only whitespace too.
leppie
Wow, comprehensive answer...
Damovisa
+2  A: 

myString.Trim().Length == 0 Took : 421 ms

myString.Trim() == '' took : 468 ms

if (myString.Trim().Equals("")) Took : 515 ms

if (myString.Trim() == String.Empty) Took : 484 ms

if (myString.Trim().Equals(String.Empty)) Took : 500 ms

if (string.IsNullOrEmpty(myString.Trim())) Took : 437 ms

In my tests, it looks like myString.Trim().Length == 0 and surprisingly, string.IsNullOrEmpty(myString.Trim()) were consistently the fastest. The results above are a typical result from doing 10,000,000 comparisons.

womp
@womp: string.IsNullOrEmpty checks the string Length too (besides doing the null check); in this situation, I would very much prefer checking it directly - option 1.
Dan C.
Yeah, I was surprised by how performant it turned out to be.
womp
A: 

Since I just started I can't comment so here it is.

if (String.IsNullOrEmpty(myString.Trim()))

if code will fail if myString is null since you can't call methods in a object that is null

So the correct syntax would be something like this.

if (String.IsNullOrEmpty(myString))
{
     myString.Trim;
    //do the rest of you code
}
Oakcool
A: 

All of these methods seem very inefficient, in terms of the lost productivity of developers wasting time on trivia.

Joe
A: 
public static bool IsNullOrEmpty(this String str, bool checkTrimmed)
{
  var b = String.IsNullOrEmpty(str);
  return checkTrimmed ? b && str.Trim().Length == 0 : b;
}
abatishchev
+2  A: 

String.IsNullOrWhitespace in .NET 4 Beta 2 also plays in this space and doesnt need to be custom written

Ruben Bartelink