I'm not stringing you along, honest

I'm not stringing you along, honest

  • Comments 20

JScript and VBScript are often used to build large strings full of formatted text, particularly in ASP. Unfortunately, naïve string concatenations are a major source of performance problems.

 

Before I go on, I want to note that it may seem like I am contradicting my earlier post, by advocating some "tips and tricks" to make string concatenation faster.  Do not blindly apply these techniques to your programs in the belief that they will magically make your programs faster!  You always need to first determine what is fast enough, next determine what is not fast enough, and THEN try to fix it!

 

JScript and VBScript use the aptly-named naïve concatenation algorithm when building strings. Consider this silly JScript example:

 

var str = "";

for (var count = 0 ; count < 100 ; ++count)

  str = "1234567890" + str;

 

The result string has one thousand characters so you would expect that this would copy one thousand characters into str. Unfortunately, JScript has no way of knowing ahead of time how big that string is going to get and naïvely assumes that every concatenation is the last one.

 

On the first loop str is zero characters long. The concatenation produces a new ten-character string and copies the ten-character string into it. So far ten characters have been copied. On the second time through the loop we produce a new twenty-character string and copy two ten-character strings into it. So far 10 + 20 = 30 characters have been copied.

 

You see where this is going. On the third time thirty more characters are copied for a total of sixty, on the fourth forty more for a total of one hundred. Already the string is only forty characters long and we have copied more than twice that many characters. By the time we get up to the hundredth iteration over fifty thousand characters have been copied to make that one thousand character string. Also there have been ninety-nine temporary strings allocated and immediately thrown away.

 

Moving strings around in memory is actually pretty darn fast, though 50000 characters is rather a lot.  Worse, allocating and releasing memory is not cheap.  OLE Automation has a string caching mechanism and the NT heap is pretty good about these sorts of allocations, but still, large numbers of allocations and frees are going to tax the performance of the memory manager.

 

If you're clever, there are a few ways to make it better.  (However, like I said before, always make sure you're spending time applying cleverness in the right place.)

 

One technique is to ensure that you are always concatenating small strings to small strings, large strings to large strings.  Pop quiz: what's the difference between these two programs?

 

for (var count = 0 ; count < 10000 ; ++count)

  str += "1234567890" + "hello";

 

and

 

For count = 1 To 10000

  str = str & "1234567890" & "hello"

Next

 

?  I once had to debunk a widely distributed web article which claimed that VBScript was orders of magnitude slower than JScript because the comparison that the author used was to compare the above two programs.  Though they produce the same output, the JScript program is MUCH faster.  Why's that?  Because this is not an apples-to-apples comparison.  The VBScript program is equivalent to this JScript program:

 

for (var count = 0 ; count < 10000 ; ++count)

  str = (str + "1234567890") + "hello";

 

whereas the JScript program is equivalent to this JScript program

 

for (var count = 0 ; count < 10000 ; ++count)

  str = str + ("1234567890" + "hello");

 

See, the first program does two concatenations of a small string to a large string in one line, so the entire text of the large string gets moved twice every time through the loop.  The second program concatenates two small strings together first, so the small strings move twice but the large string only moves once per loop.  Hence, the first program runs about twice as slow.  The number of allocations remains unchanged, but the number of bytes copied is much lower in the second.

 

In hindsight, it might have been smart to add a multi-argument string concatenation opcode to our internal script interpreter, but the logic actually gets rather complicated both at parse time and run time.  I still wonder occasionally how much of a perf improvement we could have teased out by adding one.  Fortunately, as you'll see below, we came up with something better for the ASP case.

 

The other way to make this faster is to make the number of allocations smaller, which also has the side effect of not moving the bytes around so much.

 

var str = "1234567890"; 10

str = str + str; 20

var str4 = str + str;  40

str = str4 + str4;  80

str = str + str;  160

var str32 = str + str;  320

str = str32 + str32; 640

str = str + str32 960

str = str + str4; 1000

 

This program produces the same result, but with 8 allocations instead of 100, and only moves 3230 characters instead of 50000+.   However, this is a rather contrived example -- in the real world strings are not usually composed like this!

 

Now, those of you who have written programs in languages like C where strings are not first-class objects know how to solve this problem efficiently.  You build a buffer that is bigger than the string you want to put in, and fill it up.  That way the buffer is only allocated once and the only copies are the copies into the buffer.  If you don't know ahead of time how big the buffer is, then a double-when-full strategy is quite optimal -- pour stuff into the buffer until it's full, and when it fills up, create a new buffer twice as big.  Copy the old buffer into the new buffer and continue.  (Incidentally, this is another example of one of the "no worse than 200% of optimal" strategies that I was discussing earlier -- the amount of used memory is never more than twice the size of the memory needed, and the number of unnecessarily copied bytes is never more than twice the size of the final buffer.)

 

Another strategy that you C programmers probably have used for concatenating many small strings is to allocate each string a little big, and use the extra space to stash a pointer to the next string.  That way concatenating two strings together is as simple as sticking a pointer in a buffer.  When you're done all the concatenations, you can figure out the size of the big buffer you need, and do all the allocations and copies at once.  This is very efficient, wasting very little space (for the pointers) in common scenarios.

 

Can you do these sorts of things in script?  Actually, yes.  Since JScript has automatically expanding arrays you can implement a quick and dirty string builder by pushing strings onto an array, and when you're done, joining the array into one big string.  In VBScript it's not so easy because arrays are fixed-size, but you can still be pretty clever with fixed size arrays that are redimensioned with a "double when full" strategy.  But surely there is a better way than these cheap tricks. 

 

Well, in ASP there is.  You know, I used to see code like this all the time:

 

str = "<blah>"

str = str + blah

str = str + blahblah

str = str + whatever

 

' etc, etc, etc -- the string gets longer and longer, we have some loops, etc.

 

str = str + "</blah>"
Response.Write str

 

Oh, the pain.  The Response object is an efficient string buffer written in C++.  Don't build up big strings, just dump 'em out into the HTML stream directly.  Let the ASP implementers worry about making it efficient.

 

"Hold on just a minute. Mister Smartypants Lippert," I hear you say, "Didn't you just tell us last week that eliminating calls to COM objects is usually a better win than micro-optimizing small stuff like string allocations?"

 

Yes, I did.  But in this case, that advice doesn't apply because I know something you don't know. 

 

We realized that all the work that the ASP implementers did to ensure that the string buffer was efficient was being overshadowed by the inefficient late-bound call to Response.Write.  So we special-cased VBScript so that it detects when it is compiling code that contains a call to Response.Write and there is a named item in the global namespace called Response that implements IResponse::Write.  We generate an efficient early-bound call for this situation only.  This massively increased the throughput performance of many pages -- unsurprisingly, many ASP pages are gated on string manipulation costs.  We also worked with the NT heap team to push some small-string optimizations into the NT heap.  Together those optimizations led to a 7x throughput improvement in some pages.

 

We also tried to do the trick I mentioned above of keeping a linked list (actually in our implementation it was a tree) of "to be concatenated" strings.  If you ever get your hands on the VBScript sources you'll see that there are a whole lot of conditionally compiled codepaths marked FANCY_STRINGS.  We did get fancy strings working, but it turned out that the overhead of maintaining the extra information, plus the number of times that we need to collapse the strings back down to regular strings (eg, every time one is passed by variable reference, which happens a lot), actually made them slower than just copying regular strings in most common cases.  That was a bummer after spending all summer on implementing it, but you live and learn.

 

What about JScript?  Why no early-bound Response.Write trick in JScript?  Because eval is evil.  In VBScript, if we see

 

Sub Foo()

      Dim Response

      Set Response = MyObj

      Response.Write "Hello"

 

then we don't generate the early bound call, because obviously we would be calling the wrong object!  In VBScript we can detect all these bogus situations lexically and turn off the optimization.  But in JScript...

 

var myfunc = eval;

function foo()

{

      myfunc("var Response = myObj;");

      Response.Write("Hello");

 

We have absolutely no way of detecting at compile time what that thing is going to actually call.  It is nigh impossible to do type inferencing in JScript Classic.  Thus, we couldn't do this trick in JScript because we might break an existing program. (JScript .NET has a type inferencer, which I will discuss in a later blog -- it was quite tricky to implement.)

 

Those of you really on the ball will now be wondering how we can do this in VBScript, since we added both GetRef on functions and an Eval method at the same time as this perf improvement.  Look, famous person!  Over there!

 

VBScript is "broken" in this regard.  If you alter the namespace dynamically to change the type of the Response object, we still call the original Response.Write method.  Don't expect that "bug" to ever be fixed, as it's been in the product since 1998 and we put it there deliberately.  We could do so and still sleep at night because there was no way that we could be breaking an existing VBScript program by adding this perf improvement. 

 

And finally, in JScript .NET the CLR provides an object specially optimized to handle constructing large strings without copying memory around unnecessarily.

 

var sb : System.Text.StringBuilder = new System.Text.StringBuilder();

for (var count = 0 ; count < 100 ; ++count)

sb.Append("1234567890");

 

A string builder uses a double-when-full strategy internally to ensure good performance when building up large strings, so please use it if you're doing lots of string work in a .NET language.

 

  • Raymond, Is there a reason why the Mid statement: Mid$(strVariable, 4, 1) = "a" is not supported in VBScript? It is often used in VB 6 to speed up string concatenations (its not the fastest option available (direct memory copies are faster, for instance), but it is pretty good). Seeya
  • Is is actually true that the double-when-full strategy uses twice the memory needed? The extra reserved space is never written into, so the operating system never has to touch the pages, hence never have to reserve any physical memory for it, right?
  • Matthew: First of all, you're reading Eric Lippert's blog, not Raymond Chen's. The bizzare lvalue syntax of the mid function was never added to VBScript because we never got around to it, and no one really banged down my door asking for it. I think you're the first person to mention it to me in about five years. It would have been a fair amount of work to get it working, as the syntax and semantics are so weird, and we never thought it was worth it. Soeren: The OS might never have to commit the pages to physical memory, but it certainly has to reserve them in the process heap! So what if it never commits some to physical memory? The heap is still filled up, and possibly the extra unused reserved space will cause another allocation to be fragmented badly or fall across more pages than it otherwise might.
  • I just checked -- malloc calls HeapAlloc, which does commit physical memory.
  • Eric, Whoops - sorry for calling you Raymond. ;-) Looking at: <code> For count = 1 To 10000 str = str & "1234567890" & "hello" Next </code> is there any reason why it doesn't change it to: str = str & "1234567890hello" (since they are both constants)? I assume it is not worth the effort in most cases? I am trying to think of a scenario in which doing that wouldn't work (maybe if str was Null - no, that would still work), but I can't think of any... Seeya
  • An anecdote: in a previous lifetime I was involved in the development of a web-based application that implemented its UI in DHTML and JavaScript (the server backend was JSP BTW). One of the team members wrote the code for an HTML-based tree control. This control was generated on the client-side using data downloaded from the server in the form of strings. The problem was, this tree could contain hundreds of nodes and it was taking extremely long to build. Looking at the code I discovered he was constructing a huge string representing the entire tree and then using innerHTML to render it. The tree nodes were added to this tree using string concatenation. I suggested one of the following solutions: 1. Put all the nodes in an array and then use join. 2. Add nodes to the HTML DOM independently using insertAdjacentHTML The second method has the advantage that the tree size (as text) can exceed 64K. But it also translates to lots of DOM/COM calls. I don't remember which option he chose, but the result was a huge performance improvement (something like two orders of magnitude).
  • Matthew: Constant folding at compile time is possible -- JScript .NET does so, but JScript and VBScript do not. I'll write a blog entry about the details. Dan: I was with you right up to the 64K part -- is there some 64K limit somewhere that I don't know about?
  • As I recall VBScript and JScript strings are implemented using COM BSTRs. Or more precisely, VBScript strings are BSTRs and JScript strings are objects convertible to BSTRs. In any event, when passed as arguments to COM methods (such as the browser DOM methods) they are passed in as BSTRs. A BSTR is a data structure that contains both the content of the string, as UNICODE chars, and the length of the string, as a word. Since the length is stored in a word, it is limited to 65535 bytes, 64K. As a result, if the entire HTML block is contained in a single string, this is a limitation on its size.
  • Actually it's a DWORD -- up to 2 billion chars (4 billion bytes / 2 bytes per char) of UNICODE text in a BSTR http://blogs.gotdotnet.com/ericli/commentview.aspx/853ae05f-7610-4531-ab1b-070695e61168
  • > Or more precisely, VBScript strings are BSTRs and JScript strings are objects convertible to BSTRs. No, other way around. JScript strings are implemented as BSTRs which may optionally be wrapped in an object. > Since the length is stored in a word, it is limited to 65535 bytes, 64K. As a result, if the entire HTML block is contained in a single string, this is a limitation on its size. Good heavens no! That limitation hasn't been in place since the last build of the Windows 3.1 version of VBScript. No 32-bit build of scripting ever had a limitation on the number of characters in a string. The win3.1 limitation was due to the 16 bit pointer size, not due to the BSTR length field -- I've just checked the sources, and in fact the 16 bit build of OLE Automation used a DWORD for the length. Boy, do I feel like an old-timer now.
  • Live and learn. Must have looked at an old doc and the detail stuck. Oh, and since I've been programming proffesionally since 89', I bet I'm older than you.
  • Well, I was getting paid to program in 1989, so I guess that makes me a professional -- but then again, I was also in eleventh grade... :)
  • Just to attest to the speedups possible in legacy JScript, I wrote a hex dumper in jscript using string concatination - the dump was built line-by-line and lines were collected into a single string. Modifying the code to use an array, then Join("\n")'ing it at the end provided an about 75x performance improvement for 64k dumps. Larger dumps took much longer and show a much greater improvement. Also interesting: while evaluating string performance, I noticed that "s += a + b + c" is slightly faster than "s.concat(a,b,c)".
  • > Larger dumps took much longer and show a much greater improvement. The naive algorithm gets four times slower every time the number of concats doubles, but the join algorithm gets only two times slower when you double the number of entries. Therefore, yes, you're absolutely right -- the larger the problem, the better the improvement.
  • "Well, I was getting paid to program in 1989, so I guess that makes me a professional -- but then again, I was also in eleventh grade... :)" Make a blogentry about how you started at MS and so on!
Page 1 of 2 (20 items) 12