Re: perfomance optimisations

[Adrian Robert]

> I think we should not introduce a lot more IMPs in GSWeb than we now 
> have.
> There are other areas that should be addressed.
>
> This is a example of a potential thing that could be faster.
> If somebody uses cString it is likely that he also uses cStringLength 
> on the same string.
> Both does basically the same. Has somebody an idea how to optimize 
> here?

Could you give us an idea of the type/magnitude of performance gains 
you are wishing for / expecting?  Could you post some profiling data 
for example?  I guess I'm a bit surprised that string-handling kinds of 
things are a problem.  Java works very well in the server space and 
I've never been impressed by the low-level efficiency of its 
string-handling (or for that matter the overall architecture of how it 
does things.

Here is, for example, the Java 1.3 implementation of String.getBytes():

    private byte[] getBytes(CharToByteConverter ctb) {
        ctb.reset();
        int estLength = ctb.getMaxBytesPerChar() * count;
        byte[] result = new byte[estLength];
        int length = 0;
        try {
            length += ctb.convertAny(value, offset, (offset + count),
                                     result, 0, estLength);
            length += ctb.flushAny(result, ctb.nextByteIndex(), estLength);
        } catch (CharConversionException e) {
            throw new InternalError("Converter malfunction: " +
                                    ctb.getClass().getName());
        }

        if (length < estLength) {
            // A short format was used:  Trim the byte array.
            byte[] trimResult = new byte[length];
            System.arraycopy(result, 0, trimResult, 0, length);
            return trimResult;
        }
        else {
            return result;
        }
    }

Basically the same as the Base method below, except I'm not sure what 
that 'append ""' is for.  (And this starts out interpreted, then gets 
just-in-time compiled.)

Anyway, overall Java on the server side is not super-fast compared to 
PHP or even Perl, but has won wide acceptance because servlets, JSP, 
and JDBC are simple but effective technologies, and also by being 
designed for scalability so that more hardware can easily be thrown at 
a situation with little developer effort.  This was accomplished by a 
well-designed application model where session information is 
encapsulated and threads are used and reused in lieu of object creation 
as much as possible.


On Apr 5, 2005, at 1:07 PM, David Wetzel wrote:

> - (const char*) cString
> {
>   NSData        *d;
>   NSMutableData *m;
>
>   d = [self dataUsingEncoding: _DefaultStringEncoding
>          allowLossyConversion: NO];
>   if (d == nil)
>     {
>       [NSException raise: NSCharacterConversionException
>                   format: @"unable to convert to cString"];
>     }
>   m = [d mutableCopy];
>   [m appendBytes: "" length: 1];
>   AUTORELEASE(m);
>   return (const char*)[m bytes];
> }
>
> - (unsigned int) cStringLength
> {
>   NSData        *d;
>
>   d = [self dataUsingEncoding: _DefaultStringEncoding
>          allowLossyConversion: NO];
>   return [d length];
> }