performance of TiXmlText::Parse with STL

[Dominik]

Hi

I had a a performance problem when using TinyXML to parse an xml file which contained a large (ca. 4MB) CDATA section.
The performance bottleneck was the continuous reallocation of the std::string class when adding another character to the value member variable.
At least in MS VC++ 6 the std::string object grows by a constant amount which is horrible if you read large blocks of data. Therefore I would suggest the following patch.

Index: tinyxmlparser.cpp

--- tinyxmlparser.cpp   (Revision 2750)
+++ tinyxmlparser.cpp   (Arbeitskopie)
@@ -1459,6 +1459,8 @@
                                && !StringEqual( p, endTag, false, encoding )
                          )
                {
+                       if (value.capacity()==value.length())
+                               value.reserve(value.capacity()*2) ;
                        value += *p;
                        ++p;
                }

[Lee Thomason]

Wow - the MS VC6 STL implementation leaves a lot to be desired. Interesting information.

Effectively however, your patch overrides the STL memory allocation scheme, which isn't a good idea either. A safer fix might be:

#if (_MSC_VER == 1300 ) // or some such version #
...your allocation patch...
#endif

but I am trying as much as possible to stay out of the compiler version game. I think it is better, in the main distribution to live with non-critical compiler issues.

It's a great tip for people using that compiler however. Another approach is to switch to STLPort which works fine with VC6 and is an altogether better implementation of STL.

lee