当前位置：首页 > news >正文

对expat库XML_Parse函数调用优化的测试

news 2025/9/10 7:50:17

xpat库文档中说

XML_Parse
enum XML_Status XMLCALL
XML_Parse(XML_Parser p,const char *s,int len,int isFinal);
enum XML_Status {XML_STATUS_ERROR = 0,XML_STATUS_OK = 1
};
Parse some more of the document. The string s is a buffer containing part (or perhaps all) of the document. The number of bytes of s that are part of the document is indicated by len. This means that s doesn't have to be null-terminated. It also means that if len is larger than the number of bytes in the block of memory that s points at, then a memory fault is likely. Negative values for len are rejected since Expat 2.2.1. The isFinal parameter informs the parser that this is the last piece of the document. Frequently, the last piece is empty (i.e. len is zero.)If a parse error occurred, it returns XML_STATUS_ERROR. Otherwise it returns XML_STATUS_OK value. Note that regardless of the return value, there is no guarantee that all provided input has been parsed; only after the concluding call will all handler callbacks and parsing errors have happened.Simplified, XML_Parse can be considered a convenience wrapper that is pairing calls to XML_GetBuffer and XML_ParseBuffer (when Expat is built with macro XML_CONTEXT_BYTES defined to a positive value, which is both common and default). XML_Parse is then functionally equivalent to calling XML_GetBuffer, memcpy, and XML_ParseBuffer.To avoid double copying of the input, direct use of functions XML_GetBuffer and XML_ParseBuffer is advised for most production use, e.g. if you're using read or similar functionality to fill your buffers, fill directly into the buffer from XML_GetBuffer, then parse with XML_ParseBuffer.

最后两段说，这个函数其实是XML_GetBuffer和XML_ParseBuffer两个函数的包装，再在中间插入从用户buffer到parser buffer的复制，如果read函数直接用parser buffer当缓冲区，就可以省略memcpy的操作。

我用先前的xml文件转csv程序做了个测试，
原代码expatfile.c调用XML_Parse

    char buffer[8192];int done;do {size_t len = fread(buffer, 1, sizeof(buffer), file);done = (len < sizeof(buffer));if (XML_Parse(parser, buffer, len, done) == XML_STATUS_ERROR) {break;}} while (!done);

修改后expatfile2.c调用XML_GetBuffer和XML_ParseBuffer

    char buffer[8192];int done;do {void *buff = XML_GetBuffer(parser, 8192);size_t len = fread(buff, 1, 8192, file);done = (len < 8192);if (XML_ParseBuffer(parser, len, done) == XML_STATUS_ERROR) {break;}} while (!done);

编译运行

gcc expatfile.c -o expatfile -lexpat -O3time ./expatfile /par/lineitem/xl/worksheets/sheet1.xml A1:P1000000
CSV已保存到 /par/lineitem/xl/worksheets/sheet1.csvreal	0m18.882s
user	0m18.168s
sys	0m0.324sgcc expatfile2.c -o expatfile2 -lexpat -O3time ./expatfile2 /par/lineitem/xl/worksheets/sheet1.xml A1:P1000000
CSV已保存到 /par/lineitem/xl/worksheets/sheet1.csvreal	0m18.909s
user	0m18.116s
sys	0m0.284s

测试证明，两种调用几乎没有差别，也许现在memcpy很快，体现不出来影响了。