Environment: VC5/6, NT4, CE 2.11
Once upon a day I was asked to write an application which could do some serious filtering, on a ‘;’ seperatedfile which held multiple collumns. The initial file was about 7 Mb in size, so I had to create a program which could read a file, process the read lines and produce a new output file.
So you think, what’s the big deal ?
When I wrote that program, a pentium 133 was considered a fast PC. As far as I knew, there where no generic (Microsoft provided ??) solutions for a simple task as reading text from a file. So I build my own textfile reading class.
This brings me to the part which I found most interesting, the first version of this filtering program, did the job in several seconds.
So how come the thing worked so fast ? By the optimal manner of reading a text file.
The StringFile class itself consists of 2 loops. One loop is for filling a read buffer, and the other loop is used for reading a line from this buffer. The effect of these loops is that when the file is read, its is done by reading 2k of data per turn. Further processing (finding where any given line starts and when it stops) is done inside memory, not on disk. And, as you can probably known, memory is faster then disk…so there’s my explanation for the speed of the filter program.
After some fiddling around with this class I decided to post this, so that everyone can enjoy this piece of code.
This sample show just how easy this thing works, open, read and close.
What do you want more ??
#include "StringFile.h"
BOOL ReadTextFile(LPCSTR szFile)
{
CStringFile sfText;
CString szLine;
BOOL bReturn = FALSE;
// When the given file can be opened
if(sfText.Open(szFile))
{
// Read all the lines (one by one)
while(sfText.GetNextLine(szLine)!=0)
{
printf("%s\r\n",szLine); //And print them
}
sfText.Close(); // Close the opened file
bReturn = TRUE; // And say where done succesfully
}
return bReturn;
}
Some benchmarking (trying to find optimum blocksize for reading) gave me the following results:
This shows that the optimum size for this piece of code lies around 2k blocksize. Increasing this blocksize doesn’t speedup reading, the only thing speeding up the read actions is probably improving the used code.