Wednesday, May 12, 2010

Beware of the \r\n gotcha on Windows

If (as I have just done) you decide to FastFormat to prepare intermediate text for output, and your program runs on Windows, beware of the writeln() and fmtln() API functions. Consider the following code

std::string s;

ff::writeln(s, "hello"); // s => "hello\r\n"

ff::write(std::cout, s); // writes "hello\r\r\n"

This occurs because the underlying C standard library expands each standard end-of-line (EOL) character \n to the Windows-specific EOL sequence \r\n.

Instead, use the *ln() functions on the output, as in:

ff::write(s, "hello"); // s => "hello"

ff::writeln(std::cout, s); // writes "hello\r\n"

Remember that FastFormat "appends" to sinks

A user recently posted a note on the Help forum, suggesting that FastFormat was up to 100 times slower than MFC's CString::Format(). Thankfully that's not the case, and FastFormat appears to be faster than CString::Format() in even the simple case suggested.

The reason this appeared to be so was because the test program reused the same CString instance in each call to fastformat::fmt(), resulting in concatenation to a huge size. The fault here is no doubt mine, for not having made the nature of FastFormat's formatting more clear in the documentation. So I'll try and do so briefly now.

All FastFormat formatting is appending, and this is a deliberate design decision. In part, this is to achieve consistency between immediate sinks such as strings, and stream sinks (such as std::cout, stdout, ...). Also, it's useful to be able to break apart large or complex formatting operations into statements, without sacrificing performance or expressiveness. (Note: there'll always a small performance penalty for such things, but it will be insignificant in most cases where the size and complexity of the statement demands breaking up.) Consider the following example:

std::string s;
ff::fmt(s, "{0};{1};{2}|{3};{4};{5}|{6};{7};{8}|{9};{10};{11}|", a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11);

The same can be achieved as follows:

std::string s;
ff::fmt(s, "{0};{1};{2}|", a0, a1, a2);
ff::fmt(s, "{0};{1};{2}|", a3, a4, a5);
ff::fmt(s, "{0};{1};{2}|", a6, a7, a8);
ff::fmt(s, "{0};{1};{2}|", a9, a10, a11);

In real-world cases, such clarity may be worth paying a few more extra cycles for.