Looking back on printf

2015.08.02 12:35

printf("Hello world, it's time to reexamine printf()");

From "Hello World" to the latest Linux kernel, printf molds the C programmer's mindset on formatted output. Love it or hate it, printf's influence has spawned numerous ports that unquestioningly copy its syntax and semantics in C libraries and even scripting languages. How printf earned this distinction through various strengths:

Conversion specifications elegantly insert dynamic data between static strings
Fire and forget output buffering
Providing variadic function arguments with a 1 to 1 correspondence of the conversion specifications
Being the very first function a budding C programmer calls

Although C++ iostreams, boost formatting and other alternatives exist no efforts to really rethink printf exist until now: reprint. Essential to understanding the motivation for reprint is an brutal examination of the "strengths" of printf.

Templating and conversion specifications

Templating is perhaps the most straightforward way to produce formatted output. The popularity of PHP and other templating languages is a testament to the power of templating. Every templating system requires two things:

A separator between static output and dynamic output.
A language describing the dynamic input and output specifications.

All escape characters in C strings use '\' character, but printf's separator character is the '%' character. In modern times, the use of '\b' or '\n' is a superior alternative for the following reasons:

For printing percent values, printf would require using a double escaped percent character, which always looks awkward in code.
The use of '%' diverts neophyte C programmers from truly understanding the canonical list of escape characters.

The second point is the most important. Neophyte C programmers should learn that '\' is the universal escaping character in C strings and not be confused with the '%' sign. The original uses of '\b' and '\f' have been long abandoned and should be repurposed for escaping formatted output from static output.

Subsequently, the conversion specifier syntax was built organically with seemingly randomly chosen syntax characters. One only needs to glance at the man page to see printf's resemblance to an occult magic spell book. Educating a C programmer in disciplined and deliberate programming must start with a solid foundation built on precisely specified systems. printf's syntax does not set an example to follow.

Output buffering

The use of printf works great in simple programs, where the output is almost magically shuttled away to stdout. However, this paradigm breaks down when printf becomes repurposed as a way to create formatted strings. The numerous buffer overflows attributable to sprintf are impossible to count. snprintf is almost no better, as it requires the programmer to estimate an upper bound on how much space the string will take up before printing it.

The bugs caused by sprintf and snprintf do not arise from flaws in C, but rather because the fire and forget semantics of printf are incorrectly applied to strings. Essentially, sprintf and snprintf are hacks to work on character arrays from a system that worked on buffering. Unfortunately, C programmers have been indoctrinated to use the functions because they have been portrayed as the "logical" extensions to printf.

Variadic function arguments

printf is a great way to introduce C programmers to variadic arguments. The ability to push a variable number of arguments to a function is essential to concise code. However, where printf falls short is that this is the ONLY way to marshall data to printf. This shortcoming is most pronounced when outputting all the fields of a struct value. It is necessary that the number of conversion specifiers at least matches the number of fields in the struct. However, requiring the programmer to push all the fields of a struct into variadic arguments makes printf seem cumbersome. Additionally, when the fields encode individual bits with significance, the arguments to printf become especially hairy. Not only is this a waste of code, it is a waste of stack space; ideally, the programmer would only have to specify a POINTER to the struct to print it out.

The first function

When learning a new language, the first few concepts or elements encountered shape our understanding of that language. With decades of hindsight, it is necessary to look again at how new C programmers are educated. Personally, I stress new programmers because very few existing C programmers are likely to question their foundations. There is no reason to enumerate all the reasons seasoned C programmers will cling to the old "tried and true" ways. I only have the motiviation to build a new foundation for those willing to learn.