This very minute? It's impossible. The scanner for C expressions is a program with ~1000 lines of source code (of course, it's a rough estimate). Probably, it's possible to write more compact (and fast) scanner with special methods (based on automata theory, for example).
Better start from a simple grammar. For example, consider simplest arithmetical expressions:
Tokens:: one of
Operators:: one of
+ - / *
Punctuators:: one of
Try to implement a scanner for this lexical grammar.
Input: source text string (or char array)
Output: next token code