Better solution than lex/yacc for parsing a DSL in C?

Question

One of my programs accepts a commands (like kill foo) at runtime. Think of it as a little domain-specific language. Here are a few examples:

kill
kill client
exit

But also, chained commands are allowed and whitespace is not significant before and after commands, so the following examples are also valid:

kill ; say "that was fun"
  kill  ;  kill      ; kill;

I have currently implemented this with lex/yacc (flex/bison to be specific) and that caused a lot of headache. The lexer very much depends on the context (for example whitespace tokens are generally not returned, unless after a kill keyword for example) and has many different states. The grammar used to have conflicts and I don’t really like the format in which it has to be specified (especially the $1, $2, $3, … to use arguments for non-terminals). Also, the error messages which bison provides (at parse-time) are sometimes accurate, but often not (the kill command with optional arguments leads to error messages like Unexpected $undefined, expected $end or ; for kill clont instead of kill client). Lastly, the C API for yacc is cruel (external defines all over the place).

I am not asking you to solve all the aforementioned questions (I will open separate threads with more specific descriptions and code if there is no way around lex/yacc). Instead, I am interested in alternatives to lex/yacc.

My criteria are the following:

• Input is a string (const char *), there is no output but instead some code should be called for each different keyword.
• I want to use this with C (C99).
• The software should be already included in the major linux distros or at least easy to bundle / package.
• It should be well-documented.
• The syntax for describing my language should be easy.
• It should output meaningful error messages upon parsing errors.
• Performance is not that important (of course it should be fast, but the typical use case is interactive usage, not processing tons of MB of commands).

Answer 1

As for a very simple and small grammar, I'd consider writing the lexer/parser by hand - it's often not that much work.

Virtually all linux distros ship a variant of lex/yacc. Other than that, two other widely used parser generators are lemon and antlr.