How to write a (shell) lexer by hand

Question

I'm working on a shell, a small bash-like shell, without scripting (if while ...) I have to make the lexer/parser (LL) by hand.

So the lexer will transform the command (char *cmd) to a linked list (t_list *list). And the LL parser will transform the linked list (t_list *list) to an AST (binary tree t_btree *root) with a grammar

So, I know how to make the LL parser but I don't know how to tokenize my command.

For example: ps | grep ls >> file ; make && ./a.out

=> 'ps' '|' 'grep' 'ls' '>>' 'file' ';' ''make '&&' './a.out'

Thanks.

(I don't wanna use any generator)

Answer 1

(This explains the idea hinted by Spudd86).

You need to implement a finite state machine. There are the following states:

• General state
• Inside a file name
• Inside the && token
• Inside the || token

For each state and next input character, you have to decide what is the next state, and whether to output a token. For example:

• Current state: General; character: x => next state: inside-file-name
• Current state: inside-file-name; character: space => next state: General; output the token
• Current state: inside-file-name; character: & => next state: inside-&&; output the token
• Current state: inside-&&; character: & => next state: General; output the token
• Current state: inside-&&; character: x => next state: General; syntax error
• ... (ad nauseum)

It's much boring work to work out all the rules (the fun starts when you must debug the resulting code), so most people use code generators to do that.

---

Edit: some code (sorry if the syntax is messed-up; i usually program in C++)

enum state {
    STATE_GENERAL,
    STATE_IN_FILENAME,
    ...
};

// Many characters are treated the same (e.g. 'x' and 'y') - so use categories
enum character_category
{
    CHAR_GENERAL, // can appear in filenames
    CHAR_WHITESPACE = ' ',
    CHAR_AMPERSAND = '&',
    CHAR_PIPE = '|',
    CHAR_EOF = EOF,
    ...
};

character_category translate(int c)
{
    switch (c) {
    case '&': return CHAR_AMPERSAND;
    case ' ': case '\t': case '\n': return CHAR_WHITESPACE;
    ...
    default: return CHAR_GENERAL;
    }
}

void do_stuff()
{
    character_category cat;
    state current_state = STATE_GENERAL;
    state next_state;
    char token[100];
    char token_length = 0;
    do {
        int c = getchar();
        cat = translate(c);
        // The following implements a switch on 2 variables
        int selector = 1000 * current_state + cat;

        switch (selector)
        {
        case 1000 * STATE_GENERAL + CHAR_GENERAL:
            next_state = STATE_IN_FILENAME;
            token[token_length++] = c; // append a character to a filename token
            break;

        case 1000 * STATE_GENERAL + CHAR_WHITESPACE:
            next_state = STATE_GENERAL; // do nothing
            break;

        case 1000 * STATE_GENERAL + CHAR_PIPE:
            next_state = STATE_IN_OR_TOKEN; // the first char in '||' or just '|'
            break;

        // Much repetitive code already; define a macro for the case constants?
        // Have to cover all states and all character categories; good luck...

        case 1000 * STATE_IN_FILENAME + EOF:
        case 1000 * STATE_IN_FILENAME + CHAR_WHITESPACE:
            next_state = STATE_GENERAL;
            printf("Filename token: %s\n", token);
            break;

        default:
            printf("Bug\n"); // forgot one of the cases?
        }

        current_state = next_state;

    } while (cat != CHAR_EOF);
}