AVR Freaks

I'd steal part of what Jim said:

const uint8_t message[] PROGMEM = 15,16,19,20,24;

then use it like this:

const uint8_t convert[] PROGMEM = 15,16,19,20,24;
const char message[] PROGMEM = “AVRFREAKSISTHEBEST”

uint8_t i;

int main(void) {
    for (i = 0; i < strlen(message); i++) {
        original = pgm_read_byte(&message[i]);
        encoded = pgm_read_byte(&convert[original - 'A']);
        use(encoded);
    }
}

I split that up to explain what's going on with some sensible variable names but obviously you could use:

    for (i = 0; i < strlen(message); i++) {
        use(pgm_read_byte(&convert[pgm_read_byte(&message[i]) - 'A']));
    }

though to be honest the code generated is probably identical in both cases and I think it helps the code to be self documenting by introducing the names "original" and "encoded".

PS before someone moans, yes the original uses characters up to 'T' yet the convert[] arrray currently only has entries for 5 characters ('A'..'E') and there's no checks for reading off the end of the array!

I still don't get it: why do you need the char step at all? Just store (and use) the raw values as suggested by Jim: that is, 0b010101 instead of A.

You could build them using some macros if that helps you. For example:

// Indiviual colors
#define R  01
#define G  10
#define B  11

// 3 colors set
#define col(c1, c2, c3)   col_(c1, c2, c3)
#define col_(c1, c2, c3)  0b##c1##c2##c3

// Usage example
const uint8__t colors[] PROGMEM = { col(R,R,R), col(R,G,B), … };

You could even define all of the 3³ possible values (that's only 27 of them after all), if you like better how the final code looks:

#define RRR  col(R,R,R)
#define RRG  col(R,R,G)
#define RRB  col(R,R,B)
#define RGR  col(R,G,R)
#define RGG  col(R,G,G)
#define RGB  col(R,G,B)
…

// And use them like this:
const uint8__t colors[] PROGMEM = { RRR, RGB, … };

It just concatenates stuff:

col_(A, B, C)              => 0bABC
col_(whatever, you, want)  => 0bwhateveryouwant