fgoto on binary variable length protocol with ragel - ragel

Question

I tried to write a parser to a simple binary protocol
0 or more out of frame bytes
Start of frame with STX (0x02)
Two messaje length bytes for messaje length.
One command byte .
0 or more data bytes.
One checksun byte.
The ragel script
static int len;
static unsigned char command;
static int indexx;
static unsigned char data[0x10000];
static unsigned char checksum;
%%{
machine themachine;
action out_of_frame_action { }
action start_of_frame_action {}
action first_byte_len_action { len = fc *256; }
action second_byte_len_action { len += fc; }
action command_action { command = fc; indexx=0; len--;}
action data_action {
if(!len)
fgoto check;
else
{
data[indexx++];
len --;
}
}
action checksum_action { checksum = fc; }
stx = 0x02;
first_byte_len = any;
second_byte_len = any;
command = any;
data = any*;
checksum = any;
main := (^stx* $ out_of_frame_action)
(stx > start_of_frame_action)
(first_byte_len > first_byte_len_action)
(second_byte_len > second_byte_len_action)
(command > command_action)
(data $ data_action) ;
check := (checksum % checksum_action) ;
}%%
machine state
But the result machine don't jump to check (5) status so doesn't execute checksum_action and don't return to the first state to continue parsing.
What is wrong?
Follow #Roman recomendations I post a full example. This example doesn't work either because get a wrong checksum.
#include <stdint.h>
#include <stdio.h>
static int len;
static unsigned char command;
static int indexx;
static unsigned char data[0x10000];
static unsigned char checksum;
%%{
machine themachine;
stx = 0x02;
action out_of_frame_action
{
printf("OOF %02X\n",fc);
}
action start_of_frame_action
{
printf("STX ");
}
action first_byte_len_action
{
printf("fbl[%d] ",(int)(fc));
len = 256*((unsigned char)fc);
}
action second_byte_len_action
{
len += (unsigned char )fc ;
printf("sbl[%d] Len=%d ",(int)(fc),len);
indexx=0;
len-=2; // Checksum and command are included on message len
}
action command_action
{
command = fc;
printf("CMM=%02X ", command);
}
action check_len
{
len > 0
}
action data_action
{
data[indexx++]=fc;
printf("[%02X]",(unsigned char) fc);
len--;
}
action checksum_action
{
checksum = fc;
printf(" Chk=%02X \n",checksum);
}
first_byte_len = any ;
second_byte_len = any;
command = any;
data = any*;
checksum = any;
check = (checksum % checksum_action);
main := ((^stx* $ out_of_frame_action)
(stx > start_of_frame_action)
(first_byte_len > first_byte_len_action)
(second_byte_len > second_byte_len_action)
(command > command_action)
(data when check_len $ data_action)
check
)**;
}%%
%% write data;
int main(void)
{
uint8_t buf[] = {
0x00, // OOF 00
0x00, // OOF 00
0x02,0x00,0x03,0x20,0x01,0x21, // STX fbl[0] sbl[3] Len = 3 CMM=20 [01] Chk=21
0x00, // OOF 00
0x00, // OOF 00
0x02,0x00,0x05,0x81,0x01,0x02,0x03,0x87, // STX fbl[0] sbl[5] Len = 5 CMM=81 [01][02][03] Chk=87
0x02,0x00,0x03,0x03,0x01,0x04, // STX fbl[0] sbl[3] Len = 3 CMM=03 [01] Chk=04
0x02,0x00,0x05,0x07,0x01,0x02,0x03,0x0D // STX fbl[0] sbl[5] Len = 5 CMM=07 [01][02][03] Chk=0D
};
int cs;
uint8_t *p, *pe, *eof;
p = buf;
eof = pe = buf + sizeof(buf)/sizeof(uint8_t);
%% write init;
%% write exec;
}
I change
check = (checksum % checksum_action);
by
check = (checksum > checksum_action);
but is not a solution.

Answer 1

It would be really nice if you have provided us with a sample input and a full source to verify the program as a whole
I'm not sure why your goto doesn't work, but I'd probably simplify things by not using fgoto at all. There is no need for it here, you can just use semantic condition (when)
Your machine doesn't return to the initial state because nothing tells it to.
So I've made these little changes to your original code:
--- main.c.orig 2017-12-17 11:48:49.369200291 +0300
+++ main.c 2017-12-20 22:51:11.096283379 +0300
## -12,14 +12,12 ##
action first_byte_len_action { len = fc *256; }
action second_byte_len_action { len += fc; }
action command_action { command = fc; indexx=0; len--;}
+ action check_len {
+ len
+ }
action data_action {
- if(!len)
- fgoto check;
- else
- {
- data[indexx++];
- len --;
- }
+ data[indexx++];
+ len--;
}
action checksum_action { checksum = fc; }
## -30,13 +28,13 ##
data = any*;
checksum = any;
-main := (^stx* $ out_of_frame_action)
+ check = (checksum % checksum_action);
+main := ((^stx* $ out_of_frame_action)
(stx > start_of_frame_action)
(first_byte_len > first_byte_len_action)
(second_byte_len > second_byte_len_action)
(command > command_action)
- (data $ data_action) ;
+ (data when check_len $ data_action) check)**;
-check := (checksum % checksum_action) ;
}%%
Notice how when is used for data simplifying its action and also notice the longest-match kleene star operator ** in the main that instructs it to loop back to the start after successful frame parsing. This gets us to this nice diagram, which is probably closer to what you want:

Answer 2

In your first example, fgoto consumes the checksum data so it's not available later when you want the checksum. You can use fhold to prevent it from being consumed. The checksum_action should also go back to main (and use a $ transition instead of '%', which will only occur if it's EOF.).
action data_action {
if(!len) {
fhold;
fgoto check;
} else {
data[indexx++] = fc;
len --;
}
}
action checksum_action { checksum = fc; fnext main; }
(or you could put the checksum logic in the data_action code).