Question on mg_queue_del()

[gvanem]

I want to use the mg_queue_* functions in reading from a COM-port.

Init phase like:

  #define RX_SIZE 1000 // or what-ever
  char *sio_buf  = calloc (RX_SIZE, 1);
  char *rx_q_buf = calloc (RX_SIZE, 1);
  struct mg_queue rx_q;

  mg_queue_init (&rx_q, rx_q_buf, RX_SIZE);

And use mg_queue_book() to add the whole COM receive buffer in one go.
Like:

  int len = COM_read (handle, sio_buf, RX_SIZE);
  if (len > 0 && mg_queue_book(&rx_q, &rx_q_buf, len) >= len)
  {
    memcpy (rx_q_buf, sio_buf, len);
    mg_queue_add (&rx_q, len);
  }
  //... handle empty or error

But this COM protocol is some SLIP-like protocol that needs
to de-ESCape some characters. So I'd like to read from the struct mg_queue rx_q
one byte at the time (in a state-machine loop):

  int ch;
  size_t len = mg_queue_next (&rx_q, (char**)&ch);
  if (len >= 1)
  {
    mg_queue_del (&rx_q, 1);
    return (ch);
  }
  return (0);  // empty

But the comment in src/queue.h says:

  // Removes the oldest message of len bytes. Must be called with the same len
  // returned by the preceding mg_queue_next() call.
  void mg_queue_del(struct mg_queue *q, size_t len);

it is not allowed to remove just 1 byte. I need mg_queue_del (&rx_q, len);.

So how can I do this the best way? Perhaps I need to create a
custom mg_queue_pop() for single characters?

BTW, The https://mongoose.ws/documentation/ does not mention
Must be called with the same len ...

And the documentation on mg_queue_init() etc. is not available in the search box?!

[scaprile]

Hi, Gisle,
The old code for the PPP driver in 7.21 and before, used to do something similar as what you want to do, as it de-stuffed HDLC in-place, IIRC. New code for the L2 handler does it my way, at the UART end

[gvanem]

The old code for the PPP driver in 7.21 and before, used to do something similar

Thanks, but how far back do I have to search for this?

[scaprile]

7.21 release tag
When we release, we update version.h and tag. Then HEAD keeps the latest version number as it grows with new stuff, that is why you see 7.21 in HEAD now. Yeah, confusing, but that's the way it's been for its whole history, afaik.

[gvanem]

I've dropped the idea to use mg_queue* for this. More elegant with a rx_head / rx_tail logic.
To get one byte:

  if (rx_head == rx_tail) return -1;
    int rc = rx_head [rx_tail];
    if (++rx_tail > RX_SIZE)
       rx_tail = 0;
  return (rc);

etc. But thanks anyway.

[gvanem]

BTW, I found the un-ESCaping / un-stuffing is now in mg_atcmd_rx(). That gives me some ideas for my implementation.

[scaprile]

Yeah, that's what I intended to say, it is less academic and layered but saves a lot to just unstuff at the entry point, buffer sizes become predictable and coherent again. In deeply-embedded resource-constrained devices of yore, we couldn't waste 2x buffer size for possible worst case stuffing... so we did it at the UART rx point, sometimes within the interrupt handler...