lammertb/libfins

decoding of float and double is not working

Opened this issue · 1 comments

juwalter commented

helle @lammertb

after some testing we noticed that libfins is seemingly broken for float and double

In libfins at

libfins/src/fins_01_04.c

Lines 185 to 217 in 91e9aa4

case FINS_DATA_TYPE_INT32 :
case FINS_DATA_TYPE_UINT32 :
case FINS_DATA_TYPE_BCD32 :
case FINS_DATA_TYPE_SBCD32_0 :
case FINS_DATA_TYPE_SBCD32_1 :
case FINS_DATA_TYPE_SBCD32_2 :
case FINS_DATA_TYPE_SBCD32_3 :
case FINS_DATA_TYPE_FLOAT :
if ( XX_finslib_decode_address( item[offset+a].address, & address ) ) return FINS_RETVAL_INVALID_READ_ADDRESS;
area_ptr = XX_finslib_search_area( sys, & address, 16, FI_MRD, false );
if ( area_ptr == NULL ) return FINS_RETVAL_INVALID_READ_AREA;
chunk_start = address.main_address;
chunk_start += area_ptr->low_addr >> 8;
chunk_start -= area_ptr->low_id;
fins_cmnd.body[bodylen++] = area_ptr->area;
fins_cmnd.body[bodylen++] = (chunk_start >> 8) & 0xff;
fins_cmnd.body[bodylen++] = (chunk_start ) & 0xff;
fins_cmnd.body[bodylen++] = 0x00;
chunk_start++;
fins_cmnd.body[bodylen++] = area_ptr->area;
fins_cmnd.body[bodylen++] = (chunk_start >> 8) & 0xff;
fins_cmnd.body[bodylen++] = (chunk_start ) & 0xff;
fins_cmnd.body[bodylen++] = 0x00;
recvlen += 6;
break; 

case FINS_DATA_TYPE_INT32    :
case FINS_DATA_TYPE_UINT32   :
case FINS_DATA_TYPE_BCD32    :
case FINS_DATA_TYPE_SBCD32_0 :
case FINS_DATA_TYPE_SBCD32_1 :
case FINS_DATA_TYPE_SBCD32_2 :
case FINS_DATA_TYPE_SBCD32_3 :
case FINS_DATA_TYPE_FLOAT    :

	if ( XX_finslib_decode_address( item[offset+a].address, & address ) ) return FINS_RETVAL_INVALID_READ_ADDRESS;

	area_ptr = XX_finslib_search_area( sys, & address, 16, FI_MRD, false );
	if ( area_ptr == NULL ) return FINS_RETVAL_INVALID_READ_AREA;

	chunk_start  = address.main_address;
	chunk_start += area_ptr->low_addr >> 8;
	chunk_start -= area_ptr->low_id;

	fins_cmnd.body[bodylen++] = area_ptr->area;
	fins_cmnd.body[bodylen++] = (chunk_start >> 8) & 0xff;
	fins_cmnd.body[bodylen++] = (chunk_start     ) & 0xff;
	fins_cmnd.body[bodylen++] = 0x00;

	chunk_start++;

	fins_cmnd.body[bodylen++] = area_ptr->area;
	fins_cmnd.body[bodylen++] = (chunk_start >> 8) & 0xff;
	fins_cmnd.body[bodylen++] = (chunk_start     ) & 0xff;
	fins_cmnd.body[bodylen++] = 0x00;

	recvlen += 6;

	break;

the library treats FINS_DATA_TYPE_INT32, FINS_DATA_TYPE_UINT32, FINS_DATA_TYPE_BCD32, FINS_DATA_TYPE_SBCD32_0, FINS_DATA_TYPE_SBCD32_1, FINS_DATA_TYPE_SBCD32_2, FINS_DATA_TYPE_SBCD32_3, FINS_DATA_TYPE_FLOAT exactly the same when requesting data from the PLC.

However, during decoding:

(this seems correct for e.g. uint32)

case FINS_DATA_TYPE_UINT32 :

	item[offset+a].uint32   = fins_cmnd.body[bodylen+3];
	item[offset+a].uint32 <<= 8;
	item[offset+a].uint32  += fins_cmnd.body[bodylen+4];
	item[offset+a].uint32 <<= 8;
	item[offset+a].uint32  += fins_cmnd.body[bodylen+0];
	item[offset+a].uint32 <<= 8;
	item[offset+a].uint32  += fins_cmnd.body[bodylen+1];

	bodylen += 5;

	break;

but broken here:

libfins/src/fins_01_04.c

Lines 333 to 344 in 91e9aa4

case FINS_DATA_TYPE_FLOAT :
sfloat.val_raw[0] = fins_cmnd.body[0];
sfloat.val_raw[1] = fins_cmnd.body[1];
sfloat.val_raw[2] = fins_cmnd.body[3];
sfloat.val_raw[3] = fins_cmnd.body[4];
item[offset+a].sfloat = sfloat.val_float;
bodylen += 5;
break; 

case FINS_DATA_TYPE_FLOAT :

	sfloat.val_raw[0] = fins_cmnd.body[0];
	sfloat.val_raw[1] = fins_cmnd.body[1];
	sfloat.val_raw[2] = fins_cmnd.body[3];
	sfloat.val_raw[3] = fins_cmnd.body[4];

	item[offset+a].sfloat = sfloat.val_float;

	bodylen += 5;

	break;

I cannot understand why using seemingly absolute offsets fins_cmnd.body[0]? this should also take bodylen into account or perhaps use pointer arithmetic to move fins_cmnd.body?

am I missing something?

lammertb commented

Seems like a bug TBH. The bytes should at least be aligned, not with a gap of one byte in between.