Ok, an interesting stuff, indeed.
- LED is on PB0 on my board, it flashes after reset, all programmers work fine
- I did some editing of the avrdudeconf (added resetdelay = 15, similar to uno, and added ..tool=avrdude into boards) - that causes the avrdude resets the BT pin32 (aka dtr) now before the upload (I am using a BT module).
So the upload started, but finished with errors already published here. I did about 70 uploads of a 15kB sketch and my current findings are as follows (chip date 1027, breadboard, good decoupled, short wires (5cm long Rx, Tx to BT), no crossing with other signals or crystal):
- without pull up/down resistor on RX and w/ short wires the probability of a successful upload ~10%
- with pull-up/down 10k res on Rx - about 40%
- with serial 10k in RX - 100% upload, even with a 20cm long Rx wire and while the wire touched/hold during the upload (the 20cm wire at the 1284p's Rx pin side)
- w/o the serial 10k and with 20cm long Rx wire no luck.
That behavior might be caused, except "Rx pin signal/noise sensitivity", by a reversal of the 1284p's Rx input pin direction, even for a very short time, thus it fires against the other output (ftdi's or BT's Tx), and when the current peak is not limited (ie. by a serial resistor) it flips internally into an undefined state. Hopefully the bootloader does not change the Rx direction (or switches internal pullups on/off on that pin) somewhere in the code..