Hi Piranha,
I have written a small test application with 2 own boards,
... but based on STM standard example STM32Cube_FW_G4_V1.4.0\Projects\NUCLEO-G474RE\Examples_LL\I2C\I2C_TwoBoards_MasterTx_SlaveRx_DMA_Init
... just I stripped HAL - I use my own "Abstration Layer" (which is not as "abstract" as HAL :) - I acknowledge the HAL is for sure nice and useful for STM to produce its very important multitude of demo software for any controller, but I must admit that I do not like it very much (I love the STM32 register initialization scheme and description, it is very clear somehow...)...).
... but if you use this example with 2 Nucleo 474 boards (if you have them, somehow difficult to get them in the last year...), you should see it if you would enlarge the transfer buffer aLedOn to a larger value, e. g. use my length of 1031, fill it with zeroes, and put the bytes 1, 3, 7, 0xF at the positions 250, 500, 750, 1000, then you should see it easily and fast with a basic oscilloscope... . Just of course you have to adapt the code a bit to poll the TCR bit fast to do the reload correctly (or use an interrupt programming .. I used while-loop-polling of TCR for simplicity in my first basic example .. in a final larger system of course you would use TCR interrupt, this I am just programing... ).
(as I do not like HAL very much, I would not like to try such adaption in HAL... but if you do it and it schould realy work in your program with RELOAD=0 for the last buffer, I would be stunned... then I will try it also, I have two Nucleo474 boards here, but I bet 100% against it...).
