Proposal to revamp FIFO processing

[davies46]

Looking at the FIFO processing (and as far as I can tell), the library keeps reading the device to empty the FIFO, so instead of getting a full gesture it reads in fragments (one FIFO at a time), does some quick computation on the to give two values, and accumulates those values each time a new fragment comes in.

If you look at each reading in the FIFO as X for left/right and Y for up/down, the computation on a fragment determines the entry and exit positions by checking for the first and last XY values, ignoring those that are too close to (0,0).

There's no reason I can see (other than memory footprint) for processing each fragment separately, and it's possibly not ideal to use the first/last valid point in a fragment either. I suspect it's just to save memory and CPU cycles.

If you happen to be running on a chip with some clout, and some RAM, there will be much better solutions. You might for instance string all the fragments together to get a bunch of XY points, run a least-squares fit over the data to get a gradient and look at the ordering to get a direction for your vector.

Anyway, I'm taking a look at that now :-)