Dave Reinagel wrote:
>
> I found a few strange things during debug.
> I run my slaves in interrupt mode, and at the end of each write sequence,
> I got an interrupt indicating that a read operation was required,
> which was a bogus event.  To fix that, I made a state machine to detect
> a read interrupt immediately after the end of a write operation, to
> dump the reading, and all my problems went away.

I've had hassles with Microchip's I2C slave hardware.
Such as the R/W flag is only valid after the address byte is received.
After that it becomes the ACK/NACK flag for data bytes.
So I have to copy the R/W flag somewhere safe for later data.

I also had problems with I2C masters that don't send NACK on the last byte.
So I get a surplus byte request interrupt which I have to service by
sending a dummy byte with d7 = high.

I wonder if your bug is a similar effect.

Does your bug happen frequently or rarely?
I'm having problems where I'm reading an EEPROM
which involves writing (the EEPROM internal data pointer)
and I occasionally get an error.

> But another warning...

Am I the only one getting rather cheesed off with all the
irritating weird things about the PIC that you have to find
out the hard way? I wonder how the original designers tested
the PIC before they said "yep, this design works".