Somehow nodenum was getting reset to zero (and saved to flash - which is
bad because it makes the failure permanent). So I've changed nodenum
selection to occur after we load the saved preferences (and we try to keep
nodenum stable in that case).
I'm puzzled as to how it ever got set to zero (unless there *shudder*
is some errant pointer that clobbered it). But next week I'm turning
4 byte nodenums back on, which will make this moot - because they
will always be based on macaddr and the current process where nodes
haggle with the mesh to pick a unique one-byte nodenum will be gone.
Hi, I think the problem you were having building for ESP32 was due to
a funny thing. Notice the #define for INTERRUPT_ATTR. That macro expands
to IRAM_ATTR - which is a special flag the ESP32 requires for _any_ code
that is going to be called from an ISR. So that the code is guaranteed
to be in RAM (the ESP32 uses a clever scheme where the FLASH is actually
high speed serial flash and all reads/writes are actually only happening
to a small number of pages in RAM and they have a driver that is constantly
copying blocks they need into that ram. This essentially how VM works
for desktop computers, but in their case they are paging to FLASH.
But for code that runs in an interrupt handler must _always_ be in RAM
because if you took a 'page fault' for that code being missing in RAM they
can't nicely do their clever VM scheme.
So that's all good. The problem was - apparently GCC for the ESP32 has a
a bug when that attribute is applied in the class declaration. So
I moved it out into the cpp file and all seems well now.
