Anthony Pesch

@PetteriAimonen what are the restrictions around modifying the callback interface (e.g. to return +n / 0 / -1) instead of a bool?

AFAIK, a ssize_t can always hold a size_t, right? The read() system call returns an ssize_t and takes an size_t for count. So as long as we use both, there...

Totally understood re: stable API. The fundamental problem here is that for each read, there are 3 possible results - 1. success 2. eof 3. error With the bool, we...

Awesome, I'll get the tests passing next. RE: function pointers, I guess it's worth noting we don't _really_ need pb_decode / pb_release to internally be able to call these routines....

@PetteriAimonen wanted to ask one question as I'm closing in on _most_ of the tests passing. Right now I just started running all tests using the "fast" path by running...

FWIW, as a small update, this is mostly done, I've just dragging my feet finishing up support for extensions. Trying to get that done in the next few days.

@PetteriAimonen extensions are working, but ran into one issue - is the maximum extension tag encoded somewhere i'm missing? e.g. in alltypes.proto I see `extensions 200 to 255;` but then...

Just pushed the latest work, which is ready enough to get a solid review I think. With this, all tests pass by just running `scons CCFLAGS="-DPB_ENABLE_DECODE_FAST"`. The interface from pb_decode.c...

Did a little bit of work on optimizing some of the generated code and I think a fairly common pattern is emerging for some of the more complicated lvalue types....

> I'm a bit worried that in trying to heavily optimize for speed, there will be significant code size penalties even when not using this feature. Nanopb primarily tries to...