Merge micrpython 1.17

[jepler]

I've started on this. Difficulties so far include:

• f-string implementation is not quite compatible. In particular, f"" f"" concatenation is not supported. I think we should accept this minor regression
• we should consider whether to revert our make_new slot prototype to match upstream

In my WIP branch, make VARIANT=coverage test_full now passes, but all the ports are broken.

[jepler]

Upstream uses a make_new signature of

typedef mp_obj_t (*mp_make_new_fun_t)(const mp_obj_type_t *type, size_t n_args,
    size_t n_kw, const mp_obj_t *args);

while we have

typedef mp_obj_t (*mp_make_new_fun_t)(const mp_obj_type_t *type, size_t n_args,
    const mp_obj_t *args, mp_map_t *kw_args);

I've completed the transition for one board (CircuitPlayground Express) and found the sum of the size of all the make_new functions in that build. Before, they summed to 9408 bytes in 57 routines. After, they sum to 9176 bytes also in 57 routines:

echo $((`arm-none-eabi-nm --radix=decimal --print-size build-circuitplayground_express/firmware.elf  \
 | grep make_new | awk '{print "+" (+$2)}'` ))

Assuming this size savings is accurate and the code still works, I think we want to switch back to the upstream way. I think the price is an additional creation of an args+kw sequence in some specific path, using a bit of heap memory.

[jepler]

@tannewt there's a nice size savings to reverting to upstream's make_new signature ^^^ so I think we should follow them.

Can you give me an idea what to test for subclassing native types? #1472 is about displayio.Shape. I did test that class NeoPixel(adafruit_pixelbuf.PixelBuf) still works, which I think fits the bill. The brightness= and other kwargs are successfully handled when constructing the native object.

[tannewt]

@jepler Thanks for the explanation. I think switching back is the right way to go. I don't remember what didn't work.