Change Adafruit Register libraries to use a library wide shared buffer

[grgrant]

@tannewt asked me to explore the possibility of a shared module wide buffer to reduce the memory footprint of the use of many register descriptors in a driver library. I also tried allocate on first use and it didn't make any improvement. The class-wide and module-wide shared buffer had the same improvement. I think module-wide would show improvement if any library used more than one non-inheritied class (like Struct and UnaryStruct) -- ROBits inherits RWBits. I'm not using any libraries that use both Struct and UnaryStruct.

I just want to say upfront that the savings was not as much as I had hoped for in libraries like MAX1704x, BMP5xx and TMP117. I'm perfectly OK if you want to just approve PR #62 which changes i2c_struct.UnaryStruct to allocate on init like all the other libraries. None of these are a must merge.

This module wide buffer change supercedes #62 and moves all buffers to a single bytearray buffer extended to the largest needed by the various register accessors.

Here are the measured savings I found:

I2C Struct/Bit(s)

Class	Current	Module Buffer	Saving
UnaryStruct	74*	58	22%
RO/WBits	116	84	28%
RO/WBit	90	58	36%

• This is the measurement of the Move i2c_struct.UnaryStruct's byte array to init #62 allocate in init to be equivalent to the rest of the classes

Register Struct/Bit(s)

Class	Instance	Module Buffer	Saving
UnaryStruct	80	48	40%
RO/WBits	154	112	27%
RO/WBit	96	64	33%

I've been running these versions on 4-5 of my temperature/humidity/CO2 sensors using MAX1704x and TMP117 (i2c_struct/Bit(s)) and also BMP5xx (register_Bit(s)) for about a week with no issues.

Also tagging @FoamyGuy as he was the author of the register_Struct/Bits versions

[tannewt]

One reorganization suggestion. I do think the slower execution is worth the saved memory.

[grgrant]

@tannewt

Instead of putting this in each file, put it in __init__.py and import it in each. That way all of these files will share it and the implementation will only be in one place.

Done. I've tested this library for three days with 3-TMP117/119s, 2-BMP5xx, 3-MAX1704x and 1 DS3231 (for the bcd libraries). I don't have anything that uses i2c_struct_array.

If CircuitPython ever goes multi-threaded the shared buffer might cause an issue

Edit: While working with another library I realized I should have used mem.alloc rather than mem_free to compare library growth. So numbers below are redone to be more correct.

New AdafruitRegister footprint savings for libraries

Here are the savings over the currently deployed production version of the library:

library	prd	new lib wide	Savings
BMP5xx	15488	14720	768
TMP117	12832	12816	16
MAX1704x	9952	9936	16
DS3231	9056	8928	128

Footprint measurement methodology

In most cases even if the prod/new values changed the savings stayed the same. I'm using prod and new library files compiled with the same mpy-cross-macos-10.2.1-arm64. I also found that I needed to disable everything settings.toml, boot.py, repl.py and code.py and reboot the MCU from time to time to get the most consistent numbers. I used the following snippet (changed for each library) pasted to REPL and then did a soft reset between -- unless the number was really weird and then I'd do a microcontroller.reset.

# Fixed version to use gc.mem_alloc()
# BMP5xx
import gc
gc.collect()
b=gc.mem_alloc()
import adafruit_bmp5xx
gc.collect()
a=gc.mem_alloc()
b-a

Old mem_free wrong numbers for posterity

Interesting that the savings were the same except for TMP117.

library	prd	new lib wide	Savings
BMP5xx	24416	23648	768
TMP117	21824	21760	64
MAX1704x	2096	2080	16
DS3231	1104	976	128

[tannewt]

I don't see the new changes. I suspect you forgot to push them.

[grgrant]

@tannewt

I don't see the new changes. I suspect you forgot to push them.

I pushed them on Friday and then briefly changed to draft while I did some memory footprint anomaly checks. Then I changed it back to ready for review. Here is the link from above of the Friday Push: 17117b0

I did a second push -- that appeared empty -- today and got a new CI set of tests.

[tannewt]

Thank you!