Possible fix for #11

[caternuson]

Can't just add logic in existing function since it would still hasattr. Need to dynamically provide it as needed.

Don't have an INA219 to test with original issue, but tested with an I2C FRAM and TCA:
tca_fram_test.py

import board
import adafruit_fram
import adafruit_tca9548a

tca = adafruit_tca9548a.TCA9548A(board.I2C())

fram = adafruit_fram.FRAM_I2C(tca[0])

fram[0] = 1
print(fram[0])

RPI/Blinka

pi@pizerow:~ $ python3
Python 3.7.3 (default, Apr  3 2019, 05:39:12) 
[GCC 8.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tca_fram_test
bytearray(b'\x01')
>>>

ItsyM4/CP

Press any key to enter the REPL. Use CTRL-D to reload.
Adafruit CircuitPython 4.0.2 on 2019-06-27; Adafruit ItsyBitsy M4 Express with samd51g19
>>> import tca_fram_test
bytearray(b'\x01')
>>>

[ladyada]

i thought we added it to blinka?

[ladyada]

oh right its on blinka (required) but not busio.

[caternuson]

Yep. And that all worked. New issue is that it can then fail on native CP since I2CDevice will try to call the linux func.

[tannewt]

I think we always want writeto_then_readfrom available so that drivers are written for it going forwards. (Otherwise they won't work with Blinka.)

So, on this PR detect in writeto_then_readfrom or in the constructor whether it is available and provide a substitute. Bonus points for adding writeto_then_readfrom into CircuitPython too. Thanks!

[caternuson]

Don't most drivers use I2CDevice and call write_then_readinto? It's then I2CDevice that turns that into platform specific calls. With the current logic, the simple existence of writeto_then_readfrom is seen as being linux:
https://github.com/adafruit/Adafruit_CircuitPython_BusDevice/blob/master/adafruit_bus_device/i2c_device.py#L145
Without changing that logic, this class needs to behave dynamically. So it's checking with hasattr and providing with setattr.

And in the future, when writeto_then_readfrom gets added to busio.I2C, it would still work.

[tannewt]

Yes, most do but this driver shouldn't assume I2CDevice is used (just like busio.I2C shouldn't.) It should always provide writeto_then_readfrom even if the underlying I2C bus doesn't. As is, it only adds it when the underlying implementation has it. Doing it only sometimes risks people writing code that won't work on Blinka.

I created a PR to remove stop= from CircuitPython in favor of writeto_then_readfrom here: adafruit/circuitpython#2012

[caternuson]

Is this PR even needed then? If CP's busio.I2C provides writeto_then_readfrom, then the existing version will just start working.

[tannewt]

It'll be needed for older versions. Want me to wrap it up?

[caternuson]

Sure. Works for me.

[tannewt]

@caternuson I've commited to your branch with what I think will work. Please let me know what you think. Thanks!

[caternuson]

Ah, OK. I think I was trying to avoid duplicating the logic from I2CDevice and just drive the behavior from here.

Not sure what it would be - but can you think of any issues in the future if the code in I2CDevice.writeto_then_readfrom changes and then code here is not updated to match?

[tannewt]

Ah, I think it's ok to duplicated it since it's temporary. There is no guarantee the this is used with I2CDevice so it's better to have this updated now.

I can't think of any issues but we can always fix things if it comes up.

[caternuson]

OK, cool with me. I just took care of the minor lints.

[tannewt]

Thanks! I totally missed that it had issues.