Genuine decimal numbers

I just looked at your code on Github (with a view to answering Shabaz's questions)

You obviously want to share, which is great, but your code does not make it easy.

decimal.py is 945 lines of code with no comments or explanation other than:

#

Do you have some documents or notes that explain how it works and how it might be tweaked or modified or maintained ?

MK

Have you looked at decimal_math.py or pi.py or the microbit versions microbit_root.py and microbit_pi.py

I have, They are examples of how ot use decimal.py but they don't explain how decimal.py is actually meant to work.

MK

I have, They are examples of how ot use decimal.py but they don't explain how decimal.py is actually meant to work.

MK

It just performs basic mathematics to a magnitude and precision only limited by memory, try

decimal.divide(decimal.Number(1), decimal.Number(3), 1000)

Hi,

Most of us won't have a Micro:bit ready to test this immediately, so this is why I'd hoped for some example output. Sure, we could use the normal Python version instead of the Micropython version, but there will be a significant difference so that's not as useful.

Anyway, here is a summary of Micropython floating-point for those interested. I only know this because I spent a couple of days investigating it (not on micro:bit):

* Micropython is very similar to Python, but unfortunately some math-related libraries cannot currently work with Micropython

* Micropython might use single-precision maths, it depends on how Micropython was compiled. I suspect it is single-precision on Micro:bit since rhubarbdog mentions 32-bit floats in an earlier comment.

* Micropython can be re-built for double-precision, and that's highly recommended as a first step if better precision is needed, because it's seamless, no need to treat numbers differently, the result will just be more precise. Personally I wish double precision was the default for most boards, but I believe it isn't. For Pi Pico, I've built some double-precision releases of Micropython, to save the effort building them, but the effort to build it isn't hard, it is documented here (along with the links to the built images):  Pi Pico (RP2040) MicroPython Double Precision  For micro:bit someone else might have done this too, but I don't know.

* Don't assume that Micropython calculation results will be the same as with CircuitPython, or normal Python. All are different. CircuitPython is normally the worst at it, with clear errors in places.

* Another option is to partially code in C and partially in Python, and that way theoretically you can compute using the maths library for C, however there be dragons with that method too, in particular in some scenarios it is very hard to link to the maths library for some devices (it is to do with soft floating point for some processors, that probably won't apply to Micro:bit since that has quite a powerful chip from memory).

* Even double-precision isn't always precise enough, hence the need for (say) quad precision. There are Python (not Micropython) libraries such as decimal.py which can provide great precision, for example on normal Python:

>>> 100/3
33.333333333333336

The above is obviously not entirely correct, since it has rounded up at the end.

In contrast, with decimal.py on normal Python:

>>> Decimal(100)/Decimal(3)
Decimal('33.33333333333333333333333333')

*  The decimal.py that works on normal Python doesn't work on Micropython, partially due to some features which Micropython does not support, but although some of those can be worked around, decimal.py uses a lot of resources, and it is easy to blow the amount of RAM that is available.

* There is a subset of functionality from the normal decimal.py that is nevertheless handy to have with MicroPython, I have tried one library which worked for me (I did not try on Micro:bit, only on Pico), it is called mpy_decimal.py and it works very well, providing as much precision as requested. The code has comments, as michaelkellett mentions comments are highly recommended otherwise it will be near-impossible for people to figure out how to improve your project, so perhaps you can use that as an example of how to do it, I feel (others may disagree) that author put in decent comments in that project. I have done a couple of spot-checks with that mpy_decimal library and the results are accurate too (more accurate than I could prove on my own, I had to compare with online Wolfram Alpha for that reassurance).

* I could go on, but I think those are the main points that were discovered.