Why does Micropython 'const' not accept float values?

Question

Most Micropython ports contain a 'micropython' module, which has a specific function called 'const()'. I am led to believe it behaves a lot like '#define' in C, however, it only accepts integers. Does anyone know why this is? You can declare float constants in C, so why is this unavailable in micropython?

I want to efficiently store values which are used to convert between units, but many of them are float values... (I thought about multiplying them by 10^x and then just dividing by that at runtime, but this will likely take just as long as storing the float in a variable and letting the code use the lookup table).

Any ideas why we can only declare integers with micropython.const()?

Cheers :)

Adz

"it behaves a lot like '#define' in C" #define is a pre-processor command and therefore works completely different. – Discommend

@KlausD.: MicroPython's const is recognized by the MicroPython compiler, though, so it behaves a lot more like #define than anything you could do in CPython. You can't do any complex macro stuff, but const values will be directly substituted into their usage sites (within a module, at least), and non-exported consts don't require a storage location. – Zouave

ah ok, thanks @user2357112supportsMonica and KlausD, I understand about the preprocessor commands and see the slight difference in usage. Is lack of floating-point support perhaps to do with the different space requirements for integers and floats? – Adz

MicroPython only accepts so called 'small integers' for const; 'large' integers (i.e. integers which don't fit in a machine word minus some bits) and floating point values need to be allocated on the heap. const replacement etc is done when parsing I think, before compiling. As to your actual question, i.e. why that wouldn't work for heap-allocated objects: I guess technically it could be possible but it might be too hard, not sure.. – Andraandrade

Answer 1

With the recent release of v1.19, you can apply const to any object. See PR #8548 for some of the details.

In short, the reason const was initially constrained to ints was simply because the implementation to do so was relatively straightforward.

Varner

Answer 2

This is by design, the objective of micropython.const() is to save RAM on the MCU therefore according to the docs:
the argument to const() may be anything which, at compile time, evaluates to an integer. http://docs.micropython.org/en/latest/reference/constrained.html?highlight=const()#execution-phase

Spruik

How would this save RAM though? I thought unlike CPython, Micropython floats are 32 bits (unlike CPython 64 bits equivalent to a double) so they occupy the same space as an integer? – Adz

@birdistheword99, the details are explained in the 1st paragraph of linked reference documentation. the compiler will avoid coding a lookup to the name of the constant by substituting its literal value. This saves bytecode and hence RAM. – Spruik

MicroPython integers are signed 31-bit smallint (1 bit is reserved to flag a pointer, but i can't find a document reference to that) – Spruik

thank you for the information, and I apologise for being dumb here, but going back to the original question, if the compiler saves RAM by substituting a lookup to the integer for the actual integer, why can't it do the same thing with a float? Why can't FOO = const(1.234) tell the compiler to substitute FOO with 1.234 and save bytecode the same way? – Adz