Is this an error?!

[edogaldo]

Arduino_Core_STM32/cores/arduino/stm32/clock.c

Line 99 in cf54e13

nb_loop = (((HAL_RCC_GetHCLKFreq() / 1000000)/5)*delay_us)+1; /* uS (divide by 4 because each loop take about 4 cycles including nop +1 is here to avoid delay of 0 */

In the comment you say "divide by 4" but in the code you divide by 5, is it an error?!

[fpistm]

Hi, in fact it was done by @fprwi6labs for the old F0 core:
https://github.com/fpistm/Arduino_Core_STM32F0/blob/a97ed35b1252e4fb667f354019279216de518af1/cores/arduino/libstm32f0/clock.c#L178
He probably copy/paste from F4 implementation to F0 and adapt the code without update the comment.
When I've merged all the core in one I did not seen this.

[edogaldo]

Looking at the code, precision seems not to be a main concern: the function always waits for delay_us+1 us and also the function doesn't consider the time required to calculate (((HAL_RCC_GetHCLKFreq() / 1000000)/5)*delay_us)+1 and for function call (stack allocation/deallocation, jump & back).
So, as long as the only difference is the constant loop duration (4 vs 5) and the usage of subs.w vs sub, why not simplifying and factoring everything?

[fpistm]

Yes, why not.
As I said this is linked to legacy code and probably some other clean/merge could be done elsewhere.