Delay: Don't rely on EH traits

[David-OConnor]

File in question:

cortex-m/src/delay.rs

Line 32 in 1a4e096

fn _delay_us(&mut self, us: u32) {

Typically, HALs implement their own Delay. Given it only uses cortex-m peripheral, there's no reason HALs need to keep this; this cortex-m library includes a delay.

However, it's dependent on embedded-hal::Delay. Can we make the implementation native, and add EH as an optional add-on? This way, you don't need to import the trait, or include it in your project. I'll do a PR if approved.

[David-OConnor]

In case it helps, the use case is this:

• I have a STM32 HAL library that implements its own Delay (In a similar way to other cortex-m-based HAL libraries). It uses the cortex_m systick peripheral, and no stm32-specific peripherals. I think this, and other HALs can delete their Delay modules, and direct users to use the one here, in cortex_m instead.

• I don't think user code should rely on embedded-hal traits; their main use case is in writing generic drivers. Currently, I'm unable to delete the Delay module and defer to the cortex-m Delay, since it would require user code to import embedded-hal, since the cortex-m Delay requires it.

[Disasm]

I still don't get why cortex-m Delay driver shouldn't rely on embedded-hal while HALs can rely on embedded-hal for the same delay driver.

[David-OConnor]

HALs shouldn't IMO. Generic drivers should.

[Disasm]

Yes, and which crate should provide these DelayMs/DelayUs implementations for generic drivers?

[David-OConnor]

The HALs provide them as a secondary API that wraps the native primary one (as in the PR associated with this issue). The drivers use them, and import the trait.

User code that flashes to an MCU doesn't need them.

@Rahix has a good write up here

[Disasm]

So you suggest that cortex-m has a Delay driver, but DelayMs/DelayUs should be provided by HALs? I don't understand why it should work like this. What if I have a generic Cortex-M (FPGA) core and no HAL?

[David-OConnor]

No - I'm suggesting cortex-m supply a native impl, and an embedded-hal impl - as in the PR. And HALs delete their Delay module, as I'm prepared to do. I'm currently wrapping this cortex-m Delay instead of directing users to use it, since I don't want users to need embedded-hal.

[Disasm]

Ohh, I forgot that now we have fallible-only Delay* traits. Now it makes sense.