Remove static {read,write} buffers

[andrewgazelka]

• Fixes #17
• Fixes #10
• This is essentially a copy of #13 but with the documentation change requirements that were never made. Since the mentioned PR has been stale since March, I found it fair to make another PR.

I am potentially concerned regarding a use case for non-static buffers in the previous docs in the embedded-dma:

The reason we don't require 'static in the traits themselves is because it would block implementations that can deal with stack allocated buffers, like APIs that use closures to prevent memory corruption.

Can someone provide an example of an API that use closures to prevent memory corruption? If closures have a valid use case, #18 might be a better option.

[korken89]

Is there something hindering progress here? :) Would be super to get this and const generics support

[andrewgazelka]

Thanks! Didn't know about that website.

[thalesfragoso]

Can someone provide an example of an API that use closures to prevent memory corruption?

Probably something akin to crossbeam's scoped threads (scope function).

I haven't seen any on the DMA side though, and not sure how useful it would be, i.e. you would only be able to do parallel work (w.r.t DMA) inside the "transfer closure", which wouldn't allow for doing work in different contexts (e.g. different interrupts / main). So maybe we should focus on being better suited for the common case, as #10 mentioned, it happened a few times that DMA implementations forgot to bound to 'static.

A closure based transfer could probably create its own buffer wrapper too, so there's a workaround even for that case.

[andrewgazelka]

Can someone provide an example of an API that use closures to prevent memory corruption?

Probably something akin to crossbeam's scoped threads (scope function).

I haven't seen any on the DMA side though, and not sure how useful it would be, i.e. you would only be able to do parallel work (w.r.t DMA) inside the "transfer closure", which wouldn't allow for doing work in different contexts (e.g. different interrupts / main). So maybe we should focus on being better suited for the common case, as #10 mentioned, it happened a few times that DMA implementations forgot to bound to 'static.

A closure based transfer could probably create its own buffer wrapper too, so there's a workaround even for that case.

ok, that makes sense.

[andrewgazelka]

@thalesfragoso @eldruin I'd really appreciate if this could get merged soon. Is there anything else you would like me to change? Thanks so much!

[andrewgazelka]

@eldruin done

[andrewgazelka]

@thalesfragoso are you also able to run the workflow? Thanks! :)

[bors[bot]]

Build succeeded:

• ci-linux (1.37.0, x86_64-unknown-linux-gnu)
• ci-linux (stable, thumbv6m-none-eabi)
• ci-linux (stable, thumbv7m-none-eabi)
• ci-linux (stable, x86_64-unknown-linux-gnu)