09:26 <ryan-summers[m]> <paulyoung> "Hi. What I meant was, I’m pretty..." <- Ah yes indeed, USB works in such a way as it is always host-powered. If your adapter is a USB-OTG device, it _could_ power your device, but those are pretty rare in practice. In reality, that's usually something like a smart phone

09:26 <ryan-summers[m]> <paulyoung> "Maybe I’ll just get an F4..." <- I would highly recommend against making a USB host on an embedded device. It is _very_ complicated and I don't know if it's well supported.

09:53 <MathiasKoch[m]> James Munns:... (full message at <https://catircservices.org/_matrix/media/v3/download/catircservices.org/fdCRMpIxADrRnwCqOJnBxhxy>)

09:54 <JamesMunns[m]> not generally. IIRC, those non-self-describing code paths require in the general case to act pretty "hash-mappy", which doesn't fit well at all with postcard

09:54 <JamesMunns[m]> (that's why its marked WontImplement)

09:57 <JamesMunns[m]> If you can share a reproduction, like a git repo that does the serde json core round trip, then the postcard attempted round trip, either as a bin crate so I can run it on a desktop, or as a lib with a unit test or something, I can look at it in more detail, but my gut feeling/memory of this is "postcard can't really handle this"

09:58 <JamesMunns[m]> If you make up that repro, I can at least promise a detailed response of why not, or workarounds I can think of.

09:59 <JamesMunns[m]> One potentially annoying workaround is to actually make two (very similar) structs, one intended for json, and one for postcard, and impl `from` between them, so you can do something like:... (full message at <https://catircservices.org/_matrix/media/v3/download/catircservices.org/tagMkwvuMKlFwBpgYxCqmIqZ>)

09:59 <MathiasKoch[m]> Absolutely. I have a pretty simple unit test running here, so I can just shell that out to its own repo.. hang on :)

10:00 <MathiasKoch[m]> JamesMunns[m]: > <@jamesmunns:beeper.com> One potentially annoying workaround is to actually make two (very similar) structs, one intended for json, and one for postcard, and impl `from` between them, so you can do something like:... (full message at <https://catircservices.org/_matrix/media/v3/download/catircservices.org/zelnDrJQFIWfhxEMtydCdCIc>)

10:01 <JamesMunns[m]> MathiasKoch[m]: Yeah, it'd basically just be copy paste, and one has the flatten attribute and the other doesn't, IMO? It feels unsatisfying, but if it's just one type, that might be a pretty straightforward way of doing it.

10:07 <MathiasKoch[m]> https://github.com/MathiasKoch/adjacent_enum_postcard

10:15 <JamesMunns[m]> Working on something else ATM, will try to take a peek today

10:15 <MathiasKoch[m]> No problem, just happy you want to take a peek 👍️ Thanks !

13:10 <GiacomoPasini[m]> Hey, I have a question about portable-atomics. How do I patch my entire dependency tree to use that instead of core::sync::Atomics? I guess I'm not the first one that wants to use a no_std package on a platform without atomics but I can't find an easy way to patch dependencies (don't think I can use a cargo patch on core)

13:30 <danielb[m]> 😭

13:30 <danielb[m]> are you using esp32?

13:31 <danielb[m]> * using esp32 by any chance?

13:39 <GiacomoPasini[m]> mainly targeting riscv32im

14:02 <ryan-summers[m]> I think the main way we have done this in the past is just update all the dependencies in the chain to use the atomics shim

14:02 <ryan-summers[m]> But that obviously becomes super painful with a long dep chain

14:22 <korken89[m]> Does anyone else use the `-D warnings` to get error on warnings in CI? When I do this for embedded targets the don't link properly and I'm not sure why.... (full message at <https://catircservices.org/_matrix/media/v3/download/catircservices.org/PRvKzTWdOdfczQXCTLMwWMbS>)

14:22 <korken89[m]> s/the/they/

14:23 <korken89[m]> I think it has something to do with the linker variables not being passed correctly anymore, as if the RUSTFLAGS is completely overwriting existing flags

14:25 <GiacomoPasini[m]> <ryan-summers[m]> "But that obviously becomes super..." <- are portable-atomics a drop-in replacement for their core counterparts? How stupid would be to have build script that replaces imports (if possible at all)?

14:28 <korken89[m]> On my issue: https://github.com/rust-lang/cargo/issues/5376

14:28 <ryan-summers[m]> GiacomoPasini[m]: Generally, `atomic-polyfill` is used in embedded. I think it may have came after, but it's a drop-in replacement from what I remember

14:31 <GiacomoPasini[m]> atomic-polyfill is deprecated and suggests to use portable-atomics, that's why I was looking into that

14:32 <ryan-summers[m]> Ah gotcha, they keep changing

15:41 <romancardenas[m]> https://github.com/riscv-rust/e310x-hal/blob/master/Cargo.toml

15:41 <romancardenas[m]> This is a riscv32imc board that uses atomic-polyfill, maybe it helps you

15:42 <romancardenas[m]> Here you can see how we deal with it: https://github.com/riscv-rust/e310x-hal/blob/057d29839471c9ed00fb79748a4faf0292c782ef/src/gpio.rs#L6

15:42 <romancardenas[m]> (this board is a bit weird and supports atomic operations partially, so that's why we have weird feature flags)

15:56 <GiacomoPasini[m]> thank you!

15:57 <GiacomoPasini[m]> my problem is that the project I would like to use has lots of dependencies and would be quite consuming to contribute a patch like that to each of them

15:59 <romancardenas[m]> Yes, I know... if I'm not wrong, libraries should look like this: https://github.com/romancardenas/riscv-slic/blob/main/riscv-slic/Cargo.toml

15:59 <romancardenas[m]> So they always provide a set of features to propagate the configuration of portable-atomic

16:00 <romancardenas[m]> This is very annoying though, but I don't know if there is any workaround

16:01 <dirbaio[m]> <korken89[m]> "I think it has something to do..." <- try passing linker args from build.rs, these don't get overwritten if you set RUSTFLAGS

16:01 <dirbaio[m]> https://github.com/embassy-rs/embassy/blob/main/examples/stm32c0/build.rs#L2-L4

18:29 <romancardenas[m]> Can anyone send me a few pointers to get external interrupts working in qemu for riscv? Currently I have simple applications working, but when I rely on external interrupts... They never happen 😢

18:31 <romancardenas[m]> Also delays based on polling do not introduce delay at all

20:14 <dkm[m]> <waveguide[m]> "I wonder too how many modificati..." <- In the case of AdaCore's compilers based on GCC, the modifications are very small and everything that makes sense is sent upstream.