Make libosdp no_std compatible

[Sympatron]

Depends on #16.

The feature arc is introduced, because some targets like thumbv6m-none-eabi don't support CAS operations and therefore don't support Arc. By disabling this feature these targets can be supported, too. I will probably looking into implementing get_version() and get_source_info() some other way that does not rely on CAS operations, but this way libosdp can at least be built for those targets now.

[Sympatron]

I changed get_version() and get_source_info() to return &str. That way once_cell is not needed anymore. This is fine since the C implementation returns a statically allocated const char*.

[sidcha]

So, two things,

• I added 8b8416c to make rebase on this branch easy. Please rebase it on master an the libosdp-sys commit in this PR should be dropped.
• Why does libosdp-sys need std feature at all? maybe we can make both libosdp-sys and libosdp no_std only? Do you see any downsides to this?

[Sympatron]

• I added 8b8416c to make rebase on this branch easy. Please rebase it on master an the libosdp-sys commit in this PR should be dropped.

Ok, I will do that.

• Why does libosdp-sys need std feature at all? maybe we can make both libosdp-sys and libosdp no_std only? Do you see any downsides to this?

For libosdp-sys you are definitely right. It does not use anything from std and should be no_std only.
libosdp does use no_std incompatible stuff like thiserror and std::io::Error though.

[Sympatron]

Done

[sidcha]

Apart from those two nits, LGTM. Thanks.

[sidcha]

embedded_io::ErrorKind::TimedOut cannot be put to ChannelError::WouldBlock since embedded_io defines read/write to always block, there is no equivalent to ChannelError::WouldBlock. so this comment-line can be removed.

This probably indicates a deeper issue as LibOSDP expects all read/write to be non-blocking. We need to explore ReadReady/WriteReady traits and adjust the channel callbacks to handle the outcomes correctly.

[Sympatron]

You are right. But currently we are only using embedded_io::Error, so that is mostly a problem implementers of Channel will have, not of libosdp directly.

This should probably be pointed out more clear in the doc comments of Channel though.

A quick google search revealed that none of the HALs I could find implement ReadReady/WriteReady currently. I think WriteReady is not stricly necessary, but Read alone will simply not work as expected. That is indeed problematic.

Nevertheless, I don't think this has to block this PR. It can be handled in a future PR.

[Sympatron]

BTW this is a semver breaking change, because of the change to get_version() and get_source_info().

[sidcha]

True, but I think we should wait until we have some level of API stability before enforcing SemVer.