transaction: Implement Card::reconnect() wrapper

[kgraefe]

This gives us access to Card::reconnect() from within a transaction so that we can check the card status and reset it conditionally with no interference from other processes.

This fixes a race condition I had between two processes where both processes passed the status check resulting in two card resets.

[bluetech]

Just to make sure, you tested that reconnect works as expected inside of a transaction, the transaction doesn't end? I mostly expect it not to work.

One hitch with this is that it would technically allow doing tx.transaction(). Can you check what happens in this case? If it's just an error then I guess it's OK.

TBH, I regret using Deref, I now think it would have been better to add explicit wrapper methods... So maybe instead of DerefMut we should add an explicit reconnect method?

[kgraefe]

Just to make sure, you tested that reconnect works as expected inside of a transaction, the transaction doesn't end? I mostly expect it not to work.

Yes, I have tested it. It works and solves the race condition. In fact, we have been using SCardReconnect() within a transaction for years now with no problem. (I'm converting our C application to Rust.)

That being said, I can only tell for pcscd on Linux, not Windows or MacOS.

One hitch with this is that it would technically allow doing tx.transaction(). Can you check what happens in this case? If it's just an error then I guess it's OK.

Right, I did not think about that. I did some tests:

• I could create a second transaction and drop it immediately. Thus tx.transaction() does neither return an error nor block forever (which is what I expected). The outer transaction was still usable afterwards. However, it feels like an implementation detail of pcscd and might be different on other platforms or even with other versions of pcscd.
• In all other cases the borrow checker complained at compile time because we're having two mutable references to the same card at the same time.

TBH, I regret using Deref, I now think it would have been better to add explicit wrapper methods... So maybe instead of DerefMut we should add an explicit reconnect method?

I agree.

[bluetech]

OK, makes sense. So let's add a reconnect method on Transaction instead of DerefMut.