Fix empty token on swift package-registry login (Issue #7453)

[pwallrich]

Fix empty token when adding a new package-registry and adding token interactively on release builds

Motivation:

When executing swift package-registry login it asks for a token but fails to retrieve it properly. It sends an empty string to the registry server and login therefore fails. (see #7453)

Modifications:

I've debugged the code a bit and found out, that somehow buffer and passwordPtr don't seem to be holding correct values after the readpassphrase call.

• An easy quick fix is to make sure that buffer doesn't get deallocated by adding smth like _ = buffer after the String init. This works but is not nice.
• My first try was to use buffer instead of passwordPtr to create the string. These works, but I remember reading somewhere that & can be quite nasty sometimes.
• I also tried buffer.withUnsafeMutablePointer. However readpassphrase seems to not only change the content of buffer but move it. This leads to a runtime failure Fatal error: Array withUnsafeMutableBufferPointer: replacing the buffer is not allowed

Result:

The buffer is retained, the token is properly parsed and sent to the server.

[MaxDesiatov]

Hi @pwallrich, thank you for the contribution? Would you be able to add a test for this to our test suite to prevent possible regressions in the future?

[pwallrich]

Hey @MaxDesiatov. Could be quite hard, since it only happens in release builds. But I'll try to find a way to trigger it reliably in tests.

Do you know if there already exist tests for things like that so I wouldn't have to reinvent the wheel?

[pwallrich]

Ok I've managed to write a test for it, but it's actually a bit nasty since I have to change the stdin.
Did that based on https://forums.swift.org/t/how-to-read-from-standard-input-during-unit-tests/43237/.

I also had to create a separate testTarget since the test needs to be built with optimisation and therefore can't use @testable. I don't know, how I would add to the CI.

The more I think about this bug the more I question whether reverting the change introducing readpassphrase would make sense. So use getpass again and add an additional error for users who use tokens >128 characters telling them to either use --token or --token-file. But I guess that would need a bigger discussion.

Do you have any thoughts about that?

[MaxDesiatov]

@swift-ci test