Make TryInto::Error return the original value instead of a str

[ninegua]

When using the derived TryInto/TryFrom instances, if the result is an Error, the original value is consumed (if it is not a reference). This prevents the valid use case of chaining several try_from or try_into together, at least not without potentially expensive clones.

Returning a string error message is descriptive if the user program actually prints it, but IMHO that is not a very common use case because reliable programs should always handle errors. An additional reason against string error is that it is not friendly for localization/translation.

I could change the Error type to be a tuple that contains both the string & the original value if people really must have the string back.

[JelteF]

Thanks for this PR, a few initial comments:

1. This can only be released as part of 1.0.0, since this is a breaking change.
2. Can you add a test that does the chaining?
3. The changed doctests do not compile. The easiest is usually to use cargo expand to get the right code.

[JelteF]

And I'm a bit undecided if I'd like to keep the user friendly error messages too. They are quite helpful in cases of a programming error.

[ninegua]

Took me a while, but managed to get it pass all the tests now.

They are quite helpful in cases of a programming error.

So would you rather prefer a tuple (str, T) as the Error type? Or other suggestions?

[ninegua]

One possibility is to support returning the original value through Err as an optional feature. This will not break backward compatibility.

[JelteF]

One possibility is to support returning the original value through Err as an optional feature. This will not break backward compatibility.

I think this is good idea. Could you implement that? I don't think most people will require the original type to be returned. Something like:

1. #[try_into(error_type(string))], this is the default
2. #[try_into(error_type(input))], this would enable your desired behaviour

[JelteF]

removing this from the 1.0 milestone as this doesn't need a breaking change

[JelteF]

Added back 1.0 milestone, because I think returning an error by default that actually implements Error would be really good.

[JelteF]

Closing this since there was no response from the original author, and I don't think the approach is the right one. let's continue to discuss in #173