Simplify parsing of crate versions

[dtolnay]

I get lost when I see things like Semver::parse(j["vers"].as_str().unwrap()).unwrap().

[nabijaczleweli]

What you've done there is you've pulled another, like, 2-3 crates in and made a structure you deserialise into, then out of? What? How does that simplify anything at all?

[dtolnay]

Thanks for the explanation @nabijaczleweli! I obviously disagree but I would love to understand your perspective better if you have the patience to explain.

You mentioned that there is a struct we are deserializing into then out of, but notice how the original code actually does much worse than that:

• Original code: R -> String -> JsonValue -> Option\ -> String -> Semver
• New code: R -> struct -> Semver

So that is one aspect of what I meant by simpler. The other aspect has to do with error conditions:

• Original code can fail because:

  1. Can't read from the reader
  2. Reader does not contain valid UTF-8
  3. Can't parse the JSON
  4. The "yanked" field doesn't exist
  5. The "yanked" field is not a boolean
  6. The "vers" field doesn't exist
  7. The "vers" field is not a string
  8. The "vers" string is not a semver version

• New code can fail because:

  1. The input data doesn't match what we expect it to look like

Literally that's it. The struct is a way to declare what we expect the input data to look like, and if it doesn't look like that, then that is the only way anything can fail (on line 284).

Yes serde_json has 3 new dependencies (not counting serde because you already depend on that) but all of them are extremely small. I understand minimizing dependencies so please be honest if that is the primary reason for deciding a particular way.