fix(semver): Allow unsetting build metadata

[justinmchase]

The main issue is if you pass "" as the build metadata to the semver.increment function it doesn't actually unset the build metadata as expected because of a falsy comparison of the parameter rather than a strict metadata === undefined.

This will now alter the behavior of the increment function to have no changes when you specify undefined but to actually set the build metadata to [] if you were to pass "" into the function.

Additionally, there was what I think to be a "bug" discovered in the constructor which was making it very hard to test, which is a bug where the this.raw field was set before the call to this.format(), which ends up retaining a wrong / different version than if you call increment.

Namely if you create a semver with build metadata the build metadata will be in the raw field. If you then increment it, then it will fall out of the raw field even if the build metadata is unchanged.

This happens because in the increment function, it is calling this.format() and then setting the raw field to the version but in the constructor it does the reverse. This means the raw will change based on if you've called increment or not.

Also, the whole design of this class probably needs a major revamp to be immutable. Having mutations happen on functions such as .format() is really problematic when the object is this complex. If the std lib ever hits 1.0.0 I would strongly recommend putting this chore onto the backlog.

Before

const v = semver.parse("1.2.3+0");
v.increment("patch", undefined, "");
console.log(v.format({ style: "full" })); // 1.2.4+0

After

const v = semver.parse("1.2.3+0");
v.increment("patch", undefined, "");
console.log(v.format({ style: "full" })); // 1.2.4

const v = semver.parse("1.2.3+4");
v.increment("patch");
console.log(v.format({ style: "full" })); // 1.2.4+0

[kt3k]

LGTM. Thanks for the fix!

Also, the whole design of this class probably needs a major revamp to be immutable.

This makes sense. Created an issue #3164