Right now, the Syntax spec is clear about what the value of a declaration is - it's all the tokens between the : and the ;. In particular, this includes any initial or final whitespace, which is pretty common in declarations like color: red; or in omitted-semicolon situations like:

.foo {
  color:red
}

When serializing non-custom properties, this whitespace doesn't matter; the property value has been parsed into a specialized data structure, and serialization just works off of that - it doesn't even attempt to preserve source whitespace. This means, for example, that you can be sure the first character of 'color' will be a non-whitespace char.

But the value of a custom property is the token list it parses as, so technically the initial/final whitespace is part of its value and should be serialized with it. Add to this the resolution at TPAC that custom properties should store their source representations so they can reserialize exactly as written (rather than, in some situations, serializing with extra comments inserted). On the other hand, this can result in unexpected whitespace in the serialization, as the initial/final whitespace isn't really considered to be "part of the value". (This is why I'm reporting it - one of the engineers on our team was writing some code with custom properties and got tripped up by this.)

So my question is: do we want to alter the "consume a declaration" algorithm to trim initial/final whitespace tokens? That dodges the resolution, as it won't be part of the value anymore. This will have zero effect on ordinary properties, just custom properties, and will likely bring custom property serialization slightly more in line with author expectations.