Alias $type inheritance? about community-group HOT 9 CLOSED

I think the $type alias is to allow for parsers just to know to check an object and its properties. Otherwise how would it know to check for colors? It’d have to assume a token structure (like all colors are under top level color property).

It also helps explicitly define the type of the token, since you might want to mix tokens to achieve another (like combining opacity number and RGB color to make a RGBA new color). Or for example, if you multiply fontSize and a spacing token, what is the type of the result? Might get weird for parsers.

I do think there you have a great idea of not including it every time though. It seems verbose and makes it difficult to manually parse JSON without a tool to strip the noise. Not sure how to practically achieve that without an explicit structure or some other system to accommodate for the inference of types.

Also this issue emphasizes the need for a section of the docs on parsing, and expectations the spec has - to create some consistency between tooling.

from community-group.

I think the $type alias is to allow for parsers just to know to check an object and its properties. Otherwise how would it know to check for colors? It’d have to assume a token structure (like all colors are under top level color property).

Well that’s just the thing though—parsers actually have to validate aliases currently. Aliases have to be non-circular. And non-aliases have to be valid tokens. By just accepting an invalid schema at face-value, the user will get errors with no clear action on how to fix it. So parsers actually do have to scan the alias and know it’s referring to a color anyway, and there’s no way for them to avoid having that work (or they’d be bad, hard-to-use-tools). But I was more interested in exploring whether that’s an error to show to the user “fix this!” or if it just silently works

Also this issue emphasizes the need for a section of the docs on parsing, and expectations the spec has - to create some consistency between tooling.

💯!

from community-group.

The spec lists out the following resolution order:

A token's type can be specified by the optional $type property. If the $type property is not set on a token, then the token's type MUST be determined as follows:

• If the token's value is a reference, then its type is the type of the token being referenced.
• Otherwise, if any of the token's parent groups have a $type property, then the token's type is inherited from the closest parent group with a $type property.
• Otherwise, the token's type is whichever of the basic JSON types (string, number, boolean, object, array or null) its value is.

So, in your examples

{
  "color": {
    "a": { "$type": "color", "$value": "#336699" },
    "b": { "$value": "{color.a}" }
}

This is valid, and color.b has a type of color.

{
  "color": {
    "a": { "$type": "color", "$value": "#336699" },
    "b": { "$type": "dimension", "$value": "{color.a}" }
}

This is invalid; the spec says that a token with the type dimension must be "a string containing a number (either integer or floating-point) followed by either a "px" or "rem" unit". Since {color.a} doesn't match this, this is an invalid token.

{
  "base": {
    "$type": "color",
    "a": { "$value": "#336699" }
  },
  "semantic": { "$value": "{base.a}" }
}

This is valid, and semantic would get resolved to having the type of color (since base.a has the type of color).

The spec is pretty unambiguous about these, though it might need a bit more language to acknowledged that a referenced token might inherit its type, too, either from its group or another reference.

I can’t think of a good reason for even declaring $type on aliases

Yes, almost any type on a token with a value like {base.a} will result in an invalid token. There's probably some weird edge cases where you can have a token with a fontFamily type which, according to the spec, would be valid, resulting in css like font-family: '{base.a}' ... but resolving this would probably create substantially more edge cases than it would solve.

from community-group.

There's probably some weird edge cases where you can have a token with a fontFamily type which, according to the spec, would be valid, resulting in css like font-family: '{base.a}' ... but resolving this would probably create substantially more edge cases than it would solve.

Oh I hadn’t even thought of this, but you’re right just evaluating a token’s $type by parsing its value could lead to many errors. Especially if string tokens ever were accepted (not that they are in the DTCG now, but mainly thinking of Figma variables’ current implementation where string types can refer to colors or typography values).

The spec is pretty unambiguous about these, though it might need a bit more language to acknowledged that a referenced token might inherit its type, too, either from its group or another reference.

Agreed—I think clarifying this would help. The resolution definition you provided earlier implies this, but doesn’t outright explicate it, and it could be clearer.

Thanks for clarifying! This seems like it’s not really ambiguous, then, how the spec is interpreted. More just a possible documentation TODO of “should alias types get a little more specific regarding this behavior or not”.

from community-group.

We are experiencing 2 simular problems:

1. aliases should as state include the type they refer to
2. there are not enough types to describe the data

Let's first dive into the aliases, but the second seams related to so I added that in the end.

The aliases would have a solution to include a type in xml I beleive as an attribute can hold the class. But json does not have attribute in its spec. In joson I would like to propose a way to include the type.

{
  "background": {
    "gradient": {
      "value": {
        "path": "gradient.default.1",
        "type": "color"
      },
      "type": "reference"
    }
  }
}

This reads as: "at the path background.gradient.value there is a reference to a node of type color". This way we can include the type in the json.

My second remark is about the different allowed types in the Design Tokens spec. Could someone refer me to another related issue?

Remark: When writing a parser for a color value, one needs to parse a single color as color=hex, a gradient is of form colors=[(hex, 50%),(hex, 10%),(hex, 30%)]. This represents a distinctly different schema.

To conclude might I suggest to preferably add more types to the spec but definatly use the json structure to include the reference type that has a path and a type as properties?

from community-group.

I thought a lot about the issues you mention @doozMen, and I landed on the same conclusion: the json tree lacks data.
That said, the format specification aims for user-friendliness rather than ease of parsing when tradeoffs like this one shows up. I don't think we'll get the format to host the missing pieces.

BUT, we can still specify a mapping for allowed aliasing matches among tokens:

// Example with Border token
const borderTokenTypeMapping = {
  _unionOf: [
    { _tokenType: borderTokenTypeName },
    {
      _mapOf: {
        color: colorTokenTypeMapping,
        style: strokeStyleTokenTypeMapping,
        width: dimensionTokenTypeMapping,
      },
    },
  ],
} satisfies TokenTypesMapping;

Full version of the code

Beside this off-initial-topic answer, shouldn't we close this issue?

from community-group.

I’m happy to close the issue, yeah. There might be a TODO on just adding some additional clarification in the documentation, but no spec changes are necessary IMO.

from community-group.

I thought a lot about the issues you mention @doozMen, and I landed on the same conclusion: the json tree lacks data. That said, the format specification aims for user-friendliness rather than ease of parsing when tradeoffs like this one shows up. I don't think we'll get the format to host the missing pieces.

BUT, we can still specify a mapping for allowed aliasing matches among tokens:

// Example with Border token
const borderTokenTypeMapping = {
  _unionOf: [
    { _tokenType: borderTokenTypeName },
    {
      _mapOf: {
        color: colorTokenTypeMapping,
        style: strokeStyleTokenTypeMapping,
        width: dimensionTokenTypeMapping,
      },
    },
  ],
} satisfies TokenTypesMapping;

Full version of the code

Beside this off-initial-topic answer, shouldn't we close this issue?

Thanks for taking my off-topic suggestion into account. Your suggestion to have a mapping slightly slows down the parsing and at least for me is less readable but if this is possible is there an issue I can track to when this mapping description lands in the spec?

from community-group.