[TCGC][Bug] getAllModels, Optional literal do not have correct generated name

[haolingdong-msft]

TSP definition:

model Model {
  literal: "literal";
  optionalLiteral?: "optionalLiteral";
}

Java code:
In Java's generated code, we will generate an enum for the optionalLiteral, the enum is called ModelOptionalLiteral, like below. I guess this optional literal is like anonymous model, and will need TCGC to generate a correct name for it.

public enum ModelOptionalLiteral {
    /**
     * Enum value optionalLiteral.
     */
    OPTIONAL_LITERAL("optionalLiteral");

    /**
     * The actual serialized value for a ModelOptionalLiteral instance.
     */
    private final String value;

    ModelOptionalLiteral(String value) {
        this.value = value;
    }

    /**
     * Parses a serialized value to a ModelOptionalLiteral instance.
     * 
     * @param value the serialized value to parse.
     * @return the parsed ModelOptionalLiteral object, or null if unable to parse.
     */
    public static ModelOptionalLiteral fromString(String value) {
        if (value == null) {
            return null;
        }
        ModelOptionalLiteral[] items = ModelOptionalLiteral.values();
        for (ModelOptionalLiteral item : items) {
            if (item.toString().equalsIgnoreCase(value)) {
                return item;
            }
        }
        return null;
    }

    /**
     * {@inheritDoc}
     */
    @Override
    public String toString() {
        return this.value;
    }
}

Discussed with @tadelesh offline, TCGC does not handle the name for it currently. It is better if TCGC can handle it.
@archerzz @pshao25 Does .Net have this behavior? What will .Net generate for optional literal?

[iscai-msft]

i think this is something we can add. The root of the difference is java and .net want constants returned as single-value enums, while js and python will keep it as a single value constant.

We should have an input to tcgc that allows you to toggle this behavior. If an enum is returned in this case, we will automatically return a generated name ModelOptionalLiteral as well. To confirm, do .net and java want this enum to be fixed?

[haolingdong-msft]

Hey @iscai-msft , thanks for the confirmation, for Java, it is fixed enum.

[tadelesh]

@pshao25 could you help confirm .net behavior?

[pshao25]

We are extensible enum currently, but it is very subject to change. So we don't have any request for TCGC for now.

[tadelesh]

@haolingdong-msft i want to confirm java do this for all the constants or just model properties?

[haolingdong-msft]

This is for all optional literal cases. Is there any difference between generating name for model property and for all cases from TCGC side?

[iscai-msft]

the difference is right now, the constant type doesn't have a name property on it.

It makes sense to add this behavior in tcgc though, where we add .name and .isGenerated onto constants. This way languages (like .net and java) who would like to generate an enum / object for constants have a name to go off of

For now, I would say this isn't a blocker for tcgc adoption. You can keep your current code that comes up with a name for these constants (worst case you need to look into SdkConstantType.__raw to build the name for that constant).