Reason for not supporting float/double etc in attributes?

[wonbinkanglg]

So, I wanted to declare some custom attributes for containing scaling value, in double/float.
I noticed flatc will give error claiming unsupported type. Also noted on doc, only string and int types are supported.
Any idea why?

I am considering adding double support by parsing string containing real numbers with decimal point as double, eg) "0.001". Seeing how this is rather simple matter, was wondering if there was a strict reason to forbid floats for attributes.

[Advait251206]

That's an interesting question. Looking at it from a schema/compiler design perspective, my guess is that attributes were intentionally kept simple because they're primarily used as metadata for code generation rather than data serialization.

A few possible reasons for only supporting integers and strings:

• Deterministic representation: Strings and integers have unambiguous textual forms, while floating-point values can introduce questions around precision, formatting, exponent notation, locale differences, etc.
• Compiler simplicity: Attributes are parsed and consumed by various code generators and tools. Restricting them to a small set of types keeps the implementation straightforward across languages.
• Most metadata use cases don't require floating-point arithmetic: Tooling can already store values like "0.001" as strings and interpret them however it wants.
• Backward compatibility: Adding a new attribute type sounds simple, but it may affect parsers, reflection APIs, schema tooling, and third-party consumers that currently assume only string/int values.

That said, your use case seems reasonable. If attributes are being used to drive code generation and the value is conceptually numeric (such as a scale factor, unit conversion ratio, threshold, etc.), native float/double support would be more convenient and less error-prone than encoding the value as a string and parsing it later.

I'd be curious whether this limitation was a deliberate design decision from the beginning or simply a case where nobody has needed floating-point attributes strongly enough to justify expanding the schema grammar.