Is there a reason to prohibit empty (quoted) record or union labels? I ran into this in Dhall for Java while writing some property-based round-trip tests where I had arbitrary JSON values that included objects with empty field names. I can easily filter these out, but I tried changing + to * in the equivalent of quoted-label there, and allowing empty labels doesn’t seem to break anything. Am I missing something?
@travisbrown: There’s no particular reason why they are forbidden. We just never thought to permit them
And what about the backtick itself? It’s probably not widely used, but we should allow backticks too as JSON and YAML do. Maybe add an escaped
\`
in quoted-label-char ?
I’m not sure parity with JSON or YAML field names should be the target here. The spec already explicitly prohibits non-printable-ASCII Unicode characters in labels, “mainly to minimize the potential for code obfuscation”, which seems like the right choice to me.
I guess you could argue that quoted labels should support all printable ASCII characters, and that it’s odd to exclude only the backtick, but I personally think the added complexity of supporting escapes outweighs that consideration (especially since \ is a valid character in quoted labels). Supporting empty quoted labels seems comparatively easy for implementations.
(FWIW I’m more than happy to implement escaped backticks in Dhall for Java if there’s agreement that the spec should include them, and I do agree that there’s an argument on the basis of consistency. I just don’t think I’m convinced that the change is needed.)