Grammar for Secret Storage key ID

[AndrewFerr]

Link to problem area: https://spec.matrix.org/v1.18/client-server-api/#storage

Issue
It's not explicitly stated whether the key ID used to identify Secret Storage keys may be an empty string.

At least in the JS-SDK, an empty key ID is treated as an undefined key (source), which is reasonable. Nonetheless, the spec may as well take a stance on it, as technically speaking, there is nothing that prevents an empty string from being a usable key ID.

The Secret Storage key ID is used in the following places:

• The m.secret_storage.key.[key ID] account data type
• The value of the key parameter in the m.secret_storage.default_key account data object
• The keys of the encrypted map in encrypted account data objects

[richvdh]

I agree, the spec should take a stance. I think it would be very odd if we accepted the empty string as a key ID, since it would mean storing the key itself in an account data type called m.secret_storage.key..

To be honest though, it's not just the minimum length that the spec should concern itself with: it should really specify a maximum length, and grammar. The maximum length is somewhat constrained by the fact that m.secret_storage.key.[key ID] needs to fit into the type of an account data entry (although that in itself is not constrained currently (#2356), we should probably avoid going over 255 bytes).

The opaque identifier grammar, limited to (say) 128 characters, would seem appropriate. It'd need an MSC to change it, though.