Problem

A key name is made out of a sequence of key part names, and can be constructed with keyAddBaseName/keySetBaseName. Both applications and configuration file formats might need arbitrary strings to be encoded within a key name part.

For example:

an application uses names of internal components as sections within the configuration.
a parser reads an empty string, to be encoded as base name.

Constraints

keySetBaseName (key, keyBaseName (key)) should be a NOP, which is needed for round-trips: If a storage plugin serializes what it gets with keyBaseName; keySetBaseName must lead to the same key.
support any configuration file format (i.e., any valid file of some format can be transformed to a KeySet, e.g. TOML supports empty key part names).

Assumptions

Considered Alternatives

restrict what keyAddBaseName/keySetBaseName can accept: has the downside that applications would suddenly fail when trying to set some key base names
have additional keySetBaseName* functions that make strings safe to be accepted in keyAddBaseName/keySetBaseName: seems to be a too big change in the storage plugins

Decision

keyAddBaseName/keySetBaseName never fail with any argument, so any character sequence can be escaped except of NULL bytes. The argument goes unmodified to the unescaped key name.

For arrays there is no escaping needed because an array is only an array if the metadata array is appended to the direct parent key. See array.

Rationale

hard to use it wrong API: having only the functions keyAddBaseName/keySetBaseName, without any size argument
applications and storage plugins can pass any c-string to keyAddBaseName/keySetBaseName without any further consideration

Implications

Related Decisions

Array