3.2 KiB
| cairn | date | change |
|---|---|---|
| log | 2026-08-28 | pimdir-collection-id-is-the-mailbox |
A pimdir mailbox is its collection id
The pimdir backend named a mailbox by stripping a <namespace>/ prefix off its collection id, so imap/INBOX showed and resolved as INBOX, and both spellings addressed it. That mixed an addressing key with a display convenience over a convention the store cannot be asked about: collections.id is an opaque TEXT PRIMARY KEY, io-pimdir splits it nowhere, pimdir SPEC 9.2 says the store neither parses nor validates it, and hierarchy is modelled by the parent foreign key rather than by a separator. The / belongs to Neverest, whose own spec calls the namespace internal and non-configurable.
The guessing showed in the code: the derivation only worked because it filtered to message/rfc822 first, it needed pimdir.namespace for the store it could not decide, and it needed an ambiguity error for two collections stripping to one name. Himalaya already had the answer in another backend: JMAP mailbox ids are opaque server strings carried verbatim, and [mailbox.alias] is how a user avoids typing one.
What landed
The id is the mailbox. -m imap/INBOX is the spelling, and a listing shows imap/INBOX in both columns. pimdir.namespace is gone from PimdirConfig and from config.sample.toml, resolve_namespace and the client's namespace field with it, and mailbox_name is deleted.
hub_id kept the half that was doing work. It is now a membership check over the account's mail collections: a known id passes through, an unknown one is refused naming the ones the account holds. An id passed to the store unresolved reads as a mailbox that exists and is empty, which is what the check exists to prevent, and that was never the part the stripping provided.
Mailbox.name carries the collection row's own name rather than a derived one, which is JMAP's shape. Neverest seeds that column to the id today, so both columns read the same; a store that later carries a real display name shows it with no further change here.
The listing sorts by id rather than by the derived name, there being no longer two orders to choose between.
Capabilities moved
- backends: pimdir names a mailbox the way its server does is replaced by A pimdir mailbox is its collection id
Verification
Built clean and run against the live Neverest store at ~/.local/state/neverest/posteo, which holds three namespaces (imap, caldav, carddav) under one account. mailbox list renders all sixteen mail collections as imap/… in both columns; envelope list -m imap/INBOX returns envelopes; envelope list -m INBOX is refused naming the sixteen collections the account holds. Both delta scenarios, checked against real data rather than a fixture.
No unit test was added for hub_id: it is a membership check whose two outcomes are exactly the two scenarios above, and covering it would have meant building a store fixture this crate has no harness for to re-assert what the live run already showed.
Not done
Whether Neverest should write a short display name into collections.name is a question about the producer, left open. This change reads whatever is there, so either answer lands without touching Himalaya again.