Store raw tokens instead of dictionaries in metadata

[link2xt]

Follow-up to #239 before we deploy it to nine

[hpk42]

If we want to evolve per-user settings (eg when to expire/remove messages) then we will need more than just a single setting. Also, having a dictionary quite directly maps imap key-value metadata so i think the simplification here is not that useful and requires more work if we add more keys.

…

On Sat, Mar 23, 2024 at 19:05 -0700, link2xt wrote: You can view, comment on, or merge this pull request online at: #243 -- Commit Summary -- * Store raw tokens instead of dictionaries in metadata -- File Changes -- M chatmaild/src/chatmaild/metadata.py (33) -- Patch Links -- https://github.com/deltachat/chatmail/pull/243.patch https://github.com/deltachat/chatmail/pull/243.diff -- Reply to this email directly or view it on GitHub: #243 You are receiving this because your review was requested. Message ID: ***@***.***>

[link2xt]

marshal will not work for storing multiple settings unless we also use lockfiles for each file. SETMETADATA requests can be sent concurrently, in which case we will overwrite the whole dictionary. We should use dbm then or create a folder for guid and store files like token, setting1, setting2 separately rather than trying to put everything into a single file.

[hpk42]

On Sun, Mar 24, 2024 at 11:25 -0700, link2xt wrote: `SETMETADATA` requests can be sent concurrently, in which case we will overwrite the whole dictionary.

Wasn't there something with dovecot that serializes all metadata-operations or do i misremember? If indeed we have concurrency then the current way is insufficient and i think going for guid-dir and store files like `token` etc. is the easiest as it also makes accessing it from other processes trivial. I btw tried to find a way to use the vmail-dir of a user but couldn't easily find one.

[hpk42]

meanwhile @link2xt and me discussed to store metadata-settings into a directory structure with each setting being a separate file -- preferably living inside the per-user maildir

[link2xt]

Finding per-user maildir from guid is not easy, solution at https://dovecot.org/list/dovecot/2018-May/111694.html is hacky. Using guid as a folder name is good enough for now.

[hpk42]

it seems that both SETMETADATA and GETMETADATA transmit the e-mail address to our dict-proxy,
in the begin-transaction and in the lookup, respectively.

B13\tlkj123poi@c2.testrun.org
S13\tpriv/43f5f508a7ea0366dff30200c15250e3/devicetoken\tl1kj23lk123
C13

Lpriv/43f5f508a7ea0366dff30200c15250e3/devicetoken\tlkj123poi@c2.testrun.org

so we probably don't need any hacks.

[hpk42]

#254 is merged, closing this