SQLite specific behaviour with autoincrement (https://sqlite.org/autoinc.html) is actually
already handled by SQLAlchemy, and unsetting autoincrement cause troubles when storing new
objects.
a first Alembic revision file is joined to migrate existing databases to new mapping.
This patch:
- set `nullable=False` in various places
- drop legacy `message_types` table, and use `Enum` instead
- use naming convention, and get rid of anonymous constraint (this is the main reason
why the revision file is so long)
Alembic database migration tool, which is the recommended one for SQLAlchemy has been
integrated. When a database is created, it will be used to stamp to current (head)
revision, otherwise, DB will be checked to see if it needs to be updated, and upgrade will
be triggered if necessary.
Some f-strings where used in translatable text, this has been fixed by using explicit
`format()` call (using a script based on `tokenize`).
As tokenize messes with spaces, a reformating tool (`black`) has been applied to some
files afterwards.
- fixes coroutines handling in various places
- fixes types which are not serialised by Tx DBus
- XEP-0384: call storage methods in main thread in XEP: Python OMEMO's Promise use thread
which prevent the use of AsyncIO loop. To work around that, callLater is used to launch
storage method in main thread. This is a temporary workaround, as Python OMEMO should
get rid of Promise implementation and threads soon.
this method is used by `getFiles` and `setFile` and fill `owner` when it's missing to
avoid getting accidentally files from other users.
`owner` is not filled in some case (no client, `file_id` or `parent` specified) and an
exception is raised if a component has neither `owner` nor `peer_jid` specified.
using public permissions by default makes permissions change more easy as we don't have to
recursively change all permissions of sub files when changing directory. Root directories
are private by default.
this new schema brings the new `public_id` column, which will be used to serve files via
HTTP, splits `mime_type` in 2 (`media_type` and `media_subtype`) to make indexing/media
filtering easier, and fixes indexes for `files` table.
/!\ huge commit
Avatar logic has been reworked around the IDENTITY plugin: plugins able to handle avatar
or other identity related metadata (like nicknames) register to IDENTITY plugin in the
same way as for other features like download/upload. Once registered, IDENTITY plugin will
call them when suitable in order of priority, and handle caching.
Methods to manage those metadata from frontend now use serialised data.
For now `avatar` and `nicknames` are handled:
- `avatar` is now a dict with `path` + metadata like `media_type`, instead of just a string
path
- `nicknames` is now a list of nicknames in order of priority. This list is never empty,
and `nicknames[0]` should be the preferred nickname to use by frontends in most cases.
In addition to contact specified nicknames, user set nickname (the one set in roster) is
used in priority when available.
Among the side changes done with this commit, there are:
- a new `contactGet` bridge method to get roster metadata for a single contact
- SatPresenceProtocol.send returns a Deferred to check when it has actually been sent
- memory's methods to handle entities data now use `client` as first argument
- metadata filter can be specified with `getIdentity`
- `getAvatar` and `setAvatar` are now part of the IDENTITY plugin instead of XEP-0054 (and
there signature has changed)
- `isRoom` and `getBareOrFull` are now part of XEP-0045 plugin
- jp avatar/get command uses `xdg-open` first when available for `--show` flag
- `--no-cache` has been added to jp avatar/get and identity/get
- jp identity/set has been simplified, explicit options (`--nickname` only for now) are
used instead of `--field`. `--field` may come back in the future if necessary for extra
data.
- QuickContactList `SetContact` now handle None as a value, and doesn't use it to delete the
metadata anymore
- improved cache handling for `metadata` and `nicknames` in quick frontend
- new `default` argument in QuickContactList `getCache`
- `items` was returning the dict with all values, it now returns a Deferred with actual dict items()
- `all` new method does return the dict with all values (in a Deferred)
- `adel` has been added to LazyPersistentBinaryDict
typeCheck() is not automatically called on reception by Wokkel, as a result extensions
fields values may be strings instead of the field type. TypeCheck is now explicitly
called in memory.disco to avoid that. It is not called immediately on reception as the
string value is needed to calculate the capability hash
