Normally the backend is made to have a singleton instance, and handle all profiles.
However, it may be useful to run instances independently (e.g., for development, to run
components in independent processes, for testing, etc).
This was previously possible by using a specific configuration files at backend's root.
This patch introduce a cleaner way by using an isolation mode.
Isolation mode is activated by settings the `LIBERVIA_ISOLATION` environment variable to
the name of the isolated instance. `LIBERVIA_ISOLATION_MODE` can be used to specify one of
the following mode:
- normal (default): config and shared files are in XDG in a `isolated/<name>` subdirectory
(usually `~/.local/share/libervia/isolated/<name>` and
`~/.config/libervia/isolated/<name>`). Runtime files and other instance specific data
(such as D-Bus name) are also isolated. Config uses cascade as usual.
- full: same as above, but config isn't cascading, it's using only its own isolated config
file.
- tmp: same as normal but the local directory is put in a temporary directory, and deleted
on backend startup. It stays on termination, so data can be checked if necessary (this mode is
usually used for development).
This feature will notably be useful to test severless mode on a single machine.
rel 479
- work around a deprecated behaviour in `asyncio.get_event_loop` until the fix is
integrated in a released Twisted.
- update dependencies to have versions compatible with Python 3.14
- disable `term-image` which is bounded to incompatible Pillow version. It is unmaintained
for 1 year, and will probably be removed soon in favor of `textual-image`.
The reactor is now set in the launcher, as it is the first script to be
loaded. This should avoid race conditions when some Twisted or libervia
packages are imported and end up importing `twisted.internet.reactor`,
which installs the default reactor before the AsyncIO one. The AsyncIO
reactor is still set in `twisted.plugin.libervia_backend_plugin` (but it
no longer fails if already set), in case the backend is started as a
plugin (although this is not recommended anyway)
The backend was not properly exiting in case of early exception (e.g. configuration
error). It seems to be related to the use of the `@aio` decorator, but couldn't find the
exact issue with loop. The use of `os._exit` is a workaround for now, until the fix is
inspected more deeply and properly fixed.
This is a huge configuration refactoring.
The configuration system now uses Pydantic settings, with better validation, better
support for environment variables and other goodies.
The main config files now use TOML instead of INI, this allow for better handling of
complex types, notably dict.
Plugins, components and frontends can declare their config settings in a class which will
be imported dynamically.
default location of pid file is now specific to logged user, this allow to run several
instances of Libervia by different users on the same machine without PID conflicts.
if `wait_for_service` is used in `[DEFAULT]`, the service at the specified host/port is
waited for before starting the backend. This is useful in the situation where we need to
be sure that something is started before running, e.g. in a container.
- updated installation to use the now standard `pyproject.toml` instead of legacy
`setup.py`. `setup.py` and other legacy files have been deleted.
- removed outdated README4PACKAGERS
- removed pylintrc which has not been correctly reviewed. Linter conf should go to
`pyproject.toml` now.
- [hatch](https://hatch.pypa.io) is now used as main building tool. However, thanks to the
use of standards, other tools can be used too.
- update .hgignore
- several dependencies version bump, with code update to adapt to changes.
