Version 0.0.5:
First release outside the experimental channel.
**QGIS 4**
* The plugin runs on QGIS 4 (Qt6) as well as QGIS 3.44 (Qt5).
* Ubuntu 24.04 (noble) with QGIS from apt is not supported: its system
numpy 1.x
conflicts with the numpy 2 that pandapower's dependencies require.
**Security**
* The extra parameters field of *Run power flow* no longer runs the
text as Python.
Only literal values (`algorithm='nr', max_iteration=10`) are accepted.
* The plugin no longer installs packages itself with pip. Dependencies
are installed
by the qpip plugin from `requirements.txt`.
**Fixes**
* Export skips multi-part geometries and selects them, instead of
stopping with an
assertion error.
* Errors that were silently ignored are now written to the
*Pandapower* tab of the
message log.
* The plugin package no longer contains developer scripts (`.bat`,
`.sh`).
**Requires** pandapower 3.5.6 or newer. numpy and scipy are no longer
pinned.

Version 0.0.4:
pandapower networks are a data source, not an import.
**Networks are now opened, not imported.**
* pandapower appears in the **Data Source Manager** alongside
PostgreSQL and SAP HANA.
Pick a network file, select the tables you want, press Add.
* Every pandapower table can be opened, not just `bus` and `line`.
Tables without
geometry (`trafo`, `load`, `switch`, …) open as plain attribute
tables.
* `res_*` tables can be opened in their own right, while their columns
stay merged
into the `bus`/`line` layers for styling.
* Only tables that contain rows are listed, so the dialog stays
readable.
**Known limitation**
* Expanding a pandapower `.json` directly in the **Browser** panel is
not available.
Returning a Python subclass of `QgsDataCollectionItem` from
`QgsDataItemProvider.createDataItem()` is unsafe on QGIS 3.44 — the
object loses
its Python half once QGIS stores it, and expanding the node crashes
QGIS. This
reproduces without this plugin, so it is a binding limitation rather
than
something the plugin can work around. Use the Data Source Manager
instead; the
code is kept behind `ENABLE_BROWSER_TREE` in `pandapower_qgis.py`.
**Edits are written on commit.**
* Changes are written to the `.json` when the layer edits are saved,
not on every
individual change. Rolling back writes nothing.
* All layers of one network share a single in-memory network, so one
save writes
them all and an edit in one layer is immediately visible in the
others.
* A file that changed on disk is no longer silently overwritten.
**Removed**
* The *import from pandapower* action, together with
`ppqgis_import.py` and its
dialog. The Browser and Data Source Manager replace it. Note the old
import
reprojected the network's coordinates in place; opening a network now
leaves the
stored coordinates untouched and lets QGIS reproject for display.
* The vendored `geo.py`, superseded by `pandapower.plotting.geo`
upstream.
* The asynchronous save machinery, and with it the "previous save
operation is
still running" class of errors.
*Export is unchanged* — it still builds a new network from arbitrary
QGIS layers.
**Fixes**
* Plugin no longer fails to load on a profile where the locale setting
is unset.
* Geometry edits no longer wrote into a pandas copy that was then
discarded.
* Renderer instances are no longer shared between layers (a
double-ownership crash).
* Closing a layer no longer deregisters the provider for every other
open layer.
* `pressure_level` in a layer URI was written but never read, so pipe
layers lost
their level filter.
**Requires** pandapower 3.5 or newer, which stores geodata in the
`geo` column.

Version 0.0.3:
* Feature/dataprovider was included, which allows to use pandapower
networks as a dataprovider.
* Large restructuring

None