… al) data sources

rows inserted/added was fired twice (once by the manager and once by the
model), confusing the proxy model.

Also count was wrong by 1 in beginInsertRows( parent, oldCount, oldCount + count )
and beginRemoveRows, but now these two calls have been removed completely.

Fix #56493

This behaves much nicer when row heights grow large, eg when
there's multiline content in a field or when json fields are
present

…ltValue

Previously we mixed invalid QVariant values with null QVariants,
resulting in a mix of results between Qt 5 and Qt 6 and across
different data providers

…or group layers

…ested (typically from QgsGeometryCollection::isValid())

…n initiated against editing-enabled vector layer

Avoid various errors in case of empty input layer, empty/null geometries, invalid geometries, linestrings with zero-length segments...

If we only have receieved the first part of a message, don't
emit stateChanged until we've received the rest.

Possibly refs #56460

See discussion from
#55943 (comment)

…ecks in a foolproof way

The check for ``width * depth + 31`` not overflowing INT_MAX was done
after computing it, which would be undefined behavior if that happened.
Nowadays compilers can use "impossible situations" in smart ways, and
could very well discard the check if it is done afterwards. Be on the
safe sife and check before, as done in
https://code.qt.io/cgit/qt/qtbase.git/tree/src/gui/image/qimage_p.h#n89

These are cheap to construct by design, so we can avoid
storing them and the messy cross-class access to member
variables

This prevents a bug where the save action in the Python console
keeps defaulting back to the binary QGIS install folder

- Add some typehints
- Avoid creating an unnecessary local variable

This fixes UNC paths not identified as network paths.

By ensuring that a trailing slash for directories is passed
down to the identify function.

…found (fix #54058)

This is messy, as there's no way we can possibly make the current,
stable API of this class safe. We have to resort to an opt-in
"safe" mode which exposes a non-dangerous API.

This should hopefully fix issues where the qt event loop causes
destruction of the detected connection before listener slot is
called and is able to take ownership of the signal argument...

Use a single timer instead of the QTimer static method, so that
we can stop the timeout as soon as we receive valid data and don't
risk a race where unwanted timeouts are added to the event loop
while we're processing a valid detected connection

Otherwise we end up calling QgsGpsDetector::detected multiple times
for the same device, and if an external client has taken the
detected connection in the meantime then the detector will incorrectly
move to the next device and continue trying to make connections

This ensures that there's NO change of any remaining unwanted
interaction between the GPS detector and the connection which
it no longer owns after a call to takeConnection. Otherwise
when the detector outlives the "find a connection" logic we risk
status updates from the taken GPS device from triggering unwanted
detect-a-valid-connection logic from the detector.

Use the processing context's ellipsoid instead of a hardcoded
WGS84 ellipsoid for distance calculations during network
analysis, so that the lengths used will exactly match other
measurement tools used on the same features in the same
project.

A WFS request such as
`SERVICE=WFS&REQUEST=GetFeature&VERSION=1.1.0&SRSNAME=EPSG:4326` does
not invert the axis and return the coordinates in the LON/LAT
order. For example:

<gml:Envelope srsName="EPSG:4326">
  <gml:lowerCorner>2.358 48.865</gml:lowerCorner>
  <gml:upperCorner>2.37 48.876</gml:upperCorner>
</gml:Envelope>

However, the same request without a SRSNAME parameter inverts the axis
and returns the the coordinates in the LAT/LON order:

<gml:Envelope srsName="EPSG:4326">
  <gml:lowerCorner>48.865 2.358</gml:lowerCorner>
  <gml:upperCorner>48.876 2.37</gml:upperCorner>
</gml:Envelope>

With this change, the axis is not inverted if the SRSNAME parameter is
not passed. This way, this is the same behavior as
`SRSNAME=EPSG:4326`.

This is a follow-up of #45270.

Transform the "Opacity" value calculated by Data Defined Override expression from [0-100] range to [0-1] range.

This is an attempt to fix browser slowness when network
drives becomes unavailable.

Some QDir/QFileInfo call may become slow when called on
disconneced drives on windows, by calling them in the
init method (which is called by a separate thread and when
the driver is more likely to be available) the risk
to call them when they can become slow is drastically
reduced.

Whether this will help to reduce the browser slowness
on network drives it is hard to tell but it won't certainly
hurt.

Fix #56652