Remove responsibility for credentials mutex locking from external
callers and handle appropriate locks internally. This allows the
mutex lock to be much "tighter" and avoids deadlocks when
credentials are requested while an existing credentials dialog
is being shown.
(No mutex is required protecting the credentials dialog itself
as this is ALWAYS shown in the main thread)
Fixes#20826
With recent changes in QgsNetworkAccessManager this should be
safe to resurrect (fingers crossed!). Also simplify code a lot,
because now QgsNetworkAccessManager handles waking the worker
thread after the auth request is handled.
Add a lot more tests
races and crashes when auth requests occur in non-main thread
Like the recent SSL error handling, this commit abstracts out
the network authentication handling into a thread safe approach.
Brings back QGIS 2.18's ability to directly convert a Processing model
to an equivalent Processing Python script algorithm, correctly
updated and working in the 3.x API.
Available from the model dialog, and from the right-click context
menu on an existing model.
Sponsored by Solspec
- fix parameter can get converted to plain number parameter after edits
- allow parameter to be linked to parent parameters, so that the
correct distance unit and choices are shown for the parameter
This setting, which is disabled by default and placed into an
"advanced" group on PDF/SFG export, disables the built-in
raster layer tiled rendering. While the tiling is good for
memory usage, it can cause visible "seams" in the rasters
for generated PDF/SVG files.
The setting has a tooltip warning users that disabling the
tiling results in high memory usage during exports.
Fixes#19500
of main algorithm
Using code like:
buffered_layer = processing.run(..., context, feedback)['OUTPUT']
...
return {'OUTPUT': buffered_layer}
can cause issues if done as a sub-step of a larger processing algorithm. This
is because ownership of the generated layer is transferred to the caller
(Python) by processing.run. When the algorithm returns, Processing
attempts to move ownership of the layer from the context to the caller,
resulting in a crash.
(This is by design, because processing.run has been optimised for the
most common use case, which is one-off execution of algorithms as part
of a script, not as part of another processing algorithm. Accordingly
by design it returns layers and ownership to the caller, making things
easier for callers as they do not then have to resolve the layer reference
from the context object and handle ownership themselves)
This commit adds a new "is_child_algorithm" argument to processing.run.
For algorithms which are executed as sub-steps of a larger algorithm
is_child_algorithm should be set to True to avoid any ownership issues
with layers. E.g.
buffered_layer = processing.run(..., context, feedback, is_child_algorithm=True)['OUTPUT']
...
return {'OUTPUT': buffered_layer}
For algorithms with multiple parameters depending on a vector layer parameter, the code that loads the layer in the background is called repeatedly, impacting performance. A small layer cache is implemented with these changes, so the dialog only tries to load the layer once.
