for a request. Most requests should reserve more
than a single connection (default: 3) and only
requests that are executed as nested expressions
should specify 1.
We never call this method using nullptrs, so there's no need
for this to be a pointer argument in the first place. And
having it a pointer encourages leaky code, such as the
leak this commit fixes in server.
The original commit speed up some layers, but regressed
performance for others. It was decided to revert the
dynamic queue feature, but it seems this was never actually
done.
Fixes#16239, #19203
This forces Python code and plugins to become datum transform
aware, and given that upgrading python code is easy (just
add QgsProject.instance() as a new argument to the constructor)
it's relatively painless to force this on PyQGIS users.
Also fix upgrade the easy QgsCoordinateTransform c++ constructors
where the project is available, or where using QgsProject::instance()
is safe to do.
For others, just avoid the deprecated warnings until we can
get access to the correct project instance where the transform
is being constructed.
Considering this statement:
layer.dataProvider().getFeatures(QgsFeatureRequest().setFilterFids([]))
this resulted in an iteration over the whole postgres source.
Internally, this happened implicitly when iterating over a layer with
an active edit buffer but no edited features. In such a scenario, we are
now much faster.
If set, all geometries will be reprojected from their original
coordinate reference system to the destination CRS while
iterating over features.
If a CRS has been set as the destination CRS, then the filterRect
parameter should be specified in the same CRS as this destination
CRS.
Additionally, a callback function can be specified on the request
to be called if a transform exception is encountered while
iterating over features.
This is designed to make it easier for scripts and plugins to
correctly reproject layers in an efficient and robust way, instead
of having to implement lots of repeated code themselves and
potentially missing some of the important considerations which
come with reprojecting geometries & bounding boxes.
Now, if a script wants the features from a layer in a specific
CRS, they can call:
crs = QgsCoordinateReferenceSystem('epsg:4326')
request = QgsFeatureRequest().setDestinationCrs(crs)
for f in layer.getFeatures(reqeuest):
print('geometry in 4326 is {}.format(f.geometry().exportToWkt()))
Lower the default queue size, but automatically adjust it
based on how long each fetch takes. This change keeps fetching
responsive even for slow connections or databases. The current
approach with a fixed queue size can result in very slow feature
fetching, which prevents UI updates for multiple seconds.
Switching from QVariant to QVariantList solves the underlying Qt issue:
- comparison of QVariantList objects works fine
- comparison of QVariantList objects wrapped in QVariant does not work
The extra wrapping of QVariantList into another QVariant seems unnecessary anyway,
so we may as well save a tiny bit of memory and cpu
* Remove deprecated Qgis::WKBType and API cleanup
Renames QgsWKBTypes to QgsWkbTypes
Replaces usage of the enums:
* Qgis::WKBType with QgsWkbTypes::Type
* Qgis::GeometryType with QgsWkbTypes::GeometryType
Their values should be forward compatible (a fact that was already
explited up to now by casting between the types)
Renames some SSLxxx to SslXxx and URIxxx to UriXxx
* Fix build warnings and simplify type handling
* Add a fixer to rewrite imports
* The forgotten rebase conflictThe forgotten rebase conflicts
* QgsDataSourcURI > QgsDataSourceUri
* QgsWKBTypes > QgsWkbTypes
* Qgis.WKBGeom > QgsWkbTypes.Geom
* Further python fixes
* Guess what... Qgis::wkbDimensions != QgsWkbTypes::wkbDimensions
* Fix tests
* Python 3 updates
* [travis] pull request caching cannot be disabled
so at least use it in r/w mode
* Fix python3 print in plugins
All pointer based methods have been removed.
Now we have only:
void setGeometry( const QgsGeometry& geom )
and
QgsGeometry geometry() const
Benefits include avoiding a whole lot of tricky pointer lifetime
issues, potential memory leaks, and finally closing #777, which
has survived for over 9 years!...
Impacts on PyQGIS code:
- no more need for the messy
g = QgsGeometry( feature.geometry() )
workaround, just use g = feature.geometry() instead
- IMPORTANT: you can no longer test whether a feature has geometry
using `if f.geometry():`, since QgsFeature::geometry() will
*always* return an object. Instead, use
`if not f.geometry().isEmpty():`, or preferably the new method
`if not f.hasGeometry():`
Fix#777
Fixes again editing of features with negative identifiers.
Now a test fails, but it is the test being broken this time
(testSignedIdentifiers). The test was added as part of
2bd7f446b4dd368968f23b990262c5e8a5a83f80, which fixed a crash
when using signed identifiers.
Also rename pktInt64 to pktUint64, in case we'll be able to know
in advance the database contains only positive integers (not
implemented in this commit).
Can currently be triggered by using the field calculator to update a selection.
While an iterator is active and the connection is locked, an update statement
waits unsuccessfully for the same (locked) connection.
This commit only locks the connection while an iterator is actually fetching
data and not for its entire lifetime.
subsets of attributes and the filter expression requires fields
which are not included in the attribute subset
Note: I've only fixed the providers which implement the vector
provider conformance unit tests. Other providers (eg GRASS) should
also implement a similar fix.
