Improved PySide2 support: filling QPolygonF objects directly in memory from NumPy arrays

Author: PierreRaybaut

CHANGELOG.md

@@ -1,5 +1,9 @@
 # PythonQwt Releases #
 
+### Version 0.8.1 ###
+
+- PySide2 support was significatively improved betwen PythonQwt V0.8.0 and 
+  V0.8.1 thanks to the new `qwt.qwt_curve.array2d_to_qpolygonf` function.
 
 ### Version 0.8.0 ###
 

README.md

@@ -87,10 +87,53 @@ for more details on API limitations when comparing to Qwt.
 
 ### Why PySide2 support is still experimental ###
 
-There is still a significant performance issue with PySide2 (drawing polylines with PySide2 is apparently approx. 60 times slower than with PyQt5, see [this bug report](https://bugreports.qt.io/browse/PYSIDE-1366)).
+<img src="https://raw.githubusercontent.com/PierreRaybaut/PythonQwt/master/doc/images/pyqt5_vs_pyside2.png">
 
-As a consequence, until an equivalent feature is implemented in PySide2, we strongly 
-recommend using PyQt5 instead of PySide2.
+Try running the `curvebenchmark1.py` test with PyQt5 and PySide: you will notice a 
+huge performance issue with PySide2 (see screenshot above). This is due to the fact 
+that `QPainter.drawPolyline` is much more efficient in PyQt5 than it is in PySide2 
+(see [this bug report](https://bugreports.qt.io/projects/PYSIDE/issues/PYSIDE-1366)).
+
+As a consequence, until this bug is fixed in PySide2, we still recommend using PyQt5 
+instead of PySide2 when it comes to representing huge data sets.
+
+However, PySide2 support was significatively improved betwen PythonQwt V0.8.0 and 
+V0.8.1 thanks to the new `array2d_to_qpolygonf` function (see the part related to 
+PySide2 in the code below).
+
+```python
+def array2d_to_qpolygonf(xdata, ydata):
+    """
+    Utility function to convert two 1D-NumPy arrays representing curve data 
+    (X-axis, Y-axis data) into a single polyline (QtGui.PolygonF object). 
+    This feature is compatible with PyQt4, PyQt5 and PySide2 (requires QtPy).
+    
+    License/copyright: MIT License © Pierre Raybaut 2020.
+    
+    :param numpy.ndarray xdata: 1D-NumPy array (numpy.float64)
+    :param numpy.ndarray ydata: 1D-NumPy array (numpy.float64)
+    :return: Polyline
+    :rtype: QtGui.QPolygonF
+    """
+    dtype = np.float
+    if not (
+        xdata.size == ydata.size == xdata.shape[0] == ydata.shape[0]
+        and xdata.dtype == ydata.dtype == dtype
+    ):
+        raise ValueError("Arguments must be 1D, float64 NumPy arrays with same size")
+    size = xdata.size
+    polyline = QPolygonF(size)
+    if PYSIDE2:  # PySide2 (obviously...)
+        address = shiboken2.getCppPointer(polyline.data())[0]
+        buffer = (ctypes.c_double * 2 * size).from_address(address)
+    else:  # PyQt4, PyQt5
+        buffer = polyline.data()
+        buffer.setsize(2 * size * np.finfo(dtype).dtype.itemsize)
+    memory = np.frombuffer(buffer, dtype)
+    memory[: (size - 1) * 2 + 1 : 2] = xdata
+    memory[1 : (size - 1) * 2 + 2 : 2] = ydata
+    return polyline
+```
 
 ## Installation
 

doc/images/pyqt5_vs_pyside2.png

@@ -25,14 +25,17 @@ Why PySide2 support is still experimental
 
 Try running the `curvebenchmark1.py` test with PyQt5 and PySide: you will notice a 
 huge performance issue with PySide2 (see screenshot above). This is due to the fact 
-that PyQt5 (and PyQt4) allows an efficient way of filling a QPolygonF object from a 
-Numpy array, and PySide2 is not (see code below).
+that `QPainter.drawPolyline` is much more efficient in PyQt5 than it is in PySide2 
+(see `this bug report <https://bugreports.qt.io/projects/PYSIDE/issues/PYSIDE-1366>`_).
 
-.. literalinclude:: /../qwt/plot_curve.py
-   :pyobject: series_to_polyline
+As a consequence, until this bug is fixed in PySide2, we still recommend using PyQt5 
+instead of PySide2 when it comes to representing huge data sets.
+
+However, PySide2 support was significatively improved betwen PythonQwt V0.8.0 and 
+V0.8.1 thanks to the new `array2d_to_qpolygonf` function (see code below).
 
-As a consequence, until an equivalent feature is implemented in PySide2, we strongly 
-recommend using PyQt5 instead of PySide2.
+.. literalinclude:: /../qwt/plot_curve.py
+   :pyobject: array2d_to_qpolygonf
 
 Help and support
 ----------------

doc/installation.rst

@@ -1,3 +1,4 @@
 numpy==1.19.1
 PyQt5==5.15.0
-PyQt5-sip==12.8.0
+PyQt5-sip==12.8.0
+QtPy==1.9.0

doc/requirements.txt

@@ -28,7 +28,7 @@
 .. _GitHubPage: http://pierreraybaut.github.io/PythonQwt
 .. _GitHub: https://github.com/PierreRaybaut/PythonQwt
 """
-__version__ = "0.8.0"
+__version__ = "0.8.1"
 QWT_VERSION_STR = "6.1.5"
 
 import warnings

qwt/__init__.py

@@ -31,6 +31,10 @@
 from qtpy.QtGui import QPen, QBrush, QPainter, QPolygonF, QColor
 from qtpy.QtCore import QSize, Qt, QRectF, QPointF
 
+if PYSIDE2:
+    import shiboken2
+    import ctypes
+
 import numpy as np
 
 
@@ -56,26 +60,46 @@ def qwtVerifyRange(size, i1, i2):
     return i2 - i1 + 1
 
 
+def array2d_to_qpolygonf(xdata, ydata):
+    """
+    Utility function to convert two 1D-NumPy arrays representing curve data 
+    (X-axis, Y-axis data) into a single polyline (QtGui.PolygonF object). 
+    This feature is compatible with PyQt4, PyQt5 and PySide2 (requires QtPy).
+    
+    License/copyright: MIT License © Pierre Raybaut 2020.
+    
+    :param numpy.ndarray xdata: 1D-NumPy array (numpy.float64)
+    :param numpy.ndarray ydata: 1D-NumPy array (numpy.float64)
+    :return: Polyline
+    :rtype: QtGui.QPolygonF
+    """
+    dtype = np.float
+    if not (
+        xdata.size == ydata.size == xdata.shape[0] == ydata.shape[0]
+        and xdata.dtype == ydata.dtype == dtype
+    ):
+        raise ValueError("Arguments must be 1D, float64 NumPy arrays with same size")
+    size = xdata.size
+    polyline = QPolygonF(size)
+    if PYSIDE2:  # PySide2 (obviously...)
+        address = shiboken2.getCppPointer(polyline.data())[0]
+        buffer = (ctypes.c_double * 2 * size).from_address(address)
+    else:  # PyQt4, PyQt5
+        buffer = polyline.data()
+        buffer.setsize(2 * size * np.finfo(dtype).dtype.itemsize)
+    memory = np.frombuffer(buffer, dtype)
+    memory[: (size - 1) * 2 + 1 : 2] = xdata
+    memory[1 : (size - 1) * 2 + 2 : 2] = ydata
+    return polyline
+
+
 def series_to_polyline(xMap, yMap, series, from_, to):
     """
     Convert series data to QPolygon(F) polyline
     """
-    xData = xMap.transform(series.xData()[from_ : to + 1])
-    yData = yMap.transform(series.yData()[from_ : to + 1])
-    size = to - from_ + 1
-    if PYSIDE2:
-        polyline = QPolygonF()
-        for index in range(size):
-            polyline.append(QPointF(xData[index], yData[index]))
-    else:
-        polyline = QPolygonF(size)
-        pointer = polyline.data()
-        dtype, tinfo = np.float, np.finfo  # integers: = np.int, np.iinfo
-        pointer.setsize(2 * polyline.size() * tinfo(dtype).dtype.itemsize)
-        memory = np.frombuffer(pointer, dtype)
-        memory[: (to - from_) * 2 + 1 : 2] = xData
-        memory[1 : (to - from_) * 2 + 2 : 2] = yData
-    return polyline
+    xdata = xMap.transform(series.xData()[from_ : to + 1])
+    ydata = yMap.transform(series.yData()[from_ : to + 1])
+    return array2d_to_qpolygonf(xdata, ydata)
 
 
 class QwtPlotCurve_PrivateData(QwtPlotItem_PrivateData):

qwt/plot_curve.py

@@ -33,9 +33,8 @@
 
 The ``PythonQwt`` package is a 2D-data plotting library using Qt graphical 
 user interfaces for the Python programming language. It is compatible with 
-both ``PyQt4`` and ``PyQt5`` (``PySide`` is currently not supported but it
-could be in the near future as it would "only" requires testing to support 
-it as a stable alternative to PyQt).
+both ``PyQt4``, ``PyQt5`` and ``PySide2`` (see documentation for more information 
+on a performance issue due to PySide2 itself when plotting huge data sets).
 
 The ``PythonQwt`` project was initiated to solve -at least temporarily- the 
 obsolescence issue of `PyQwt` (the Python-Qwt C++ bindings library) which is 
@@ -49,6 +48,22 @@
 higher level features to the `guiqwt` library.
 
 See `README`_ and documentation (`online`_ or `PDF`_) for more details on the library and `changelog`_ for recent history of changes.
+    
+The following example is a good starting point to see how to set up a simple plot widget::
+
+    import qwt
+    import numpy as np
+
+    app = qtpy.QtGui.QApplication([])
+    x = np.linspace(-10, 10, 500)
+    plot = qwt.QwtPlot("Trigonometric functions")
+    plot.insertLegend(qwt.QwtLegend(), qwt.QwtPlot.BottomLegend)
+    qwt.QwtPlotCurve.make(x, np.cos(x), "Cosinus", plot, linecolor="red", antialiased=True)
+    qwt.QwtPlotCurve.make(x, np.sin(x), "Sinus", plot, linecolor="blue", antialiased=True)
+    plot.resize(600, 300)
+    plot.show()
+    
+.. image:: https://raw.githubusercontent.com/PierreRaybaut/PythonQwt/master/doc/images/QwtPlot_example.png
 
 .. _README: https://github.com/PierreRaybaut/PythonQwt/blob/master/README.md
 .. _online: https://pythonqwt.readthedocs.io/en/latest/