Skip to content

fix Example App launch of VideoSpeedTest#1599

Merged
j9ac9k merged 5 commits intopyqtgraph:masterfrom
pijyoi:choose_binding
Feb 22, 2021
Merged

fix Example App launch of VideoSpeedTest#1599
j9ac9k merged 5 commits intopyqtgraph:masterfrom
pijyoi:choose_binding

Conversation

@pijyoi
Copy link
Copy Markdown
Contributor

@pijyoi pijyoi commented Feb 22, 2021

Fixes the following:

  1. ExampleApp could not choose binding to launch VideoSpeedTest
  2. setGraphicsSystem is obsolete in Qt5
  3. VideoSpeedTest RawGLImageWidget frame rate capped by vsync (this should have been a side effect of converting from QGLWidget to QOpenGLWidget)
    • on Windows 10, instantiating RawImageGLWidget also caps the frame rates of the other two (ImageItem, RawImageWidget)
    • on Linux, doesn't affect the other two

@pijyoi
modify environment to choose binding
e09a397 
the Example App allows the user to choose a binding and a graphics
system to use to execute an example.

issue 1: setGraphicsSystem is obsolete and the method no longer exists.
thus selecting a graphics system causes an error.

issue 2: the choice of binding to use is passed as an extra command
line argument to be parsed by initExample. this is problematic for
examples that use argparse, such as VideoSpeedTest.py

issue 3: if the user has set PYQTGRAPH_QT_LIB, that takes precedence
over the choice made in the Example App.

this patch fixes the above 3 issues by setting PYQTGRAPH_QT_LIB in
the child process' environment and removing the old parsing of the
command line arguments
@pijyoi
remove calls to setGraphicsSystem
61616ff 
in fact all are no-ops
@pijyoi
remove graphics system combobox
a657dea
@pijyoi
regenerate templates
fbd1e89
@pijyoi
disable opengl vsync for benchmarking
d8e826e
@j9ac9k
Copy link
Copy Markdown
Member

j9ac9k commented Feb 22, 2021

Hi @pijyoi thanks for this PR! Had no idea there was a vsync cap issue on the benchmark, that certainly was problematic.

I'm messing with this PR, and noticing some major performance differences, unlikely due to anything in this PR, but curious if you've seen something equivalent, especially for Qt6 based variants (this is with otherwise default settings). All numerical values are in FPS, I put NA for the RawGLImageWidget for some of my test environments as they are not in Python 3.9, which you need python 3.9.1+ for pyopengl to work on macOS 11.

Binding GraphicsView + ImageItem RawImageWidet RawGLImageWidget
PyQt5 5.12 340 390 NA
PyQt5 5.15 360 380 NA
PyQt6 270 290 240
PySide2 5.12 300 350 NA
PySide2 5.15 360 410 NA
PySide6 240 250 200

I see no reason not to merge this PR, but wasn't sure where else to post this benchmark discrepancy

@j9ac9k
Copy link
Copy Markdown
Member

j9ac9k commented Feb 22, 2021

Also on PySide6, I can trigger a segfault when I close the two windows that come up from the "command-line usage" example (no segfault on pyside2, pyqt6)

Using PySide6
Fatal Python error: Segmentation fault

Current thread 0x000000010c4e7e00 (most recent call first):
<no Python frame>

@j9ac9k
Copy link
Copy Markdown
Member

j9ac9k commented Feb 22, 2021 •
edited
Loading

Also realized I get the segfault on master as well.

@j9ac9k
Copy link
Copy Markdown
Member

j9ac9k commented Feb 22, 2021

The issues I brought up are outside the scope of this PR. This PR LGTM, merging.

@j9ac9k j9ac9k merged commit c1900aa into pyqtgraph:master Feb 22, 2021
@pijyoi
Copy link
Copy Markdown
Contributor Author

pijyoi commented Feb 23, 2021

Ubuntu 20.04 64-bits using system Python 3.8, all packages from pypi.org.
No PySide2 5.12 wheel available for Python > 3.7
At least on my test system, values seem to be close together enough across different bindings.

Binding GraphicsView + ImageItem RawImageWidet RawGLImageWidget
PyQt5 5.12 590 720 500
PyQt5 5.15 600 700 480
PyQt6 600 700 480
PySide2 5.12 NA NA NA
PySide2 5.15 580 710 480
PySide6 580 690 470

Also on PySide6, I can trigger a segfault when I close the two windows that come up from the "command-line usage" example (no segfault on pyside2, pyqt6)

I have seen segfaults when running the examples on different systems. Seems to depend on Python version, binding, operating system etc. When they do happen, it's reproducible, not random. I don't get segfaults running the examples on my Windows 10 + Python 3.7 and Linux + Python 3.8 systems.

At one time, I was trying to make test_examples.py give each example a chance to end by itself but that ended up triggering many segfaults on the CI. So in the end, I thought that maybe using process.kill() to terminate each example was intentional in order to not have to deal with these segfaults.

I suspect one reason (but not the only) is that the examples are written in an interactive manner, instead of wrapping up things in a master parent widget like a QMainWindow.

@pijyoi pijyoi deleted the choose_binding branch February 23, 2021 03:39
@j9ac9k
Copy link
Copy Markdown
Member

j9ac9k commented Feb 23, 2021

Woah, those numbers... I'm on macOS 11, with no discrete GPU; I do have some other systems I can test on, which I may do...

on macOS I've seen huge performance boosts going from Qt 5.12 to Qt 5.15, but apparently that does not appear to be the case on Ubuntu.

We can continue this discussion elsewhere, I find it interesting, but I am not seeing anything that indicates an issue with the library.

@pijyoi pijyoi mentioned this pull request Mar 3, 2021
Merged
@j9ac9k j9ac9k mentioned this pull request Mar 9, 2021
Closed
@pijyoi pijyoi mentioned this pull request Mar 23, 2021
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@pijyoi @j9ac9k