#pyqtgraph | Python | Page 4

fervent vale May 22, 2024, 3:27 AM

#

oh, I was wrong about PyQt5. on RPI3 32-bits, it does support GLES. So what it doesn't support is ANGLE. Or perhaps it only supports exclusively one or the other

#

so yeah, RawImageWidget worked out of the box on RPI3 Bullseye 32-bits using distro-provided PyQt5 5.15

#

Bullseye comes with Python 3.9 though

#

it supports VAO because it has the GL_OES_vertex_array_object extension

fervent vale May 22, 2024, 10:41 AM

#

ha.. RawImageGLWidget can be implemented with QOpenGLTextureBlitter (in a new branch called texture-blitter). All the effort in modern_opengl branch was just a learning exercise...

rough furnace May 22, 2024, 11:24 AM

#

still a worthwhile learning exercise no doubt! I don't imagine all these different methods have notably different performance

rough furnace May 22, 2024, 2:46 PM

#

A while back you linked to a gist showing how to pass cupy arrays to opengl directly on the GPU (I could be remembering wrong) do you think that would be an effort worth trying on RawImageGLWidget? I can certainly help test on windows.

fervent vale May 22, 2024, 11:21 PM

#

https://github.com/pyqtgraph/pyqtgraph/discussions/2652

GitHub

Using PyTorch, CuPy and PyQtGraph for visualizing Images on the GPU...

Hi, I'm trying to display/visualize an image that is of type cupy.ndarray. The project I'm working on loads an image using PIL Image.open and converts it to a PyTorch torch.tensor on the CU...

rough furnace May 24, 2024, 8:44 PM

#

going through the numpydoc linter, wow, this thing is picky! I mean, it will make the docs better, but I thought I had ImageItem docs in decent shape but yeah... it has other opinions lol

rough furnace May 25, 2024, 4:09 PM

#

...what on earth, made a documentation change, now triggering segfaults all over the place

#

https://github.com/pyqtgraph/pyqtgraph/actions/runs/9236666607/job/25412765856?pr=3028

#

self._xp = cp.get_array_module(image) if cp else numpy```

instead of `numpy` it should have been `np` but wow, what a test failure!

fervent vale May 26, 2024, 1:10 AM

#

For setLookupTable, we don't want to advertise that the library permissively accepts ndarrays that are not uint8

rough furnace May 26, 2024, 1:11 AM

#

I'll say the docs require it...

fervent vale May 26, 2024, 1:12 AM

#

Either that or coerce it with np.asarray

#

Although then you have to consider np or cp

rough furnace May 26, 2024, 1:13 AM

#

pyright is really not happy about the self._xp bit, as it could be None

#

how's this?

        Parameters
        ----------
        lut : array_like or callable
            `lut` can be a callable that accepts the current image as an argument and
            returns the lookup table to use.  If using a np.ndarray, ensure the dtype
            is `np.uint8`

fervent vale May 26, 2024, 1:13 AM

#

The Callable part seems to be a feature exclusively for use with HistogramLUTItem

rough furnace May 26, 2024, 1:14 AM

#

you think I should not advertise callable in the docs?

#

I could remove it from the docs as an accepted type and leave a blurb about it in the Notes section.

rough furnace May 26, 2024, 1:18 AM

#

fervent vale Although then you have to consider np or cp

actually, this is handled in _ensure_proper_substrate method, I could coerce to self._xp.uint8 there...

fervent vale May 26, 2024, 1:18 AM

#

Based on my read of HistogramLUTItem, Callable exists to provide different lut lengths depending on image dtype

#

But HistogramLUTItem has now been changed to return 256 by default, I think

rough furnace May 26, 2024, 1:20 AM

#

I'm good w/ not advertising callable in the docs.

fervent vale May 26, 2024, 1:21 AM

#

I was reading about Array API and it says that it considers asarray to be an anti-pattern

rough furnace May 26, 2024, 1:21 AM

#

fervent vale I was reading about Array API and it says that it considers asarray to be an ant...

huh...

fervent vale May 26, 2024, 1:21 AM

#

For interoperability between libraries besides np and cp

#

NEP 47

#

It's also the rationale for moving away from numpy scalars

#

They should be single element numpy arrays instead of numpy scalars

rough furnace May 26, 2024, 1:24 AM

#

numpy scalars likely break various things in our library, I know ksunden ran into some issues there a while back

fervent vale May 26, 2024, 1:25 AM

#

So in the future, array agnostic code should use array api

#

So regarding asarray, the array api NEP says that inputs should just be ndarray. Don't allow python lists or any other convertible type

#

In the future anyway...

#

For getLevels, I'm not sure that it returns an ndarray. I think it returns whatever was set. The library coerces it to ndarray on use, if only to check whether it was multidimensional

#

That's the reason for the slowdown in numpy 2.0

#

User sets it using python list. Library coerces it to ndarray (double). Values are used as scalars. Ends up the image data gets upcasted to double

#

I think the Callable thing predates setColorMap

#

Or in any case, there's some kind of coupling between ImageItem and HistogramLUTItem

rough furnace May 26, 2024, 1:42 AM

#

fervent vale In the future anyway...

I'm good w/ putting warnings in the docs to only accept ndarray and that future support for array_like will be phased out

rough furnace May 26, 2024, 2:03 AM

#

"let me add the sphinx spelling extension and run it"

2000 spelling mistakes later

"ok... maybe I won't push this one to CI just yet"

#

Have to hit the hay, I'll incorporate your other feedback here shortly, I'm good w/ trying to phase out array_like and mandating ndarray ... I can put that on the docs, and add a section in the Notes or Warning about it currently working w/ array_like but that won't work indefinitely. Actually, I would even be good w/ emitting a warning.

fervent vale May 26, 2024, 2:08 AM

#

I think setLevels just leave it. The internal conversion to ndarray is to check whether it was 2d and is an implementation detail

rough furnace May 26, 2024, 9:39 AM

#

the type of levels is confusing, but as long as self._xp is not None (which it gets set to cp or np after the first call to setImage with an image, it gets set to np), it should either be None or ndarray: https://github.com/pyqtgraph/pyqtgraph/blob/master/pyqtgraph/graphicsItems/ImageItem.py#L148-L154

runic umbraBOT May 26, 2024, 9:39 AM

#

pyqtgraph/graphicsItems/ImageItem.py lines 148 to 154

if self._xp is None:
    self.levels = levels
    self._defferedLevels = levels
    return
if levels is not None:
    levels = self._xp.asarray(levels)
self.levels = levels```

rough furnace May 26, 2024, 9:40 AM

#

the fact that self._xp can be None is really annoying, should just default to np and rotate to cp is present.

#

similarly if autoLevels is provided, it uses self._xp.asarray to set the levels to.

rough furnace May 26, 2024, 10:08 AM

#

WARNING: Found 1066 misspelled words

😬 (many of the words identified as misspellings are not, it's still like ymax , setText and things like that. setText should have been grabbed by my script to build out dictionaries tho...)

fervent vale May 26, 2024, 11:12 AM

#

I don’t think I understood why there is a need to defer levels

#

But the large amount of statefulness in the name of caching makes the code complicated

#

Refactoring the fast path functions out to be stateless actually made it easier to reason and to optimize

rough furnace May 26, 2024, 11:23 AM

#

I think I'm done w/ this PR, I'll let it sit for a bit and see if other folks have suggestions. There are some other annoyances I have about ImageItem (the self._xp bit being None being the main one), but that is a an issue for a separate PR.

rough furnace May 26, 2024, 11:50 AM

#

also taking recommendation for what page should the docs improve for next.

rough furnace May 26, 2024, 12:11 PM

#

ok, discovered other stuff I don't like in this PR looking at the rendered output, .... this PR will never end 😅

rough furnace May 26, 2024, 12:56 PM

#

because I can't leave it alone, decided to incorporate an inheritance diagram. It handles switching between light and dark mode well.

runic umbraBOT May 26, 2024, 7:18 PM

#

failmail :ok_hand: applied timeout to @brave rock until <t:1716751712:f> (10 minutes) (reason: emoji spam - sent 34 emojis).

The <@&831776746206265384> have been alerted for review.

rough furnace May 29, 2024, 12:02 PM

#

started working on the docs for PlotDataItem last night, and ugh, ... this one is going to be a doozy. so much code to handle arguments coming in in any way shape or form. Definitely will have some API breakage... also using the opportunity to remove MetaArray support from PlotDataItem.

#

so yeah, not strictly working on the documentation, also redoing the function signatures ...way too much reliance on *args and **kwargs with a mountain of code to handle every combination of arguments/data-structures...

anyway thinking the next release may be 1.0.0 ... tired of zero-ver 😛

fervent vale May 29, 2024, 1:25 PM

#

It would be at least 0.14.0 to remove MetaArray anyway

rough furnace May 29, 2024, 1:30 PM

#

i'm not doing another patch release, I've gotten no reports of any other regressions

rough furnace May 30, 2024, 12:53 AM

#

wow, I'm going to break PlotDataItem in so many ways here. So little is documented, already encountering bugs just doing a close look-over...

also really hating self.opts

fervent vale May 30, 2024, 2:11 AM

#

If api is to be broken, one good one to break may be GraphicsItem::pos whose only role is to wrap the QPointF into a Point

#

It is one of the things requiring python inheritance to work

#

The mixin on the left or on the right issue

#

In the library, only ROI relies on it returning a Point.

rough furnace May 30, 2024, 2:15 AM

#

Make a PR, I’ll merge it

#

Make a solid description for Why tho, Luke will complain

#

Maybe give Martin (outofculture) a heads up too

fresh garnet May 31, 2024, 1:26 AM

#

I made an perfect download manager in python using requests

rough furnace May 31, 2024, 10:27 AM

#

fresh garnet I made an perfect download manager in python using `requests`

Looks great but this is not the right channel for this content

rough furnace May 31, 2024, 11:27 AM

#

this is so nice, see Phil announce a new pre-release on the riverbank pypi server, a couple of mouse clicks later, the CI is running that wheel...

#

should have done this years ago

fresh garnet May 31, 2024, 4:56 PM

#

rough furnace Looks great but this is not the right channel for this content

ok

rough furnace May 31, 2024, 11:43 PM

#

suppose this is getting better...

the fact that all these arguments are under **kargs is driving me up the wall.

fervent vale Jun 1, 2024, 12:47 AM

#

I think the library itself uses a boolean array for the connect argument.

rough furnace Jun 1, 2024, 1:11 AM

#

heh, finding other mistakes in the docs, but i'll correct that one ... numpydoc linter is actually... not as unhappy as I thought it would be... still nowhere near "complete" ... going to push a draft PR tonight tho

fervent vale Jun 1, 2024, 2:29 AM

#

It's not so much that it was supposed to be a boolean array

#

It's that boolean arrays convert to the int32 value required when building the serialized format

#

So an uint8 would work fine too

fervent vale Jun 1, 2024, 3:22 AM

#

as an exercise, I am implementing PCMI in OpenGL

fervent vale Jun 1, 2024, 3:50 AM

#

many years ago, I implemented something similar in OpenGL < 2.0

#

The QtOpenGL classes are really quite helpful: VAO, VBO, Texture, Shader

rough furnace Jun 1, 2024, 4:03 AM

#

what's PCMI?

fervent vale Jun 1, 2024, 4:10 AM

#

PColorMeshItem

rough furnace Jun 1, 2024, 4:21 AM

#

Oh right

#

The thing that really sucks with the OpenGL, at least modern OpenGL is doing thick lines… that I imagine will involve some work

fervent vale Jun 1, 2024, 4:33 AM

#

Oh I don't need to draw the edges

#

It's for visualising data on non-regular grids

rough furnace Jun 1, 2024, 4:46 AM

#

Right but doesn’t PCMI have an edge pen parameter allowing you to specify a thick line?

fervent vale Jun 1, 2024, 4:49 AM

#

That's not the useful part for me. With high density data, drawing any edges would blot out the image

rough furnace Jun 1, 2024, 5:54 PM

#

do you have a preference towards the bullet list, or the table format for the arguments?

rough furnace Jun 1, 2024, 6:23 PM

#

not sure what's w/ the extra bold font in the bulleted list... but I'm thinking I prefer the bulleted list vs. the table.

rough furnace Jun 2, 2024, 12:33 AM

#

https://pyqtgraph--3039.org.readthedocs.build/en/3039/api_reference/graphicsItems/plotdataitem.html#pyqtgraph.PlotDataItem

low hemlock Jun 3, 2024, 4:50 PM

#

Hi, I am trying to plot serial data that is voltage and current obtained from arduino using python's library PyQt using the graph widget - PlotWidget from PyQtGraph in real time.
So, data is continuously getting acquired based on the sample rate.
Let's say sample rate 10 ms (milli second), that means every 10 ms, data will be acquired and the same has to be plotted in real time.

So, PlotWidget want us to send data in list in setData method:
self.data_line.setData(x, y)

where x is a list containing timestamps and y is a list containing voltage values and the size of x and y has to be same for PlotWidget.

so, as data keeps on coming, these lists will get exceptionally large affecting the memory. Can we do something so that x gets the latest timestamp only and y gets the list of all the voltage values that we are obtaining from arduino, so that memory gets saved.

Is there any other method or library in which this is possible????

rough furnace Jun 3, 2024, 7:33 PM

#

You'll have to prune x and y yourself, the library won't (and shouldn't) make guesses on how much data you want to plot or how far back you want to go.

#

you could append the new data to a Queue (https://docs.python.org/3.12/library/queue.html#queue.Queue) and specify the maxsize to something reasonable, that way old data will fall off on its own.

onyx saddle Jun 3, 2024, 8:46 PM

#

rough furnace you could append the new data to a Queue (https://docs.python.org/3.12/library/q...

queues don't drop extra data, it blocks puts instead

#

You might be thinking of deque

rough furnace Jun 3, 2024, 8:47 PM

#

oh, yeah...deque... was going to say, one of those data structures drops old data

fervent vale Jun 4, 2024, 12:56 AM

#

You could decouple the plot update from the sampling rate. Have the acquisition append to some list. Have a QTimer set to the desired screen update rate. Whenever the timer fires, keep the last N newest samples and draw them

rough furnace Jun 4, 2024, 11:31 AM

#

was just running your PR, and can't help but feel silly for having the workflow

qimage = QImage("filename.png")
image = pg.functions.ndarray_from_qimage(qimage)
...
imageItem = pg.ImageItem(image)
...

wonder if it would be worth it for ImageItem to accept a QImage as its input, not just a numerical array.

fervent vale Jun 4, 2024, 11:41 AM

#

depends, is it to be considered a pre-rendered QImage with no more adjustments possible?

#

that violates many design assumptions of ImageItem

#

maybe the user would be better served by QGraphicsPixmapItem

rough furnace Jun 4, 2024, 3:21 PM

#

about to nump up min numpy to 1.24...

fervent vale Jun 4, 2024, 4:00 PM

#

in functions_qimage.py, there are 2 calls to xp.ascontiguousarray that would no longer be needed on numpy >= 1.24

#

don't know whether cupy is matching numpy behaviour there

rough furnace Jun 4, 2024, 4:01 PM

#

I'll remove those calls and test on my windows box before it gets packed up

fervent vale Jun 4, 2024, 4:16 PM

#

actually, that feature only requires numpy >= 1.23

rough furnace Jun 4, 2024, 4:17 PM

#

oh, even better, I'll try and test tonight

fervent vale Jun 4, 2024, 4:19 PM

#

i think it was the colorMaps example running with numba enabled that would fail on numpy < 1.23 if those lines were removed

#

i.e. it is actually a non-recommended usage that would be needed to hit that codepath

#

i.e. a combination of col-major and 512-length lut

rough furnace Jun 4, 2024, 4:32 PM

#

going back to API breakages, I guess I would need to do another release before 1.0.0 to slap warnings on future API breakages where no warnings currently exist.

fervent vale Jun 5, 2024, 3:57 AM

#

In order for the RawImageGLImage to display the peegee image (that makes use of translucent pixels) properly, we would need to add glClearColor and glBlendFunc

rough furnace Jun 5, 2024, 9:50 AM

#

Are they only available via pyopengl?

fervent vale Jun 5, 2024, 10:01 AM

#

It's available through the self.glfn

fervent vale Jun 5, 2024, 10:31 AM

#

It's a bit strange, the peegee image. The transparent pixels are white (255,255,255,0)

#

The drop shadow is implemented using alpha

rough furnace Jun 5, 2024, 10:55 AM

#

@wide prism ☝️

#

which peegee image specifically?

#

(chatting w/ Nils in another channel, discord is blocked by his proxy)

rough furnace Jun 5, 2024, 11:22 AM

#

fervent vale Jun 5, 2024, 11:34 AM

#

The 512@2x png. But it should be the same for the rest

rough furnace Jun 5, 2024, 11:35 AM

#

Nils requested you ese if these two are any different.

fervent vale Jun 5, 2024, 11:49 AM

#

looks too blurry to tell

#

the drop shadow is not visible against a dark background

#

ok, I have pushed a PR with the blending enabled. w/o blending enabled, the drop shadow gets rendered as a black border

#

The image below is w/o blending

#

alpha is ignored. thus the white transparent pixels show up

#

the drop shadow is implemented with black pixels with some non-zero alpha e.g. (0, 0, 0, 50)

#

and was intended to get mixed with the background to achieve the drop shadow effect

#

this one is with blending, the drop shadow is faintly visible

#

the white parts are now not from the image but from the background (which was set to white)

rough furnace Jun 5, 2024, 12:08 PM

#

oh wow, that looks ...much better

fervent vale Jun 5, 2024, 12:20 PM

#

there's a bleed through effect under WSL2 wayland

#

the translucent portions used to implement the drop shadow effect will show the contents of things behind the window

rough furnace Jun 5, 2024, 12:54 PM

#

if it's only on WSL2 Wayland, I would say categorize it as an upstream issue and move on... (or maybe leave a comment somewhere)

fervent vale Jun 5, 2024, 2:35 PM

#

It's fixed now

rough furnace Jun 5, 2024, 3:06 PM

#

🎊

rough furnace Jun 5, 2024, 3:32 PM

#

looking at the diff between your commits and the gl docs, looks like the difference is you specifying the dstAlpha value, and not using the default GL_ZERO?

rough furnace Jun 5, 2024, 7:08 PM

#

ugh, so many undocumented classes, I need to run sphinx apidoc to build the basic files... future documentation work I suppose

fervent vale Jun 5, 2024, 8:41 PM

#

glBlendFunc uses the provided settings for both the rgb colors and the alpha. It does not leave srcAlpha and dstAlpha as GL_ONE and GL_ZERO.

rough furnace Jun 5, 2024, 8:51 PM

#

in unrelated news, netflix seems to use pyqtgraph: https://github.com/Netflix/OpenVPCal

GitHub

GitHub - Netflix/OpenVPCal

Contribute to Netflix/OpenVPCal development by creating an account on GitHub.

rough furnace Jun 6, 2024, 2:31 AM

#

I think plotdataitem docs are done… next up, removing MetaArray

rough furnace Jun 7, 2024, 7:40 PM

#

any preference on how to breakup functions.py?

I was thinking of starting with two files.

numpy_to_qt.py where it's functions for coverting numpy arrays to Qt objects, and conversely a qt_to_numpy.py for the opposite direction.

that won't get all the methods out of functions.py but would likely trim down a lot of it.

fervent vale Jun 7, 2024, 7:59 PM

#

isosurface, isocurve and traceImage probably belong in their own separate file

rough furnace Jun 7, 2024, 8:01 PM

#

also low-key terrified of how broken flowchart/library/*.py is... there must be no test coverage there at all.... (but that's another problem)

fervent vale Jun 8, 2024, 12:03 AM

#

Canvas was marked to be removed after 2023/09

rough furnace Jun 8, 2024, 12:59 AM

#

Forgot about that one… will take that one out next.

fervent vale Jun 8, 2024, 11:37 AM

#

I have managed to port the long experimental opengl code in PlotCurveItem to use OpenGL 2.0 shaders in my own branch "curve-qtopengl".

#

Still uses GL_LINE_STRIP, which is available even in modern opengl

#

similar to RawImageGLWidget, it uses QtOpenGL only

rough furnace Jun 8, 2024, 12:25 PM

#

nice!!!!!!!

#

if you merge that, I'd like to give a shot at implementing thick lines using a geometry shader: https://github.com/mhalber/Lines?tab=readme-ov-file#geometry-shader-lines

There are some fantastic resources posted for webgl, which i'm hoping could be adapted: https://mattdesl.svbtle.com/drawing-lines-is-hard ...I'd like to give that a shot at some point

fervent vale Jun 8, 2024, 12:32 PM

#

is it only macos that does not implement thick lines for opengl?

#

it is "optional" to support thick lines

rough furnace Jun 8, 2024, 12:34 PM

#

I'm not sure what the limitation on thick lines across platforms, but broadly speaking, I remember the specification of thick lines not being consistent or working as expected...

#

definitely "optional" to support thick lines ... this is mostly something I was thinking would be a good thing for me to try to implement at some point

fervent vale Jun 8, 2024, 12:35 PM

#

while testing, I found that PlotSpeedTest changes the GraphicsView viewport widget for any change in settings

#

due to Interact

#

any change in any setting will result in a call to useOpenGL(bool)

rough furnace Jun 8, 2024, 12:36 PM

#

ugh

fervent vale Jun 8, 2024, 12:36 PM

#

which results in a fresh QWidget or fresh QOpenGLWidget being assigned as a viewport

rough furnace Jun 8, 2024, 12:36 PM

#

that doesn't seem ideal!

#

back to opengl merging criteria, I'm going to take a page from the matplotlib mentality regarding merging docs, "if it's an improvement, it gets merged" ... I won't block a PR because it's not "feature complete"

fervent vale Jun 8, 2024, 12:49 PM

#

the life-cycle of opengl objects is even less well-defined when using QGraphicsView

#

My main testing script for PlotCurveItem + opengl is the MouseSelect example

rough furnace Jun 8, 2024, 1:10 PM

#

sounds like as good of a place as any. Looking at the diff in the PR, should all the stuff you have in plotcurveitem remain there, like the OpenGLState and GLC dict, or should that go in a more generic location?

fervent vale Jun 8, 2024, 1:13 PM

#

OpenGLState is just a class to hold the opengl objects specifically needed for PlotCurveItem

#

the GLC dict is largish... but the intent is not to duplicate PyOpenGL

#

if there's a common need in the future, it could be factored out

rough furnace Jun 8, 2024, 1:21 PM

#

if the lifecycle is really confusing, especially across platforms/qt-bindings, would help to have a test suite that checks if buffers are destroyed and so on?

fervent vale Jun 8, 2024, 1:26 PM

#

there's no mechanism to do a cleanup on termination

#

On PySide, cleanup does not get called on termination

rough furnace Jun 8, 2024, 1:35 PM

#

oh, that doesn't sound ideal!

fervent vale Jun 8, 2024, 1:40 PM

#

the issue is that cleanup must be performed with the same QOpenGLContext that was used to create the objects

#

if that context is destroyed before the cleanup, the cleanup will fail

rough furnace Jun 8, 2024, 1:54 PM

#

curious what test coverage will go up to when we remove the canvas module 😂

rough furnace Jun 8, 2024, 2:27 PM

#

And 47% test coverage, oooof

mortal panther Jun 9, 2024, 12:41 PM

#

hey there

rough furnace Jun 9, 2024, 1:20 PM

#

mortal panther hey there

Want to make a PR improving the test code coverage? 😂

rough furnace Jun 9, 2024, 2:17 PM

#

pijyoi, you have any thoughts on un-vendoring colorama and just adding it to our setup.py/pyproject.toml as a dependency? Years ago it was more common to take the pyqtgraph directory and embed it in a project, but I've been trying to change the project structure to conform to more standard practices.

mortal panther Jun 9, 2024, 2:38 PM

#

rough furnace Want to make a PR improving the test code coverage? 😂

what do u mean

rough furnace Jun 9, 2024, 2:46 PM

#

mortal panther what do u mean

I just set up test coverage to PyQtGraph, it gives a percentage of how much of the codebase is run during the test suite. The number is quite low. If interested in contributing to the project, expanding the test coverage would be a great place to start!

fervent vale Jun 9, 2024, 3:37 PM

#

rough furnace pijyoi, you have any thoughts on un-vendoring colorama and just adding it to our...

Was it only used for some debug output? I only recall it used for the remote graphics. Is there a way to express that it is not needed for normal usage of pyqtgraph?

rough furnace Jun 9, 2024, 4:16 PM

#

I thought it was used in the console module too

#

Yes, we can make it optional, but it’s a bit annoying, would be something like pip install pyqtgraph[something]

#

If we really want it optional, best would be to emit a warning when colorama is not installed but could be used, but have a valid codepath absent colorama.

rough furnace Jun 9, 2024, 4:46 PM

#

Actually may not even need a warning, just document colorama would be utilized if present

rough furnace Jun 9, 2024, 6:55 PM

#

We have precedent with doing the same thing with scipy

fervent vale Jun 9, 2024, 11:01 PM

#

If it's a small dependency, then just add it unconditionally

#

Actually the vendored version of colorama is long overdue for update. Windows 10 terminal has native support for ansi sequences

fervent vale Jun 10, 2024, 12:32 PM

#

https://gist.github.com/pijyoi/d42d728d4cd54e847a485fa2a8a94ae0

Gist

demonstrate bug in PyQt5 versionFunctions getting deleted

demonstrate bug in PyQt5 versionFunctions getting deleted - pyqt5_opengl_functions_deleted.py

#

I have an MRE which I think demonstrates an issue with PyQt5

#

it seemingly only manifests itself for QOpenGLWidget used as a viewport to QGraphicsView

#

to summarize the issue: if two (or more) QGraphicsItem separately store the result of versionFunctions(), all except one of the returned versionFunctions() instances will get deleted

rough furnace Jun 10, 2024, 1:00 PM

#

relayed this to the mail list

#

thanks for the MWE!

rough furnace Jun 11, 2024, 2:06 AM

#

I see no reason not to merge pyqtgraph/pyqtgraph#3061 but as you've worked w/ ImageItem way more than I have, I'll let you chime in 😛

supple leafBOT Jun 11, 2024, 2:06 AM

#

GitHub

PROpen [pyqtgraph] #3061 Support downsampling of image data with NaNs

fervent vale Jun 11, 2024, 2:41 AM

#

The has_nans keyword is not propagated into the multiple axes codepath. Maybe also make the new argument keyword-only

rough furnace Jun 11, 2024, 10:30 AM

#

now that I've unvendored colorama, debating if I want to undo cprint as well, and just try and use colorama elsewhere... probably not.

rough furnace Jun 13, 2024, 12:22 PM

#

Last time I heard from bbc131 he said he wasn’t doing much work with PyQtGraph but he was still active. Hopefully he’ll chime in.

fervent vale Jun 13, 2024, 1:08 PM

#

the other thing is that paintGL casts the coords to float32

#

so that's yet another trade-off that the unsuspecting user may not be willing to accept

rough furnace Jun 13, 2024, 1:28 PM

#

True, but I think that's a thing for opengl somewhat across the board no? I would be good putting a warning in the docs next to the useOpenGL option.

fervent vale Jun 13, 2024, 1:35 PM

#

how about decoupling the decision to make paintGL non-experimental from the paintGL PR itself?

rough furnace Jun 13, 2024, 1:54 PM

#

I’m good with that

rough furnace Jun 14, 2024, 1:17 PM

#

I'm thinking more about expanding the test suite, but I'm struggling to think of how to test GraphicsItems are displaying as intended in ways other than pixel by pixel comparisons, which I would love to get away from. I mean, I can check attributes easily enough, I can check that plots are rendered correctly, but perhaps doing more image comparison won't be the worst thing in the world...

#

@plush fulcrum how do you all manage that for matplotlib?

plush fulcrum Jun 14, 2024, 1:25 PM

#

It’s a mix of strategies, though a fair number of image comparisons… which have drawbacks (we have to add tolerance because sometimes different cpu architectures do things slightly different, we are stuck on an old version of freetype because upgrading changes how text is drawn, so all of our test images would need to be regenerated)

rough furnace Jun 14, 2024, 1:27 PM

#

pyqtgraph has image comparisons, we got the tolerance down quite a bit by disabling anti-aliasing but yeah, we still have to have some fuzziness (or we could of course save images on each platform, which is something I really don't want to do lol)

rough furnace Jun 14, 2024, 2:40 PM

#

Some changes in guvectorize in numba 0.60.0

https://numba.readthedocs.io/en/stable/release/0.60.0-notes.html#new-features

fervent vale Jun 16, 2024, 1:15 AM

#

https://lists.debian.org/debian-devel/2018/11/msg00457.html

#

Extremely long thread about debian Qt arm32 using GLES vs arm64 using Desktop OpenGL

#

In short: there's only enough resources (time and effort) to provide either GLES or Desktop Qt packages for a particular architecture

#

Many existing desktop programs are coded to OpenGL, providing Qt as Desktop OpenGL allows those programs to run

#

Using RPI5 and pyqtgraph as an example: because Qt is provided as Desktop OpenGL on arm64, the old pyqtgraph opengl examples run out of the box on Raspberry Pi OS 64-bits

rough furnace Jun 16, 2024, 10:11 AM

#

I read through a bit of that, I’m surprised that the effort level is so high to try and include both but as it’s not my field I certainly won’t judge. Also surprised the situation hasn’t changed in 6 years.

rough furnace Jun 16, 2024, 11:25 AM

#

I think I'm misunderstanding something, if we're stuck with desktop opengl 2.0 on the Pi, then it's not feasible to write/compile our own shaders, right?

fervent vale Jun 16, 2024, 11:28 AM

#

RPI5 supports OpenGL 3.1

#

That would be GLSL 140

rough furnace Jun 16, 2024, 11:40 AM

#

huh, ok I did misunderstand, I thought we needed OpenGL 3.3 to write our own shaders.

#

oh looks like w/o opengl 3.3, we need to use fixed-mode pipeline

fervent vale Jun 16, 2024, 11:44 AM

#

No

#

#3055 doesn't make use of the fixed pipeline at all

#

OpenGL 3.3 Core profile enforces that fixed pipeline features are removed. So it's convenient to be sure you didn't slip in any old code

rough furnace Jun 16, 2024, 11:52 AM

#

ahhhhhh

#

appreciate the explanation

fervent vale Jun 16, 2024, 12:16 PM

#

pyqtgraph has some old shaders that make use of fixed pipeline

rough furnace Jun 16, 2024, 1:06 PM

#

Phil replied to me on the opengl issue w/ the example you provided:

fervent vale Jun 16, 2024, 1:22 PM

#

context has to be cached because all OpenGL objects (VAO, VBO, Shaders) are tied to the context in which they were created

#

if the context changes, all the objects have to be recreated

#

conversely, if the context remains the same (the common case), you can then continue using the objects

#

In a "normal" application, the context wouldn't ever change

#

In Qt, re-parenting causes a change in context

#

It's possible to not cache glfn and simply re-requesting it each time paintGL is called

#

The assumption is that Qt caches it under the hood and that it is cheap to re-request it

rough furnace Jun 16, 2024, 2:00 PM

#

Could we implement setParentItem (or whatever that method is) and recreate the context after super().setParentItem(parent) is called?

rough furnace Jun 16, 2024, 2:15 PM

#

I think just throwing a warning in the docs is with limitations on pyqt5 in that regard should be sufficient and assume we are dealing with “normal” applications

fervent vale Jun 16, 2024, 2:42 PM

#

Maybe calling keeping a reference to "context" as caching is the wrong term

#

What #3055 is doing by keeping a reference to "context" is merely to detect a change in "context"

#

Reparenting is just one way that a context could change

#

The reference to "glfn" could properly be called caching

#

Creating a list of function pointers to opengl functions could well be an expensive operation

#

And Qt is likely to cache it itself

#

In addition to the function pointers, PyQt populates the returned functions object with a HUGE bunch of opengl constants

#

So if PyQt doesn't cache the functions object itself, it would be expensive to re-request it each time

rough furnace Jun 16, 2024, 3:05 PM

#

If we need to keep a reference but not a Python reference, could we call QObject.setProperty(“context”, context) and retrieve it that way as well?

I would test myself but I’m not at a usable computer for a while…

fervent vale Jun 16, 2024, 3:10 PM

#

There's no issue with keeping a reference to context

#

The issue is that calling context.versionFunctions() kills any previous returned functions

#

And we don't know if subsequent calls to versionFunctions is cheap due to internal caching

#

#3055 already has a workaround by manually instantiating the opengl functions

rough furnace Jun 16, 2024, 11:40 PM

#

looking over your PRs, should I request glMultiDrawArrays be included in PyQt{5,6}?

rough furnace Jun 16, 2024, 11:59 PM

#

anything else you plan on adding to that PR? I'm good w/ merging as is; although I wouldn't mind more comments inserted so I can perhaps follow along a bit more (and maybe try and implement some simple graphicsitems in opengl in the future)

fervent vale Jun 17, 2024, 12:05 AM

#

https://doc.qt.io/qt-6/qopenglfunctions.html

QOpenGLFunctions Class | Qt GUI 6.7.1

The QOpenGLFunctions class provides cross-platform access to the OpenGL ES 2.0 API.

#

glMultiDrawArrays isn't listed there

#

That could explain why it's not included by PyQt

#

Which parts of the PR need more commenting?

fervent vale Jun 17, 2024, 12:36 AM

#

actually, for PySide2, if the functions were obtained via context.functions(), glMultiDrawArrays isn't inside too

#

context.functions() returns a set of functions that are availble in both GLES and OpenGL

#

it helps to ensure that you are writing code that work for both GLES and OpenGL

rough furnace Jun 17, 2024, 1:51 AM

#

Before the bit about the stenciling and quantization (you referenced the issue) works be good to see the bit of code intended to work around that issue

#

I don’t imagine there is an easy way to test the code as being compatible with gles and OpenGL

fervent vale Jun 17, 2024, 2:38 AM

#

Could you elaborate?

#

The original solution suggested in the issue was not general enough

#

It assumed there was only one ViewBox in the window

#

The idea is to find the drawable area in window coordinates

#

The old code transforms the drawable area to world coordinates, which in the vertex shader gets transformed back to window coordinates

#

Due to float32 precision, this back and forth transformation is lossy

#

In the current code, as you pan your plot, the stencil coordinates remain constant

#

In the old code, as you pan your plot, the stencil coordinates will change

rough furnace Jun 17, 2024, 2:53 AM

#

Ahh, that makes more sense

fervent vale Jun 17, 2024, 2:53 AM

#

Using Windows ANGLE was the easiest way to test that it works on GLES

#

The other change suggested in the issue was to move the origin for the data. However that is something that has more corner cases than I would be comfortable implementing in this PR, which is large enough as it were

#

The current code should be quite understandable, in my opinion

#

If you have existing applications, it should run with useOpenGL and enableExperimental

rough furnace Jun 17, 2024, 3:26 AM

#

Yeah I need to sit down and read through the diff more closely when I don’t have a cross-continental move looming

fervent vale Jun 17, 2024, 3:35 AM

#

I think the only parts that still look similar are the stencil function calls

rough furnace Jun 17, 2024, 3:42 AM

#

I do agree the code is clearly written. The suggestion for more comments was for my own benefit when I attempt to add OpenGL support for a simpler GraphicsItem i could read about intent and gotchas a bit better. Not a lot of examples of “modern OpenGL” to work with in the library that represent how we want/should have the implementation work. So comments pointing out edge cases that are being addressed, or gotchas along the way would be quite beneficial imo

rough furnace Jun 20, 2024, 8:44 AM

#

that float32 quantization issue is the same thing the SVG exporter sees, I attempted to do a similar thing by moving the data about the center (I also rescaled the data to fit between [-1, 1] but I couldn't get it working right w/o having to call .setData()

fervent vale Jun 20, 2024, 11:17 AM

#

Scaling doesn't change the precision of floating point

rough furnace Jun 20, 2024, 11:26 AM

#

No, but the way the svg writter worked it was truncating values and appending e+00<whatever> to the string being written to the SVG file.

fervent vale Jun 20, 2024, 11:57 AM

#

in the last svg example in #2661, adding 1e7 was sufficient to break svg

#

for OpenGL (w/o the origin shift applied yet), it needs 1e8 and more to trigger the quantization effects

#

for me, the applications where OpenGL curve plotting used to break down were those using DateAxisItem since the "now" offset is 1.7e9

rough furnace Jun 20, 2024, 12:13 PM

#

are you planning on adding anything else to your plotcurveitem PR? I'm good w/ merging as is, but you've kept adding stuff 😂

#

(don't get me wrong, the functionality/fixes you've added are great, they weren't things I was even aware of is all)

fervent vale Jun 20, 2024, 12:20 PM

#

It's done

#

it's almost at parity with the non-opengl codepath

#

no step, no filling

#

connect and pairs have different behavior in the presence of non-finites

#

that's fine, because in my opinion, having non-finites in those modes is undefined behavior

#

in other words, I consider the non-finite behavior that the non-opengl codepath has for those two modes to be undefined behavior

rough furnace Jun 20, 2024, 12:30 PM

#

i have a lingering ~~method~~ memory of drawing text in opengl didn't use subpixel aliasing or something like that...

fervent vale Jun 20, 2024, 12:30 PM

#

yes, the text rendering looks less sharp

rough furnace Jun 20, 2024, 12:31 PM

#

googling around about that a while ago, seems like freetype is the typical solution there, I am open to adding freetype as an optional dependency but you certainly don't need to add support to that for this PR

fervent vale Jun 20, 2024, 12:32 PM

#

the thick lines mode implementation for the non-opengl codepath was trying to mimick bug-for-bug the thin lines mode implementation

#

particularly for the connect and pairs mode

rough furnace Jun 20, 2024, 1:08 PM

#

As I was going through the docs for PlotDataItem, I sort of developed a hatred for how pyqtgraph tries to handle every conceivable scenario of data coming in in a variety of ways (dicts of lists, lists of dicts, recarrays) and having defined behavior for how we handle NaN's in every way .... yeah I'd love to reduce the code complexity and improve correctness at the expense of not handling every possible case and making the users have to do potentially more work...

rough furnace Jun 21, 2024, 5:34 AM

#

<@&831776746206265384> 👆

proper summit Jun 21, 2024, 5:47 AM

#

!ban 370195313140695044 spreading malware

runic umbraBOT Jun 21, 2024, 5:47 AM

#

failmail :ok_hand: applied ban to @thorn obsidian permanently.

haughty quartz Jun 21, 2024, 9:31 PM

#

How would I have two plots on the same graph but two different datas? As in, one line would be runtime (in ms) and the other line would be frames per second
So there would be a label on the left for runtime (ms) and label on the right for fps

#

line

#

if that makes sense

#

Like a hydrograph from geography - https://mammothmemory.net/images/user/base/geography/River landscapes/hydrograph-3a-in-river-landscapes-geography.3a8257c.jpg

rough furnace Jun 22, 2024, 3:52 AM

#

If you can reuse the same axes/scale this is pretty easy as you can call a number of different plot item methods https://pyqtgraph.readthedocs.io/en/latest/api_reference/graphicsItems/plotitem.html#pyqtgraph.PlotItem.plot

(Also check out the multidataplot method)

rough furnace Jun 22, 2024, 7:17 AM

#

@torn coyote https://jupyterlite-sphinx.readthedocs.io/en/latest/

Numpy has a PR to add interactive examples to their docs (I guess SciPy already does that?)

We should experiment if we can get Jupyter-rfb to work with this and see if we can do interactive examples as well

#

https://github.com/numpy/numpy/pull/26745

GitHub

ENH, DOC: Add support for interactive examples for NumPy with `jupy...

Description
This PR adds support for WASM-powered interactive examples with JupyterLite and jupyterlite-sphinx in the NumPy documentation.
A similar enhancement was made for the documentation for S...

rough furnace Jun 22, 2024, 7:47 AM

#

Hmm Jupyter lite doesn’t have server interaction, so I guess this runs via pyiodide and webassebly which I don’t think qt bindings support (yet?)

haughty quartz Jun 22, 2024, 1:20 PM

#

rough furnace If you can reuse the same axes/scale this is pretty easy as you can call a numbe...

Yes but I don't want to reuse the same axes

rough furnace Jun 22, 2024, 1:28 PM

#

haughty quartz Yes but I don't want to reuse the same axes

You can add a right hand axes I think but the values will be the same. To have different scales on the same plot is quite tricky. The library doesn’t help you out there, there are some PRs intended to implement that but we haven’t merged either of them

rough furnace Jun 29, 2024, 4:31 AM

#

@fervent vale I'm about to merge pyqtgraph/pyqtgraph#3068 unless you plan on other changes

supple leafBOT Jun 29, 2024, 4:31 AM

#

GitHub

PROpen [pyqtgraph] #3068 PlotCurveItem: implement OpenGL curve fill

fervent vale Jun 29, 2024, 4:35 AM

#

no more

#

I was looking at the PlotCurveItem regular paint() path. There's some redundant logic on avoiding creation of shadowPen added there by #816. setPen, setShadowPen ensure that the pens are either only None or a QPen.

#

I already removed mkPen, mkBrush from the paintGL path

rough furnace Jun 29, 2024, 4:54 AM

#

nice,... so much of that code has just been addition after addition (largely done by me 😬 ) if you see the opportunity add some test code to add coverage, I highly encourage that,. ... our test coverage is < 50% 😬

Still not sure how to vastly increase the test coverage of the test items w/o doing more image to image comparisons which I'm not particularly keen on...

#

if you hve thoughts on how to improve test coverage on the graphic items, please share!

#

almost too bad we are stuck w/ the fill stuff and can't convert PlotCurveItem to use drawPolyline

fervent vale Jun 29, 2024, 5:20 AM

#

The QPainterPath is also used by the mouse click detection

rough furnace Jun 29, 2024, 5:33 AM

#

oh! yeah, that's needed 😬

fervent vale Jun 29, 2024, 5:36 AM

#

so in a model where everything (curve, fill, mouse shape) revolves around QPainterPath, anything else becomes extra overhead.

rough furnace Jun 29, 2024, 6:51 AM

#

last I checked, the .drawPainterPath usage was almost as fast as .drawPolyLine ...if the mouse interaction is dependent on QPainterPath, staying w/ that makes the most sense.

Now I'm curious about mouse interaction and thick lines (since that uses .drawLines)

fervent vale Jun 29, 2024, 6:53 AM

#

if you use thick lines, then a QPainterPath is still generated for mouse shape

#

it's lazily created only if mouse shape is needed

#

okay i have refactored the pen / shadowpen thing

rough furnace Jun 29, 2024, 8:09 AM

#

unrelated to pyqtgraph but you have much experience w/ wifi mesh networks? In our assigned housing, running ethernet cabling is out of the question unfortunately, and the walls are pretty thick, so I'll likely need something w/ a whole lot of wifi-mesh APs... I have some ubiquiti network equipment, but haven't really used it in mesh mode before (does it work well?) open to other recommendations.

fervent vale Jun 29, 2024, 8:23 AM

#

rough furnace unrelated to pyqtgraph but you have much experience w/ wifi mesh networks? In ou...

No, I don't have any WiFi mesh at home. I do have 2 APs connected using ethernet over power though.

rough furnace Jun 29, 2024, 8:45 AM

#

fervent vale No, I don't have any WiFi mesh at home. I do have 2 APs connected using ethernet...

That’s the setup I had at my home in the states. It worked great.

untold matrix Jun 30, 2024, 2:14 PM

#

what channel is there for help?

onyx saddle Jun 30, 2024, 3:36 PM

#

untold matrix what channel is there for help?

#❓｜how-to-get-help

slender mortar Jul 2, 2024, 2:04 PM

#

@rotund lantern

#

python help needed

rough furnace Jul 2, 2024, 2:26 PM

#

slender mortar python help needed

wrong channel, look at #❓｜how-to-get-help

slender mortar Jul 2, 2024, 2:26 PM

#

mb

fervent vale Jul 3, 2024, 1:48 AM

#

HistogramLUTWidget hardcodes useOpenGL=False (instead of using whatever was configured in the config options) since the beginning of git history.

#

PlotCurveItem paintGL has implemented enough to render the stuff used by HistogramLUTItem

#

So it would be possible to change HistogramLUTWidget to useOpenGL=None

#

There are 2 bundled examples: ImageAnalysis and ImageView

#

The former uses HistogramLUTItem so will use whatever OpenGL config was set. The latter uses HistogramLUTWidget so disregards OpenGL config

rough furnace Jul 3, 2024, 4:03 AM

#

Setting useOpenGL to none makes sense to me

rough furnace Jul 3, 2024, 10:00 AM

#

<@&831776746206265384> ☝️

vivid crescent Jul 3, 2024, 10:02 AM

#

!ban 1156042416919429210 upwork scam

runic umbraBOT Jul 3, 2024, 10:02 AM

#

:incoming_envelope: :ok_hand: applied ban to @lean lava permanently.

young vessel Jul 3, 2024, 3:11 PM

#

.

#

Air

rough furnace Jul 7, 2024, 1:09 PM

#

not going to lie, was curious when a opengl implementation of PCMI was going to pop up 😄

fervent vale Jul 7, 2024, 1:58 PM

#

The bundled PColorMeshItem example doesn't really lend itself to benchmarking. Increasing the number of quads does not increase the density, but rather lengthens the data, which makes the output look not nice

rough furnace Jul 7, 2024, 1:58 PM

#

I'd like to keep that example (it does look nice) but definitely up for adding another more suitable example to the benchmark section.

#

although I am struggling to think of what would be a good example for benchmarking...

fervent vale Jul 7, 2024, 2:21 PM

#

I think the equation needs to be normalised so that increasing the size should only change its density

fervent vale Jul 8, 2024, 1:44 AM

#

2 million quads at 25 fps (only Z varying), that's good enough for my use case

rough furnace Jul 8, 2024, 3:06 AM

#

That’s a lot of quads! Nice work!

half jewel Jul 8, 2024, 2:52 PM

#

@alpine token Please don't advertise your trading bot on the server. It isn't on topic (we aren't a crypto server). I've removed your message.

alpine token Jul 8, 2024, 4:03 PM

#

hi lol thanks for not banning me. As you can see, my name says it all. And luckily that link is expired anyways...

fervent vale Jul 9, 2024, 1:48 AM

#

Surprisingly, pyqtgraph is in the top ~4000 pypi downloads?

half jewel Jul 9, 2024, 3:14 AM

#

That's pretty impressive 🎉

rough furnace Jul 9, 2024, 3:29 AM

#

I knew we were top 1% since pypi required my account have 2FA

fervent vale Jul 9, 2024, 3:50 AM

#

https://hugovk.github.io/top-pypi-packages/

Top PyPI Packages

A monthly dump of the 8,000 most-downloaded packages from PyPI

#

pyqtgraph is at rank 4082

rough furnace Jul 9, 2024, 3:52 AM

#

Thanks for the link, was going to ask for the source!

#

On reddit, i saw a thread asking about niche Python packages and PyQtGraph was one of the highest upvoted packages listed haha

fervent vale Jul 10, 2024, 12:09 AM

#

The new implementation now goes to 4 million quads at 50 fps

rough furnace Jul 10, 2024, 3:18 AM

#

by "new implementaiton" I assume you mean the diff from the 2nd commit in your PR about IBO and "flat shading"? Saw that commit as I was getting ready to go to bed last night, ...

oh, also love the use of the enum flags.

#

what's IBO short for? is this where you reuse the points in the triangle that overlap?

fervent vale Jul 10, 2024, 3:36 AM

#

Index Buffer Object. Another name is Element Buffer Object. Mesh is specified as indices, so shared vertices can be reused

#

However, sharing vertices is not enough, because colors are interpolated from the color of each vertex of the triangle

rough furnace Jul 10, 2024, 3:38 AM

#

that i did know from the little tinkering I did w/ opengl a while back... got some nifty gradients which I was confused by but then realized why it was happening

fervent vale Jul 10, 2024, 3:38 AM

#

So in the old implementation, vertices can't be shared for this reason

#

You need the flat shading feature

#

So that the colour of each triangle comes from only 1 vertex

#

So previously you needed 6 vertices per quad, I.e. you needed to upload 6 times the number of Z values

rough furnace Jul 10, 2024, 3:40 AM

#

ahh ok, so performance boost in that last commit is largely from not having to compute gradients across 6 vertices

fervent vale Jul 10, 2024, 3:42 AM

#

Not wasting cpu time creating a luminance buffer that is 6x bigger

rough furnace Jul 10, 2024, 3:44 AM

#

I have zero feel for opengl related performance optimizations, I just imgine just lessening buffer sizes and moving computation into the shader code whenever possible is the primary sort of thing to do.

fervent vale Jul 10, 2024, 3:50 AM

#

So for 4 megapix, that's 16 MB to upload per frame. (or 96MB if without IBO)

#

Moving rescale data into shader was a big win too

#

For a static image, you could effectively do change in levels and lut with no uploads

rough furnace Jul 10, 2024, 5:04 AM

#

~~how are you testing your branch?~~ disregard just saw the initial post in your PR

fervent vale Jul 10, 2024, 5:20 AM

#

My real use case is visualisation of live sensor data that is not rectangular

rough furnace Jul 10, 2024, 5:21 AM

#

Can you describe what you're measuring? Most of the "live" sensor data I've worked with I either want to treat as line plots, or convert to a frequency domain and view as a spectrogram or power density of some kind ...

fervent vale Jul 10, 2024, 5:33 AM

#

Say a plot where the grid is polar

rough furnace Jul 10, 2024, 5:34 AM

#

really wish Qt gave some way to handle non-linear transformations...

fervent vale Jul 10, 2024, 5:34 AM

#

I can't find any open source data, so I can't make a freely available example

rough furnace Jul 10, 2024, 5:35 AM

#

it's fine, I've worked w/ some polar plot data (radars/fire-control systems) before long ago

fervent vale Jul 13, 2024, 5:27 AM

#

#3093 is puzzling

#

I mean, sure, on a non-Qt based OpenGL app, it's up to the developer to create either a Desktop or a ES context

#

But on a Qt app, surely Qt should do the right thing and create whatever context the library was compiled against to begin with?

rough furnace Jul 13, 2024, 5:37 AM

#

sounds reasonable to me, wonder if this issue is new... are you able to reproduce the issue yourself?

fervent vale Jul 13, 2024, 5:37 AM

#

The "right" thing occurs on my systems

rough furnace Jul 13, 2024, 6:40 AM

#

Guess it’s a PySide bug?

fervent vale Jul 13, 2024, 10:18 AM

#

It would be a Qt bug

#

One thing I noticed from the user's eglinfo output is that for the Nvidia device, "EGL client APIs" list OpenGL_ES before OpenGL

#

so it could be that Qt's default surface is "DefaultRenderableType", which probably chooses the 1st item it finds from the system

#

after all, the 1st item "probably" is considered the "preferred" selection

#

I had hopes that one might be able to toggle to GLES in order to test out GLES on a Desktop system

#

but it doesn't work, it will crash

fervent vale Jul 14, 2024, 1:07 AM

#

A new finding: on Linux and using pypi provided PySide6, it is possible to use either OpenGL and OpenGL ES by setting the RenderableType beforehand. It crashes on Windows if you ask it to use OpenGL ES.

#

That simplifies testing of the GLES codepath greatly.

#

It probably wouldn't work on macOS

#

There are some other issues to resolve though: the versionFunctionsFactory does not work for GLES and an alternative has to be found

#

is this only true for the wheels provided on pypi (I.e. Compiled by the Qt company). How about distro provided versions.

rough furnace Jul 14, 2024, 3:30 AM

#

suppose we should check conda packages, ... would this be something that we could/should do in CI since we already test conda/pypi on each platform?

fervent vale Jul 14, 2024, 8:59 AM

#

It's something that you would just check offline rather than using the CI

rough furnace Jul 14, 2024, 9:01 AM

#

sure, but I don't have access to the 3 major platforms and with conda/pypi distributed packages in each right now 😛 was thinking CI might be able to verify that quicker since that's already setup (I know it doesn't test the linux distro variants of Qt)

fervent vale Jul 14, 2024, 9:04 AM

#

it doesn't work with PyQt bindings

#

it's a PyQt limitation, not a Qt limitation

#

or rather the issue can be broken down into 2 parts: 1) does Qt allow creation of GLES surface; 2) does the binding include the OpenGLFunctions

#

PyQt's issue is with (2)

#

It assumes that ES2 and Desktop are mutually exclusive

#

Unless conda or distros patch PyQt for this niche use-case, I think it's safe to assume that PyQt doesn't have this feature across the board

rough furnace Jul 14, 2024, 9:11 AM

#

conda definitely does patches on PyQt and PySide

#

not sure if they would patch this case or not...

#

even more confusing is the patches on the conda defaults channel will be different from the conda-forge channel

fervent vale Jul 14, 2024, 9:13 AM

#

The other niche use-case is that of using Qt's OpenGLFunctions

#

in Python

#

ah... the CI has already failed for conda PySide6

rough furnace Jul 14, 2024, 9:49 AM

#

for additional context, pyside2 is dead to me... if you get to the point that something works except on pyside2, ... yeah skip that test, or disable that functionality for pyside2...whatever is easier

fervent vale Jul 15, 2024, 12:00 PM

#

on a macOS running Qt, what version of OpenGL do you get?

rough furnace Jul 15, 2024, 12:04 PM

#

isOpenGLES: False
VENDOR: Apple
RENDERER: Apple M2 Max
VERSION: 2.1 Metal - 88.1

huh, ... what's w/ 2.1, macOS can run OpenGL 4.1

fervent vale Jul 15, 2024, 12:06 PM

#

that's what I saw in some forums too

#

but the CI macOS runners also show 2.1

#

So by right, the pcmi peegee wavy script was not using the more optimized flat shading path on macOS

rough furnace Jul 15, 2024, 12:08 PM

#


Run python -m pyqtgraph.util.glinfo
openGLModuleType: 0
isOpenGLES: False
VENDOR: Apple Inc.
RENDERER: Apple Software Renderer
VERSION: 2.1 APPLE-21.0.19
GLSL_VERSION: 1.20

#

pyside shows the same thing...

#

except openGLModuleType: OpenGLModuleType.LibGL

fervent vale Jul 15, 2024, 12:10 PM

#

0 == LibGL

rough furnace Jul 15, 2024, 12:11 PM

#

yeah I don't think I'm seeing the performance optimizations for the flat shading; man this is a bummer

fervent vale Jul 15, 2024, 12:11 PM

#

I read that macOS supports OpenGL 4.1 Core, not OpenGL 4.1 Compatibility

#

So the default might be 2.1, but possibly special setup is needed to get 4.1 Core

drifting perch Jul 16, 2024, 7:42 PM

#

yeah its flat shading

tawny silo Jul 17, 2024, 5:48 PM

#

hi
i am using pyqtgraph with pyside, it works fine for sometime but later the graphs freezes because i am updating them on a different thread, any other way to use threading with pyqt

rough furnace Jul 17, 2024, 7:01 PM

#

The general guidance is you need to update the graph on the gui thread (main thread), you can use threading to prep our collect the data to plot but you need to plot on the main thread.

There is a remote plot in the library that does allow for plotting on smother thread but it’s got other complexities and i wouldn’t recommend it

fervent vale Jul 17, 2024, 10:12 PM

#

https://github.com/pyqtgraph/pyqtgraph/discussions/2164

GitHub

Is it okay to use setData() on another thread? · pyqtgraph pyqtgrap...

Is it okay to use setData() on another thread? I digged in the docs and nothing about thread-safety was written in there.

tawny silo Jul 18, 2024, 9:00 AM

#

rough furnace The general guidance is you need to update the graph on the gui thread (main thr...

thank you

rough furnace Jul 19, 2024, 12:17 PM

#

Wow, what a good day to not work at Crowdstrike (or be a customer of crowdstrike)

runic umbraBOT Jul 20, 2024, 3:18 PM

#

:incoming_envelope: :ok_hand: applied timeout to @alpine lance until <t:1721489305:f> (10 minutes) (reason: duplicates spam - sent 4 duplicate messages).

The <@&831776746206265384> have been alerted for review.

rough furnace Jul 22, 2024, 2:00 PM

#

huh, this might be the first time I've seen this failure (it does look familiar so maybe just the first time i've seen it in a while) https://github.com/pyqtgraph/pyqtgraph/actions/runs/10042142913/job/27751887171?pr=3107

GitHub

Bump pytest from 8.2.2 to 8.3.1 in /.github/workflows/etc · pyqtgra...

Fast data visualization and GUI tools for scientific / engineering applications - Bump pytest from 8.2.2 to 8.3.1 in /.github/workflows/etc · pyqtgraph/pyqtgraph@68f3911

#

_____________________________ test_timeslide_snap _____________________________

    def test_timeslide_snap():
        count = 31
        frames = np.ones((count, 10, 10))
>       iv = pg.ImageView(discreteTimeLine=True)

tests\imageview\test_imageview.py:22: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
pyqtgraph\imageview\ImageView.py:136: in __init__
    self.ui.setupUi(self)
pyqtgraph\imageview\ImageViewTemplate_generic.py:32: in setupUi
    self.histogram = HistogramLUTWidget(self.layoutWidget)
pyqtgraph\widgets\HistogramLUTWidget.py:22: in __init__
    self.item = HistogramLUTItem(*args, **kargs)
pyqtgraph\graphicsItems\HistogramLUTItem.py:107: in __init__
    self.vb = ViewBox(parent=self)
pyqtgraph\graphicsItems\ViewBox\ViewBox.py:230: in __init__
    self.updateViewLists()
pyqtgraph\graphicsItems\ViewBox\ViewBox.py:1752: in updateViewLists
    nv = sorted(ViewBox.NamedViews.values(), key=view_key)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

view = <[RuntimeError('Internal C++ object (ViewBox) already deleted.') raised in repr()] ViewBox object at 0x1ccd029f500>

    def view_key(view):
>       return (view.window() is self.window(), view.name)
E       RuntimeError: Internal C++ object (ViewBox) already deleted.

pyqtgraph\graphicsItems\ViewBox\ViewBox.py:1749: RuntimeError

fervent vale Jul 23, 2024, 1:25 AM

#

https://gist.github.com/pijyoi/75eb055a15a9b1fc1131efc64a5d643e

Gist

PyQt glDrawElements binding issue

PyQt glDrawElements binding issue. GitHub Gist: instantly share code, notes, and snippets.

#

An MWE demonstrating a PyQt OpenGL binding issue

#

The equivalent in PySide6 works but involves a few changes other than just changing the binding import

#

The issue is in PyQt5 too, but for the sake of brevity, the script isn't compatible with PyQt5

terse hearth Jul 29, 2024, 11:52 PM

#

Can I ask a question about pyqtgraph here? I need help with implementing something

fervent vale Jul 30, 2024, 1:26 AM

#

Sure

rough furnace Jul 30, 2024, 2:24 PM

#

thanks for modernizing the GL stuff @fervent vale 🙏

is the usage of GL_SELECT problematic/difficult to replace with a non-fixed pipeline equivalent?

fervent vale Jul 30, 2024, 2:27 PM

#

Yes, it looks difficult

#

I thought GLVolumeItem would be difficult, but in the end it wasn't

#

Didn't even need to understand the core logic...

rough furnace Jul 30, 2024, 2:30 PM

#

not sure if you've chatted w/ the folks on the modern gl discord server, but they are super helpful for opengl related questions (not just specific to moderngl)

fervent vale Jul 30, 2024, 2:32 PM

#

The first one (GLMeshItem) to be converted was the most difficult

#

I think the question people would ask is why even go to OpenGL 2.1 programmable pipeline only

rough furnace Jul 30, 2024, 2:36 PM

#

There's incentive for both wanting to write our own shaders, and for running on the most amount of hardware we can

fervent vale Jul 30, 2024, 2:44 PM

#

BTW, the GLSurfacePlot animated example already looks different across different desktop gpus

#

On Intel, you will see a purplish colour on the negative amplitudes

#

On nvidia and amd, it appears black

rough furnace Jul 30, 2024, 2:51 PM

#

This reminds me of something I read in the OpenGL subreddit about different default values on different OpenGL drivers… anyway not a big deal

fervent vale Jul 30, 2024, 2:52 PM

#

it's use of undefined behavior

#

pow(x, y) is undefined for x < 0

#

and the 3d sinc function goes from -10 to +10

rough furnace Jul 30, 2024, 2:53 PM

#

oh, right... yeah makes sense

fervent vale Jul 30, 2024, 2:54 PM

#

so on one driver, it probably does pow(abs(x), y) instead, since x couldn't possibly be < 0

#

and on another it may just return a 0

#

oh, I didn't notice before. GLScatterPlotItem (and by extension GLGraphItem) didn't use to work on the RPI5. And also the grid lines used to be not visible

#

now they all work!

rough furnace Jul 30, 2024, 3:09 PM

#

nice!!!!!

#

something about high performance visualizations on really weak hardware like that is super slick

fervent vale Jul 30, 2024, 3:32 PM

#

Only the GLLinePlot example looks bad by default on the RPI5. Need to turn off anti-aliasing to look good

#

The grid used (on master) to have the same sort of look, and would only be visible if I disabled antialiasing

#

But now that GLGridItem uses GLLinePlotItem under the hood, I am not sure why the grid is visible but the curves still look the same

terse hearth Jul 30, 2024, 10:13 PM

#

terse hearth Can I ask a question about pyqtgraph here? I need help with implementing somethi...

I have a very quick question idk if it's possible but is there a way to make these grid lines less transparent

#

I have a class subclassing PlotWidget with the pen lines as ```py
x_grid_pen = pg.mkPen(color=grid_color, width=3)
y_grid_pen = pg.mkPen(color=grid_color, width=3)

plot_item.getAxis('bottom').setPen(x_grid_pen)
self.plot_left_axis.setPen(y_grid_pen)
``` grid_size is "g" by the way

#

nevermind I got it I had to add alpha=1 in showGrid

#

another quick question, is there a way to make the border between the graph and ticks a different colour? Right now it's green but I want it to be yellow

terse hearth Jul 30, 2024, 10:36 PM

#

I meant the thicker green lines that show as a border so the vertical and horizontal line that shows the level (not the grid)

fervent vale Jul 31, 2024, 4:26 PM

#

I have a private branch "check-core-profile" to test what needs to be straight out eliminated in order to run under Core Profile

#

minimally, what's left that needs to be eliminated are gl{Push,Pop}Attrib()

#

the rest like itemsAt() don't get activated if you don't use them

#

I am not sure what the side effects of removing gl{Push,Pop}Attrib() will be

#

much of the code just enables states that it needs w/o disabling them at end of use

#

part of that may have relied on gl{Push,Pop}Attrib to restore the state to a clean slate

#

all the OpenGL examples are standalone, meaning we don't get to see any bad interactions between GLGraphicsItems

#

(the exception being GLGridItem which gets used in almost all the examples)

rough furnace Jul 31, 2024, 6:00 PM

#

terse hearth another quick question, is there a way to make the border between the graph and ...

I don't think there is a way to do this easily off the top of my head; pretty sure the paint method draws all those with the same color...maybe a way to do it if you subclass AxisItem and override the paint method.

#

i'm not sure about gl{Push,Pop}Attrib ....my first thought would be to ask the modernGL folks, usually if you can link them to some existing code they're super helpful

rough furnace Aug 1, 2024, 12:19 PM

#

for NaN stuff, looks like SciPy is about to roll out an API, which I wonder if it would be helpful to look at mirroring that: https://docs.scipy.org/doc/scipy/dev/api-dev/nan_policy.html

fervent vale Aug 2, 2024, 2:55 PM

#

My all-in-one opengl testing script

rough furnace Aug 2, 2024, 5:43 PM

#

Missing the grid item? 😛

fervent vale Aug 2, 2024, 6:04 PM

#

Volume isn't inside. Anything else not inside are just derivative classes

rough furnace Aug 2, 2024, 6:34 PM

#

oh, I can see the grid now that i'm on my desktop, on mobile, it downsampled the resolution so much I couldn't see the lines in the background

fervent vale Aug 3, 2024, 2:15 AM

#

now volume is represented too

fervent vale Aug 3, 2024, 6:16 AM

#

woot, pyqtgraph now runs on OpenGL ES

rough furnace Aug 3, 2024, 6:20 AM

#

Nice work!!!!!!

fervent vale Aug 4, 2024, 11:39 AM

#

apparently GL_SELECT picking doesn't work with shaders

#

so it's not a question of whether GL_SELECT is modern or not

#

it no longer works with the programmable pipeline

#

so even if the code is left behind, it no longer works

#

I had an example in #2650 that demonstrates picking of GLImageItem(s)

#

it no longer works once the commit to draw GLImageItem using shaders is included

rough furnace Aug 4, 2024, 11:46 AM

#

ugh....

fervent vale Aug 4, 2024, 11:53 AM

#

even the example in #2650 is contrived

#

I am not sure if it worked before at all before for the other GLGraphicsItem

fervent vale Aug 5, 2024, 12:43 PM

#

That's strange, even though Matplotlib 3.9.1 has been yanked, Windows CI runners still manage to download the source tarball and build it (and fail during the tests)

rough furnace Aug 5, 2024, 2:12 PM

#

wonder if there is some caching going on ... i'm assuming that's the source of the windows CI failures?

fervent vale Aug 5, 2024, 2:32 PM

#

I was wrong about the GL_SELECT not working with shaders

#

It stopped working with shaders on my Windows AMD gpu laptop

#

However it still works on 3 other (Linux) systems that I tested on: WSL2 llvmpipe, RPI5, Intel Iris Plus

rough furnace Aug 5, 2024, 5:55 PM

#

I can test on macOS and windows with a NVIDIA GPU (and windows with amd GPU on my MacBook running boot camp)

fervent vale Aug 5, 2024, 10:07 PM

#

If you are testing, the script from #2650 needs the makeCurrent() added.

fervent vale Aug 5, 2024, 11:21 PM

#

rough furnace wonder if there is some caching going on ... i'm assuming that's the source of t...

https://github.com/pyqtgraph/pyqtgraph/blob/master/.github%2Fworkflows%2Fetc%2Frequirements.txt

GitHub

pyqtgraph/.github/workflows/etc/requirements.txt at master · pyqtgr...

Fast data visualization and GUI tools for scientific / engineering applications - pyqtgraph/pyqtgraph

#

Because of dependabot, pyqtgraph has pinned matplotlib 3.9.1 even though it got yanked

fervent vale Aug 6, 2024, 12:18 AM

#

OpenGL GL_SELECT

rough furnace Aug 6, 2024, 4:18 AM

#

alright, merged the change lowering the matplotlib version, feel free to rebase from master to make CI green again

digital void Aug 7, 2024, 6:57 PM

#

/me for calculator science engineering script

terse hearth Aug 10, 2024, 12:10 AM

#

I have a quick question when I make a pen using pyqtgraph, how do I set the line style at a later stage in code? I figured that since pyqtgraph returns a QPen object, I can call setStyle on it but this doesn't work; it doesn't do anything: ```py
self.x_grid_pen = pg.mkPen(color=grid_color, width=1)
self.y_grid_pen = pg.mkPen(color=grid_color, width=1)

and a little bit later in the code: ```py
        self.performance_tab.memory_grid_widget.x_grid_pen.setStyle(Qt.DotLine)
        self.performance_tab.memory_grid_widget.y_grid_pen.setStyle(Qt.DotLine)

        self.performance_tab.cpu_grid_widget.x_grid_pen.setStyle(Qt.DotLine)
        self.performance_tab.cpu_grid_widget.y_grid_pen.setStyle(Qt.DotLine)
``` am I missing something? Is this enough information?

terse hearth Aug 10, 2024, 12:37 AM

#

nevermind I got it just had to re-set the pen using setPen for the plot axes

terse hearth Aug 10, 2024, 2:38 AM

#

I have another question, how do I make these ticks have a label such as Mb/s, B/s, Gb/s for network usage tracking:

#

for the left plot axis

#

Also how do I prevent there from being TOO many ticks, I don't want it to look like this when a bunch of values happen

rough furnace Aug 10, 2024, 6:47 PM

#

Can’t remember off the top of my head how to do the units bit but I know that’s supported. What you have with the overlapping numbers shouldn’t be happening tho

terse hearth Aug 10, 2024, 7:25 PM

#

rough furnace Can’t remember off the top of my head how to do the units bit but I know that’s ...

Is this a bug?

#

right now I have to run this function to prevent all these ticks from happening but I'm not sure if I like this way of doing it, I was hoping there was something natively supported in pyqtgraph ```py
def set_equal_tick_spacing(self, n_ticks):
min_y, max_y = self.viewRange()[1]
if 0 <= max_y <= 32:
return # too small for change

    interval = (max_y - min_y) / n_ticks
    ticks = [(min_y, str(min_y))]
    for i in range(1, n_ticks + 1):
        tick_value = min_y + i * interval
        ticks.append((round(tick_value, 3), str(int(tick_value))))

    self.plot_left_axis.setTicks([ticks])```

rough furnace Aug 10, 2024, 7:28 PM

#

If you can create a minimum example showing that, definitely make an issue.

terse hearth Aug 10, 2024, 8:05 PM

#

terse hearth I have another question, how do I make these ticks have a label such as `Mb/s`, ...

well for now if anyone knows how to do this let me know

rough furnace Aug 10, 2024, 8:10 PM

#

terse hearth well for now if anyone knows how to do this let me know

Do none of the examples do this?

terse hearth Aug 10, 2024, 8:31 PM

#

rough furnace Do none of the examples do this?

I've tried searching I couldn't find it

terse hearth Aug 10, 2024, 8:39 PM

#

terse hearth right now I have to run this function to prevent all these ticks from happening ...

when I try to do it manually with this function (setTicks) using the correct unit labels the graph works for a little bit then breaks and shows nothing for the tick values

#

for reference this is the newly updated function I'm talking about ```py
def set_equal_tick_spacing(self, n_ticks):
min_y, max_y = self.viewRange()[1]
min_y = 0 if min_y < 0 else min_y
if 0 <= max_y <= 32:
return # too small for change

    #print(min_y, max_y)
    interval = (max_y - min_y) / n_ticks
    ticks = [(min_y, str(min_y))]
    for i in range(1, n_ticks + 1):
        tick_value = min_y + i * interval
        tick_value_speed, tick_value_speed_label = sys_utils.get_memory_size_info(
            int(tick_value)
        )
        tick_value_speed_full = f"{tick_value_speed} {tick_value_speed_label}/s"
        ticks.append((round(tick_value, 3), tick_value_speed_full))

    print([ticks])
    self.plot_left_axis.setTicks([ticks])``` `tick_value_speed_label` should be something like MB or GB or something like that

#

to make it easier to understand, it's basically settings the ticks to something like py [[(0, '0 B/s'), (1739.582, '1.74 KB/s'), (3479.163, '3.48 KB/s'), (5218.745, '5.22 KB/s'), (6958.326, '6.96 KB/s')]]

#

perhaps the labels on the right are the problem? When I remove them and just set the number and label to be equal it works just fine but it's not what I'm going for py (1739, "1739")

fervent vale Aug 11, 2024, 12:24 AM

#

Are you asking how to set units? There's a setLabel method for that that automatically prepends SI prefixes

#

https://github.com/pyqtgraph/pyqtgraph/discussions/2767

GitHub

plotItem.setLabel (units = 'what are all the possible values and de...

hi, I have been trying to get a reference list of all the acceptable values for units such as example>> units ='s' there are a few examples of some letters used however without explan...

terse hearth Aug 11, 2024, 1:03 AM

#

If I do that as an example py self.setLabel("left", units="MB/s") I get this

#

but I was more going for something like

#

is this possible?

fervent vale Aug 11, 2024, 1:28 AM

#

Not to my knowledge, I don't think that's built-in functionality. For your case, I suspect that the library code hasn't taken into account that the custom string could be longer than the length of the digits

terse hearth Aug 11, 2024, 1:31 AM

#

terse hearth for reference this is the newly updated function I'm talking about ```py def...

but it's weird because this function works, it does exactly what the second picture is showing but only shows the correct values SOMETIMES otherwise it doesn't show it at all

#

which means it should be possible to do

terse hearth Aug 11, 2024, 1:33 AM

#

terse hearth to make it easier to understand, it's basically settings the ticks to something ...

I think the problem is just the fact that it doesn't understand how to handle floating point numbers for the labels, so you can't have something like 3.45 KB/s, it must be rounded to 3 KB/s

terse hearth Aug 11, 2024, 1:34 AM

#

terse hearth I think the problem is just the fact that it doesn't understand how to handle fl...

actually nevermind that's not the issue at all

terse hearth Aug 11, 2024, 1:34 AM

#

fervent vale Not to my knowledge, I don't think that's built-in functionality. For your case,...

I think that this is right

#

actually nevermind I'm really not sure

#

I think the problem actually might be that the left axis width is too small for the tick to display, so pyqtgraph just doesn't draw it at all

#

yep that was the problem making it thicker fixed it

#

I really wish there was a native way to do this in pyqtgraph though, adding a custom unit for the actual ticks instead of a label would be really cool

#

one thing that just bothers me is when something like this happens, I'm really not sure why it happens, I think it might be when a large sudden jump of values gets plotted maybe? Like from 0 to 10 to 1000000 in a very short amount of time

#

anyways that just gets fixed when I call setTickSpacing with major and minor values

terse hearth Aug 11, 2024, 2:17 AM

#

One extra thing how do I make the plot go ABOVE the grid, not behind

terse hearth Aug 11, 2024, 2:45 AM

#

(fyi this is how the grid is being set) ```py
self.x_grid_pen = pg.mkPen(color=grid_color, width=1)
self.y_grid_pen = pg.mkPen(color=grid_color, width=1)

plot_item.showGrid(x=True, y=True, alpha=0.9)
self.plot_bottom_axis.setPen(self.x_grid_pen)
self.plot_left_axis.setPen(self.y_grid_pen)
also the plot item is being set Z Value as wellpy
plot_item.setZValue(10)``` but this doesn't work

anyone know?

rough furnace Aug 11, 2024, 4:05 AM

#

This comes up a fair amount so do a search on the issue tracker, you may need to set the z value of the axis items to something lower

terse hearth Aug 11, 2024, 4:12 AM

#

rough furnace This comes up a fair amount so do a search on the issue tracker, you may need to...

Will do, I’ll search the issues on the repository and see if I can find something there

rough furnace Aug 11, 2024, 5:27 AM

#

Maybe the Si unit functionality im thinking about is only for SpinBoxes and doesnt exist for AxisItem

rough furnace Aug 11, 2024, 7:59 AM

#

@terse hearth sorry you're running into so many issues 😅 ...the grid issue keeps coming up. GridItem is made the way it was for performance reasons, but that was a long time ago, since then, PyQtGraph has had major performance improvements (thanks pijyoi!) so we could probably take a performance penalty and make GridItem behave like other GraphicsItems, and not have it coupled to AxisItem as tightly

rough furnace Aug 11, 2024, 8:37 AM

#

https://github.com/pyqtgraph/pyqtgraph/actions/runs/10285073379/job/28462499843?pr=3122

nothing like an unusual version number to break your custom handling code; at least I know that I have access to the packaging library in the test suite so I should be able to fix this

terse hearth Aug 11, 2024, 4:36 PM

#

rough furnace <@940691242495385722> sorry you're running into so many issues 😅 ...the grid i...

Thanks I really appreciate the work you guys are doing for pyqtgraph it's a very useful tool. The performance penalty for GridItem and GraphicsItems should maybe be an option in the future

rough furnace Aug 11, 2024, 5:34 PM

#

I’m not clicking that lol; post a GitHub gist

pulsar sinew Aug 11, 2024, 5:35 PM

#

Deleted as a precaution. Post GitHub.

rough furnace Aug 18, 2024, 6:27 PM

#

Pijyoi, you done with 3109?

fervent vale Aug 18, 2024, 10:04 PM

#

rough furnace Pijyoi, you done with 3109?

Yes it's done.

half jewel Aug 26, 2024, 10:21 PM

#

!cban 928416998285787156 Cryptocurrency marketing is not allowed.

runic umbraBOT Aug 26, 2024, 10:21 PM

#

:incoming_envelope: :ok_hand: applied ban to @umbral hare permanently.

rough furnace Aug 28, 2024, 6:52 PM

#

https://aus.social/@virtualwolf/113033965970498667

If I had a 1TB memory machine would love to test PyQtGraph performance on this image:

https://aus.social/@virtualwolf/113033965970498667

aus.social

VirtualWolf

VirtualWolf (@virtualwolf@aus.social)

Holy shit, the Rijksmuseum used a 100MP Hasselblad camera to take almost eight and a half THOUSAND photos of the whole of Rembrandt's The Night Watch, for a total image size of 717 GIGAPIXELS. 😳

It's on their website as a zoomable image and you can zoom in so far you can see the individual cracks in the paint: https://www.rijksmuseum.nl/en/stor...

fervent vale Aug 29, 2024, 5:44 AM

#

The file size is 5.6TB, so 1TB may not cut it...

rough furnace Aug 29, 2024, 6:29 AM

#

ahh you're right, I read the gigapixels as gigabytes... silly me!

craggy bobcat Aug 29, 2024, 6:48 AM

#

what is pyqt used for

rough furnace Aug 29, 2024, 7:08 AM

#

craggy bobcat what is pyqt used for

making gui's using the Qt bindings in python

agile pebble Aug 31, 2024, 1:16 PM

#

Short question, is is possible to directly access the pixelbuffer of an scene object?

rough furnace Aug 31, 2024, 5:39 PM

#

agile pebble Short question, is is possible to directly access the pixelbuffer of an scene ob...

Not as far as I know, probably need to convert the widget to a QImage, and then you can access pixels, but it would be a copy so changes won’t take effect on the screen.

sacred oriole Aug 31, 2024, 5:50 PM

#

rough furnace https://aus.social/@virtualwolf/113033965970498667 If I had a 1TB memory machin...

wow that is one of the most zoomable digital images I have ever seen, the king though is google earth

#

wow pyqtgraph looks amazing I would like to give it a try today

rough furnace Sep 1, 2024, 7:19 AM

#

sacred oriole wow pyqtgraph looks amazing I would like to give it a try today

if you do find it fills a niche in your work, would love to know more about what kind of tasks you have pyqtgraph do... the library seems to appeal to people in experimental science or engineering, but I'm sure it has way more uses outside of those spaces.

fervent vale Sep 1, 2024, 7:21 AM

#

It seems to be used by people plotting stock market prices

rough furnace Sep 1, 2024, 7:23 AM

#

I've seen some mention of it being used for stock ticker stuff, but ...actually haven't seen that recently... (of course that doesn't mean it's not being used as such)

fervent vale Sep 1, 2024, 7:25 AM

#

#3141

rough furnace Sep 1, 2024, 7:26 AM

#

heh... ok I stand corrected 😄

fervent vale Sep 1, 2024, 7:30 AM

#

#2927

sacred oriole Sep 1, 2024, 7:49 AM

#

fervent vale #3141

what is this

rough furnace Sep 1, 2024, 7:50 AM

#

pyqtgraph/pyqtgraph#3141

supple leafBOT Sep 1, 2024, 7:50 AM

#

GitHub

IssueOpen [pyqtgraph] #3141 pyqtgraph draws an extra line per series when using a custom downsampling class

rough furnace Sep 1, 2024, 7:50 AM

#

pijyoi was referencing an issue to show a recent example of the library being used in the financial domain

sacred oriole Sep 1, 2024, 8:08 AM

#

rough furnace if you do find it fills a niche in your work, would love to know more about what...

I only tried their example, looks good

#

hey where can I get some stock market data?

rough furnace Sep 2, 2024, 1:31 PM

#

<@&831776746206265384> 👆

vivid crescent Sep 2, 2024, 1:36 PM

#

!cban 658989146488176640 advertising a roblox cheat

runic umbraBOT Sep 2, 2024, 1:36 PM

#

:incoming_envelope: :ok_hand: applied ban to @thorn stone permanently.

rough furnace Sep 3, 2024, 9:23 AM

#

<@&831776746206265384> 👆

vivid crescent Sep 3, 2024, 9:24 AM

#

!compban 638193267216875530

runic umbraBOT Sep 3, 2024, 9:24 AM

#

failmail :ok_hand: applied ban to @sick warren until <t:1725701057:f> (4 days).

agile pebble Sep 4, 2024, 4:33 PM

#

Hmm, any hints on how to the debug a pyside6 application crashing silently on windows (only very irregular, no traceback, heavy use of pyqtgraph). i have a feeling the crashs have some interaction with the parameter tree system but not sure yet

rough furnace Sep 4, 2024, 4:48 PM

#

Are you running the application from the console? If not I would so you might get a traceback (or maybe get a segfault notice in which case the debug will be far tougher)

#

You could wrap your whole application inside a try/except and when catching the unhandled exception log is to disk or something

#

If you’re getting an intermittent segfault, that’s going to be a lot tougher to debug. If you’re using qtpy you could also switch bindings to pyqt6 and see if you still get the crash, but first order of business is get a trace back or figure out if it’s a segfault

agile pebble Sep 4, 2024, 5:00 PM

#

It is latter :-(. And threads are involved :(. Good Idea with using PyQt6 for testing! But since I was annoyed by typing issues I switched from qtpy to pyside6 directly making it slightly harder.

#

Any experience with attaching a debugger to a python-qt process?

#

By the way, the recent change of values to items in pyqtgraph lead to a lot of silent breakage.

#

Maybe raising an exception directly when values is given would be preferable.

rough furnace Sep 4, 2024, 5:23 PM

#

Is that in parameter trees? I honestly haven’t used them that much, sorry for the breakages tho 😦

#

You think you’re getting a segfault? If so first order of business is make it easy to reproduce which I know is easier said than done

agile pebble Sep 4, 2024, 5:38 PM

#

It is seemingly random, but especially after interaction with some parameter trees, but this may be completely accidental. Sadly the python debugger does not catch anything, also the clean up events connected to app.aboutToQuit are not fired and the shell does not print any error

#

https://github.com/pyqtgraph/pyqtgraph/issues/2954 this one here, while the change is understandable, I think it should raise after it encouters an values key in the dict

GitHub

Parameter Lists : porting to 0.13.4 · Issue #2954 · pyqtgraph/pyqtg...

Documenting some behavior changes to Parameter list types since 0.13.4. Previously, it was possible to use values. Now limits must be used. (limits was added in 0.12.3) Previously, it was possible ...

#

If there interest I can do a PR for that

rough furnace Sep 4, 2024, 5:48 PM

#

definitely interest! I've been a bit slow to review PRs, I just had an international move and the kids have been home with short summer camps, they just started school again so I should be getting more time to work on the library.

but you don't catch anything even when you do something like

try:
    qApp.exec()
except Exception:
    print("cause some unhandled exception")
else:
    print("Exited the event loop cleanly")

?

agile pebble Sep 4, 2024, 5:59 PM

#

I use an exception handler which raises the exception from the main loop

agile pebble Sep 4, 2024, 7:03 PM

#

(all of numerous bugs show a nice error diaglog and show up in the log).

#

I may try the python faulthandler module

#

BTW Huge thanks for taking over the maintance of the pyqtgraph package, it powers all of our lab. And do not worry about being slow, since this is open source nobody has the right expect anything

rough furnace Sep 5, 2024, 2:05 AM

#

If you think you’re running into an intermittent segfault, those are the worst to debug certainly. The faulthandler is not a bad place to start, it might give you some insight. Also sometimes different platforms can give a little more information when being run from a console

rough furnace Sep 5, 2024, 2:27 AM

#

Was just thinking we eliminated a segfault in some of the conda package bindings but I can’t remember if it was pyside6 or what; but it involved QImage creation… not sure if that fix was in the latest release or not. That segfault only occurred in conda installed versions of the bindings (so another thing to try is to use a wheel binding if using a conda package)

fervent vale Sep 5, 2024, 2:57 AM

#

Yes, that fix is in 0.13.7

rough furnace Sep 5, 2024, 4:19 AM

#

@agile pebble the only other suggestion I have for debugging this segfault is create a log handler that writes to a file (or streams to a console) and just completely cover your library/application with log messages about what method is being executed/etc... and potentially evne use QSignalSpy to trace the emission of Signals ... eventually you will notice a method that is being run, but isn't being finishing... And let's just hope it's occurring in the same place 😄

agile pebble Sep 5, 2024, 7:29 PM

#

I had a already logging, but to be honest not very deep. Faulthandler helped me quite a lot.

#

I still not fully understand what happened, but somehow a call to cffi.cast() lead to the crash. But in ran fine like in like 99.99999% of all calls

agile pebble Sep 5, 2024, 7:49 PM

#

Working around that call seemed to fix the issues, and while it would be nice to know exactly what went wrong I sadly do not have to time to further debug this

#

Maybe some threadsafty issue

rough furnace Sep 6, 2024, 5:12 AM

#

Wow, you zero’s in on it, nice! Wonder if there was something different about the value you were casting that triggered the segfault.

coral maple Sep 8, 2024, 11:26 PM

#

Hi. I.need some one for help . Statistic and econometrics with python

rough furnace Sep 9, 2024, 3:37 AM

#

coral maple Hi. I.need some one for help . Statistic and econometrics with python

Look in #❓｜how-to-get-help

rough furnace Sep 10, 2024, 4:25 PM

#

<@&831776746206265384> 👆

fervent vale Sep 14, 2024, 4:21 AM

#

With regards to issue #2062, what is the purpose of an AxisItem responding to mouse events only to forward them to its linked ViewBox? Why not just remove the mouse event methods in AxisItem and let the events get naturally delivered to the ViewBox?

#

The downside is that the area occupied by the AxisItem no longer responds to mouse events, whereas previously those events get delivered to the underlying ViewBox

#

But that area is small

#

So a regular user would normally just click on some empty area to interact with the ViewBox rather than on the axes

rough furnace Sep 14, 2024, 4:30 AM

#

I have previously used mouse events on the axis item to zoom/pan one specific axis only, and have found that to be a good workflow.

#

(Away on a trip to Petra right now so I’ve only skimmed the comments in the issue, so I could be missing something).

#

Maybe less so for panning but more so for wheel events

fervent vale Sep 14, 2024, 7:18 AM

#

So it might make sense to just remove mouse click and drag events for Axisitem, but to retain wheel events

rough furnace Sep 14, 2024, 1:36 PM

#

That feels awkward as well, I need to spend a little time and read the issue more closely.

fervent vale Sep 15, 2024, 12:41 AM

#

The phenomenon is that drawing grid lines increases the bounding rect of AxisItem, which together with ViewBox having a zValue of -100, makes mouse events get delivered to AxisItem ahead of other items.

#

ViewBox having a zValue of -100 was surprising and had the effect of making its children stacked behind AxisItem

#

The user is unaware of these implementation details and only knows that enabling grid lines causes some mouse events to seemingly not get delivered to their item

fervent vale Sep 15, 2024, 1:06 AM

#

In PlotItem, there's a line that sets the zValue of AxisItem to 0.5 together with an erroneous comment about setting other items' zValue to greater than 1 in order to stack them above the AxisItem

#

That line is unaware that the base zValue of other items is in fact -100. A zValue >= 101 is needed to stack them above AxisItem

rough furnace Sep 15, 2024, 3:49 AM

#

GridItem/AxisItem interaction strikes again!

wide prism Sep 15, 2024, 1:47 PM

#

Would there be a point in filtering events by position? AxisItem knows where it draws the actual axis line. Having it handle any event ON or OUTSIDE of that line seems like it might be an understandable rule.

rough furnace Sep 15, 2024, 4:27 PM

#

Probably the best intermediate solution’s

rough furnace Sep 16, 2024, 5:48 AM

#

That’s not PyQtGraph code, not sure what it is, but I think you have the wrong channel.

fervent vale Sep 16, 2024, 2:42 PM

#

Does the following look like the right statement to insert into AxisItem's mouse event handlers?

if not self.mapRectFromParent(self.geometry()).contains(event.pos()):
    return

#

I gather from the code in boundingRect() that self.mapRectFromParent(self.geometry()) is supposed to be the axis sans the grid lines

wide prism Sep 16, 2024, 4:31 PM

#

From sticking color bars into the plots, I seem to remember that there is an internal gridlayout with the main plot area in the middle, and each axis in an adjacent grid cell. So if that's where the 'geometry' comes from, that seems like it might make sense?

fervent vale Sep 16, 2024, 10:07 PM

#

Now if you put it that way, then that means that AxisItem draws outside of its layout.

fervent vale Sep 17, 2024, 12:34 AM

#

I think the more elegant thing to do would be to override the shape() method.

wide prism Sep 17, 2024, 4:32 AM

#

See the docs on ```
hideOverlappingLabels

True (default for horizontal axis): Hide tick labels which extend beyond the AxisItem’s geometry rectangle.

False (default for vertical axis): Labels may be drawn extending beyond the extent of the axis.

(int) sets the tolerance limit for how many pixels a label is allowed to extend beyond the axis. Defaults to 15 for hideOverlappingLabels = False.```

#

Both x- and y-axis will overdraw into the bottom left "quadrant". This option is used to manually remove labels on the y-axis that are expected to do that. Otherwise the text from both axes just piles up there 🙂

#

I think I am trying to say that the current methods are not particularly elegant and that it could be quite reasonable to try out any alternative ideas you have.

fervent vale Sep 17, 2024, 4:42 AM

#

I did try out overriding shape() and it seems to work

#

The boundingRect of AxisItem includes the area covered by ViewBox

#

The default implementation of shape() is just boundingRect()

#

So overriding shape() allows us to specify that the grid doesn't belong

#

Qt uses shape() to determine collisions

#

So with the right shape, the AxisItem doesn't get a "hit"

wide prism Sep 17, 2024, 4:45 AM

#

That does sound very reasonable and, indeed, more elegant than collecting the events and then rerouting them.

#

Possible problems I could imagine are

things breaking with auto-expanding the ticklabel space
things breaking when rejecting tick labels that exceed a limited ticklabel space
(- things breaking with setting tickTextWidth / tickTextHeight)

#

But that should just be a matter of making sure they check for the right property (e.g. boundingRect(), not shape())

fervent vale Sep 17, 2024, 4:49 AM

#

boundingRect() is still used for paint damage detection

#

shape() seems to be dedicated to collision detection and mouse selection

wide prism Sep 17, 2024, 4:53 AM

#

I am sure Qt will be good, if there's any headache, it is in generateDrawSpecs, but it looks like that is correctly checking for the boundingRect():

br = self.boundingRect()
if not br.contains(rect):
   continue
textSpecs.append((rect, textFlags, vstr))

#

The logic gives me a bit of a headache: If you've already overdrawn your assigned box to include this space, then you can ALSO use that space to draw label text in.

#

But I see nothing in your fix to shape() that would obviously change behavior, sooooo.... that looks like a good idea to me?

#

Hmm. One possible issue:
I seem to remember that there were some previous experiment with collision boxes before. There may have been a use case where someone was drawing the labels on the inside of the y-axis, which works because of the weird bounding box.
Doing that might collapse shape() to be just one (or worst-case zero) pixel wide, which isn't helpful for interacting with the axis.

#

Is it worth considering giving shape a few pixels that overlap into the plot?

#

...keeping it non-zero width no matter what happens?

fervent vale Sep 17, 2024, 5:02 AM

#

shape() is just the mapRectFromParent(self.geometry())

wide prism Sep 17, 2024, 5:03 AM

#

...is that the cell it is originally supposed to be contained in ?

fervent vale Sep 17, 2024, 5:03 AM

#

Yes

#

It's the area given to it by the QGridLayoutItem

#

It's also the default bounding rectangle in the absence of grid lines and text over painting

#

def shape(self):
    rect = self.mapRectFromParent(self.geometry())
    qpath = QtGui.QPainterPath()
    qpath.addRect(rect)
    return qpath

wide prism Sep 17, 2024, 5:16 AM

#

That sounds like a very good starting point.

I am deep into hair-splitting and nit-picking here:
When zooming in and out, you can interact with a single axis by 'pointing the mouse at that axis' and scrolling the wheel.

But is the intuition for that to point at the axis line? In that case, having a few pixels of buffer into the main plot might avoid some frustration.

Or is it easily understood that you are scrolling the numbers?

#

Maybe we start with your elegant solution and then add as needed. 😄

fervent vale Sep 17, 2024, 5:19 AM

#

Actually I only learnt very recently about mouse wheeling over the axis

#

But my intuition was just to position the mouse over the numbers

#

Or clear of the boundaries anyway

wide prism Sep 17, 2024, 5:20 AM

#

If we get around to adding that to the docs, that might be a good way to describe it.
https://pyqtgraph.readthedocs.io/en/latest/user_guide/mouse_interaction.html

rough furnace Sep 17, 2024, 6:38 AM

#

The last batch of visitors have left, we just went to the major tourist sites within the country, and the kids are in school. I no longer have excuses (besides I need to catch up on some sleep). @wide prism would you be so kind to open an issue for the suggested docs change? As I go through converting docs to numpydoc I should see that issue and at the very least patch it then

rough furnace Sep 17, 2024, 9:23 AM

#

https://czaki.github.io/blog/2024/09/16/preventing-segfaults-in-test-suite-that-has-qt-tests/

wide prism Sep 17, 2024, 9:32 AM

#

Done. Sorry to just file an issue instead of a PR.

rough furnace Sep 17, 2024, 11:44 AM

#

ha, I really don't mind

#

sometimes we come across a "oh we should do this one thing" and I don't want to get that lost to the ether, and I want to keep the suggestions coming, so that can mean the middle ground is just opening a small issue with it

gusty bear Sep 18, 2024, 8:18 PM

#

crazy

fervent vale Sep 21, 2024, 12:55 AM

#

PyQt glDrawElements binding issue

olive ivy Sep 29, 2024, 3:32 PM

#

..

gilded dawn Oct 7, 2024, 12:31 AM

#

that crazy

#

so crazy

haughty brook Oct 11, 2024, 5:36 AM

#

Sorry to put the message here. I want to do open source in python based projects. Can anyone help me with that. Guide me which is easy and where should I start, or like help me get started.
I really want to be part of this but I don't know where to begin from.

rough furnace Oct 11, 2024, 9:30 AM

#

haughty brook Sorry to put the message here. I want to do open source in python based projects...

Usually pick a project you want to contribute to, read the contributing guide and look for issues tagged with "good first issue" and go from there.

rough furnace Oct 13, 2024, 6:37 AM

#

have a few hours to spare this morning, gonna start chipping away at that PR queue!

rough furnace Oct 13, 2024, 6:04 PM

#

Ok that was a lot of PRs merged… haven’t even looked at the issue tracker yet… another PR I want to take a closer look at too

rough furnace Oct 14, 2024, 9:49 AM

#

@wide prism if you get bored, mind giving this a once over? I'd feel better if it had a test, but expanding the tests of PlotDataItem to accomodate this would likely invove introducing parametrization where there currently isn't any ... low-key hate having to ask contributes to add that much in the way of testing...

rough furnace Oct 18, 2024, 5:40 PM

#

Man getting pings every 12 hours to merge two PRs from the same person is sure really motivating me to get right on it!

runic umbraBOT Oct 29, 2024, 4:21 PM

#

:incoming_envelope: :ok_hand: applied timeout to @undone sonnet until <t:1730219492:f> (10 minutes) (reason: emoji spam - sent 24 emojis).

The <@&831776746206265384> have been alerted for review.

rough furnace Nov 4, 2024, 1:06 PM

#

<@&831776746206265384> 👆

#

@wide prism just came across this lecture on color theory, can't help but think about your work in this space.... talking about HSL, HSV, Lab and so on... https://www.youtube.com/watch?v=6T_4mVZcozs&list=PL7ddpXYvFXspUN0N-gObF1GXoCA-DA-7i&index=13

YouTube

Marc Levoy

Marc Levoy - Lectures on Digital Photography - Lecture 13 (04May16)...

This is one of 18 videos representing lectures on digital photography, from a version of my Stanford course CS 178 that was recorded at Google in Spring 2016. A web site that includes all 18 videos, my slides, and the course schedule, applets, and assignments is http://sites.google.com/site/marclevoylectures

To help caption this lecture, follow...

▶ Play video

sweet comet Nov 14, 2024, 9:46 PM

#

🤔

eager stirrup Nov 19, 2024, 1:40 PM

#

:-;

#

;-;

dapper yarrow Nov 26, 2024, 8:11 AM

#

I have an application where I need to embed an pg.ImageView in a QGraphicsScene, where the ImageView becomes part of a node-based processing pipeline.
I can do this just fine, but when I do I cannot interact with the ImageItem or HistogramLUTWidget using the left mouse button.
Got any ideas?

Minimal example: https://paste.pythondiscord.com/HN3A

rough furnace Nov 26, 2024, 9:40 AM

#

Huh, are you integrating into a QGraphicsScene or pg.GraphicsScene? I’ll try and take a look after I wrap up my work today.

dapper yarrow Nov 26, 2024, 9:50 AM

#

into a QtWidgets.QGraphicsScene
The use case is this kind of thing (image from internet), but with an pg.ImageView in one of the nodes

rough furnace Nov 26, 2024, 11:36 AM

#

I know in our GraphicsScene (which we inherit from QGraphicsScene) we add some niceties for mouse-interaction but mouse stuff should still largely work with a QGraphicsScene

I believe the issue is ultimately that ImageView is a QWidget, and not subclassed from QGraphicsItem .

A while ago I did this UML diagram of pyqtgraph's major classes and how they relate to Qt's classes we inherit from.

https://pyqtgraph.readthedocs.io/en/latest/api_reference/uml_overview.html

#

Appreciate the minimal example, I would change the ImageView to an ImageItem and see if you get the mouse behavior you're wanting (even though it's not exactly what you want).

#

hmm.. but QGraphicsScene has the addWidget() method...

#

perhaps try scene.addItem(image_view..imageItem) ?

dapper yarrow Nov 26, 2024, 12:58 PM

#

rough furnace Appreciate the minimal example, I would change the `ImageView` to an `ImageItem`...

The thing I really want is the HistogramLUTItem (because it is great!), so just the ImageItem does not really get me to where I want.
I find it interesting that the mouse wheel works in the HistogramLUTItem but click+drag does not

rough furnace Nov 26, 2024, 1:02 PM

#

dapper yarrow The thing I really want is the `HistogramLUTItem` (because it is great!), so jus...

I assume ColorBarItem https://pyqtgraph.readthedocs.io/en/latest/api_reference/graphicsItems/colorbaritem.html#pyqtgraph.ColorBarItem won't do it for you? Can you not add the HistogramLUTItem directly? also maybe consider making GraphicsLayout of HistogramLUTItem/ImageView (that isn't a Widget) and adding that, as a GraphicsLayout can be added like any other GraphicsItem

dapper yarrow Nov 26, 2024, 1:23 PM

#

hmmm, I'll have to think about it.
The framework I'm working in has nodes on a QGraphicsScene and the nodes are widgets (with layouts etc.) using QGraphicsProxyWidget.
I would really like to have pyqtgraph visualizations inside nodes, which I think means they have to exist as widgets inside the node, and not on the QGraphicsScene directly (although this is otherwise a good idea).

I'll look more at it tomorrow. I'm thinking I will try to figure out why the mouse click events are not registered, and alternatively see if I can capture the events in a parent and then forward them to the right place.

rough furnace Nov 26, 2024, 2:08 PM

#

There was another instance of odd behavior w/ QGraphicsProxyWidget ...but can't remember the specifics. It's quite possible (likely?) that the library has not implemented something correctly causing some incompatibilities w/ QGraphicsProxyWidget ... I wouldn't be surprised if use of it drags down performance significantly though. If you're wanting the histogramLUTItem and the ImageItem next to each other, I would consider adding them in a pg.GraphicsLayout and adding the layout via scene.addItem(layout) instead of trying to get the whole ImageView in there.

rough furnace Nov 29, 2024, 12:16 PM

#

<@&831776746206265384> 👆

pulsar sinew Nov 29, 2024, 12:16 PM

#

!cban 1272239087415201824 Discord link spam

runic umbraBOT Nov 29, 2024, 12:16 PM

#

:incoming_envelope: :ok_hand: applied ban to @grizzled rain permanently.

opaque jacinth Dec 5, 2024, 4:23 AM

#

wait was a library banned or a user?

rough furnace Dec 5, 2024, 9:06 AM

#

opaque jacinth wait was a library banned or a user?

this channel for some reason seems to be major target of spam on this server, so it's quite common for a user to come into this channel, post spam, I flag the moderators and the user is banned. It's unfortunate but jjust how it goes I sometimes

opaque jacinth Dec 5, 2024, 9:46 AM

#

rough furnace this channel for some reason seems to be major target of spam on this server, so...

creepy

#

noice

idle pulsar Dec 20, 2024, 1:41 PM

#

all is one and one is all

hazy trout Dec 21, 2024, 4:30 AM

#

I just want to follow the thread; it's too far up to scroll. Is there a follow button?

hazy trout Dec 21, 2024, 4:31 AM

#

idle pulsar all is one and one is all

❤️ the slogan

lost fractal Dec 21, 2024, 6:43 AM

#

help

rough furnace Dec 22, 2024, 11:47 AM

#

Pijyoi thanks for reviewing that xml PR

runic sphinx Dec 22, 2024, 1:37 PM

#

I guess that works fine

fervent vale Dec 23, 2024, 12:13 AM

#

As I understand the PR, there's 1) serialization and 2) an xml format. Every current Parameter present in the library is already serializable, if you want to save it to disk, you could pickle it without using xml. The issue seems to be that the author has other new and more complex Parameters that may not be easily serialized with the current pyqtgraph infrastructure.

rough furnace Dec 23, 2024, 5:07 AM

#

@mortal grotto feel like reviewing that PR?

rough furnace Jan 13, 2025, 2:12 PM

#

oh no, linux CI is ❌ due to

E: Unable to locate package libegl1-mesa

fervent vale Jan 14, 2025, 1:07 AM

#

It seems to work if you change libegl1-mesa to libegl1

rough furnace Jan 14, 2025, 3:42 AM

#

Thanks for the tip. I’ll make the change in a little bit

rugged ruin Jan 16, 2025, 9:07 PM

#

hey guys is there any repo on github or any website that showcases what we can do with pyqtgraph in the fields of image processing and signal processing ? Thanks in advance. i just knew about this library and I am trying to find useful tutorials on the specified fields

rough furnace Jan 17, 2025, 4:38 AM

#

rugged ruin hey guys is there any repo on github or any website that showcases what we can d...

I havent seen tutorials specifically aimed at those fields but my signal processing work is what brought me to pyqtgraph to begin with. I used pyqtgraph to render spectrograms, power spectral density, all while having interactive components so users can tweak regions they’re interested in.

Pyqtgraph’s origins are from being able to analyze images taken from microscopes and perform various processing tasks.

#

Truth is we have a variety of examples. Some simple, some more complex, but we don’t have written up tutorials.

rough furnace Jan 17, 2025, 7:50 AM

#

Thanks to @torn coyote for bringing this to my attention, but pyqtgraph is being used at NASA as a front end for their trajectory simulation toolbox. A large component being the performance. Thanks @fervent vale for your work on improving performance of the library over the years!

Link to the paper: https://ntrs.nasa.gov/citations/20240014320

POST Explorer: A Design Space Exploration Tool for POST2 - NASA Tec...

Recent improvements for the Program to Optimize Simulated Trajectories II (POST2) have included the development of an application programming interface (API). This API allows POST2 simulation inputs to be directly manipulated from other applications (such as MATLAB or Python), and the outputs from POST2 are streamed directly to the external appl...

calm hound Jan 20, 2025, 11:22 AM

#

Hello Everyone

#

I am new here need the information about what this PROJECT is all about ?and how can I involve in there to contribute anything in my capacity?
Can somebody explain or provide me document to review and study please.

rough furnace Jan 20, 2025, 11:25 AM

#

you're going to have to be way more specific. What are you looking to contribute to specifically? Our documentation isn't hidden.... and neither is our issue tracker.

calm hound Jan 20, 2025, 11:28 AM

#

rough furnace you're going to have to be _way_ more specific. What are you looking to contribu...

Actually it am new to discord so I am exploring here , but anyway I found the docs

runic umbraBOT Jan 20, 2025, 1:28 PM

#

You are not allowed to use that command here. Please use the #bot-commands channel instead.

half jewel Jan 28, 2025, 2:40 AM

#

This is not the right channel. Also, the rules explicitly prohibit recruitment and also paid work.

clear abyss Jan 28, 2025, 2:41 AM

#

Appologies!

rough furnace Feb 3, 2025, 2:43 PM

#

@wide prism I'm using a lot of your colormap work in converting colormaps into best fit polynomial curves as I'm trying to add support for a lot of these colormaps for a work thing, but I can't use lookup tables, I have to provide an algebraic equation ... feels a little ridiculous in some cases to use a 9th order polynomial just to express a color, but what can you do 😅

fervent vale Feb 4, 2025, 1:07 AM

#

Could you check for macOS whether the GLGraphItem example draws the nodes? I suspect it doesn't get drawn on the master branch

rough furnace Feb 4, 2025, 3:58 AM

#

original and zoomed in results on macOS

fervent vale Feb 4, 2025, 4:18 AM

#

Thanks. So it works on macOS. There's a bug in GLScatterPlotItem in its decision to enable GL_POINT_SPRITE. By right, CompatibilityProfile can only be present for OpenGL >= 3.2. Since macOS defaults to OpenGL 2.1, that should have been NoProfile.

#

The reason it works for most systems is that the driver can provide a higher GL version than that requested. I.e. Qt defaults to requesting 2.0 and most drivers would then just provide a 4.x version with CompatibilityProfile

#

But since macOS only has CoreProfile for anything above 2.1, I would have thought that the provided context on macOS would be 2.1 NoProfile

rough furnace Feb 4, 2025, 5:53 AM

#

Happy to test, appreciate the explanation!

fervent vale Feb 4, 2025, 2:52 PM

#

Okay, finally found out why things were “working” despite the bug. In some version of Qt after 6.5, QOpenGLWidget’s backend was switched to QRHI (OpenGL). This backend has already enabled GL_POINT_SPRITE for versions of OpenGL where it is needed. So I tested with PySide 6.5 with a NoProfile context, and indeed the nodes were not drawn. And with PySide 6.8, the nodes were drawn.

#

This was quite puzzling, because I could have sworn enabling GL_POINT_SPRITE was required a few months back

#

PR3236 does get the nodes drawn on PySide 6.5

karmic kernel Feb 4, 2025, 3:52 PM

#

rough furnace original and zoomed in results on macOS

Woah! Cool!

vernal owl Feb 5, 2025, 7:08 AM

#

@fervent vale @rough furnace does pyqtgraph have a discord server

rough furnace Feb 5, 2025, 7:09 AM

#

vernal owl <@862941683460210688> <@369662814488363009> does pyqtgraph have a discord server

no, we just use this channel

vernal owl Feb 5, 2025, 7:09 AM

#

oh i see

rough furnace Feb 5, 2025, 7:10 AM

#

vernal owl oh i see

I don't have time to admin/moderate a server 😂 ... the mods here have been wonderful hosts for us.

vernal owl Feb 5, 2025, 7:15 AM

#

rough furnace I don't have time to admin/moderate a server 😂 ... the mods here have been won...

tht makes sense

rough furnace Feb 5, 2025, 7:06 PM

#

@fervent vale I had identical FPS results on my (macos) laptop, 15 FPS to 120 FPS. I always get a little weirded out when performance improvements like that are identical across platforms.

fervent vale Feb 5, 2025, 11:25 PM

#

There was a further simplification to the calculations that I abandoned as I thought that would need more verification. gl_Position.w is already a z distance between the camera and a vertex. That would allow not sending the camera position and the view transform to the shader

#

For now, it's hopefully just a pure translation of cpu code to shader code

rough furnace Feb 7, 2025, 1:02 PM

#

linux-arm64 CI runners are available on github actions, probably should add those:

https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/

rough furnace Feb 7, 2025, 2:59 PM

#

@fervent vale I'm considering bumping the minimum Qt6 version up from 6.2+ to something ...much newer. I'm open to suggestions on suggested versions.

Ubuntu 24.04 looks to send PyQt6 6.6.1 with apt. 22.04 doesn't seem to have a python3-pyqt6 package.

fervent vale Feb 7, 2025, 3:03 PM

#

The very latest pyqt6 and pyside6 wheels for aarch64 are for glibc 2.39, which means ubuntu 24.04 and fedora

#

RPI5 OS which is debian 12 can't use them

rough furnace Feb 8, 2025, 5:43 AM

#

hmm..both my apple and windows machines have little endian byte order, i'll check the linux-24.04-arm64 runner, unless they're big endian, I'm not sure there will be any benefit to testing pyqtgraph on there.

copper lily Feb 10, 2025, 3:16 PM

#

o

rough furnace Feb 10, 2025, 6:43 PM

#

<@&831776746206265384> 👆

fervent vale Feb 12, 2025, 2:30 PM

#

Reading through your 2021 conversation in Modern OpenGL discord channel, some of the issues mentioned there have since been overcome. The OpenGL code no longer uses fixed pipeline, it runs on OpenGL ES, shaders have been updated to GLSL 120.

rough furnace Feb 12, 2025, 2:31 PM

#

4 years ago, I can't remember 4 days ago 🙃

#

you've done amazing work making pyqtgraph's opengl functionality actually work much better!

fervent vale Feb 12, 2025, 2:32 PM

#

It’s still OpenGL 2.1, not OpenGL 3.x. Would require some big refactoring to go to OpenGL 3.x.

#

Intel drivers on Windows require a #version directive if you are on OpenGL 3.x context even if your shader is OpenGL 2.0 only

#

One bigger obstacle is actually the use of QPainter over painting

rough furnace Feb 12, 2025, 2:35 PM

#

QPainter over painting?

fervent vale Feb 12, 2025, 2:35 PM

#

Over-painting

#

Used in GLTextItem and GLGradientLegendItem

rough furnace Feb 12, 2025, 2:36 PM

#

oh oh oh, now I follow

fervent vale Feb 12, 2025, 2:36 PM

#

QPainter clobbers some OpenGL state

#

The requirement of #version directive means you need to prepend a different version string if you want to support OpenGL ES

#

The current shaders.py doesn’t support such a usage

#

OpenGL 3.x requires use of VAO. The current code relies on VAO not being required.

#

One trick of migrating to OpenGL 3.x is to create 1 single VAO, bind it, and forget about it. That satisfies the VAO requirement

#

However, QPainter clobbers the OpenGL state. So you can’t do the above VAO trick

rough furnace Feb 12, 2025, 2:46 PM

#

when I asked that of modernGL a few years back, it was mostly wondering if it would be worth the effort to use moderngl as a dependnecy, as it would likely be easier to use that raw opengl which I didn't know the first thing about.

fervent vale Feb 12, 2025, 2:50 PM

#

trimesh could actually replace the MeshData class

rough furnace Feb 12, 2025, 4:19 PM

#

I’d be good with that, if it can reduce our codebase and expands functionality… suppose the trick would be to make it an optional dependency

rough furnace Feb 21, 2025, 8:48 AM

#

pijyoi, you closed your distance to camera PR? You find an issue with it? (was about to start going through PRs and notice you closed that one).

fervent vale Feb 21, 2025, 9:12 AM

#

I am going to start making some GLGraphicsItems work with strict Core Profile. I will fuse the closed PR with that change

#

Strict Core Profile will not understand GLSL lower than 140. That includes macOS

#

Maybe in a future release, minimum OpenGL version can be bumped up to 3.1

#

That would require requesting a non-default version on macOS

rough furnace Feb 21, 2025, 9:21 AM

#

I'm good w/ doing that whenever, I doubt there are that many platforms that pyqtgraph is used on that can't run OpenGL 3.1.

Regarding macOS I'm still not understanding the underlying issue. macOS supports up to OpenGL 4.1 (granted it's been deprecated forever)

#

I believe you've tried to explain to this to me before 🙃

#

(is this one of those "but Qt for macOS is built with OpenGL 2.1?)

fervent vale Feb 21, 2025, 9:22 AM

#

On Linux and Windows, the default behavior is that the driver will give you the highest version it can. On macOS, the default is 2.1

rough furnace Feb 21, 2025, 9:23 AM

#

I don't imagine opengl gives you a way to search for the underlying platform and conditionally set the version based on that?

fervent vale Feb 21, 2025, 9:23 AM

#

On Linux and Windows, you will probably get at least OpenGL 4.3 Compatibility Profile

#

On macOS, if you request for it, you can only get OpenGL 4.1 Core Profile

#

Compatibility profile lets you use old shaders from OpenGL 2.x. Core Profile is not required to support OpenGL 2.x shaders

#

So for macOS, you can't incrementally upgrade your shaders. If you want to use newer features, all your old shaders need to be updated

rough furnace Feb 21, 2025, 9:28 AM

#

Is the thing holding us from going to a Core profile the amount of work involved, or were the Qt compatibility issues as well?

fervent vale Feb 21, 2025, 9:29 AM

#

So Qt's recommendation is that the OpenGL profile be set at an application level

#

Rather than on a per-widget level

#

We could set the profile just for GLViewWidget, but if someone embeds it into their other QWidget, I don't know what happens

rough furnace Feb 21, 2025, 9:30 AM

#

or was it that the version of Qt we would run on embedded devices wouldn't work w/ Core profile? ... (my apologies, I know you've explained this to me before, no doubt it would do me good to get more familiar w/ working w/ OpenGL)

fervent vale Feb 21, 2025, 9:31 AM

#

As a library, it shouldn't be pyqtgraph that sets the profile globally

rough furnace Feb 21, 2025, 9:31 AM

#

fervent vale We could set the profile just for GLViewWidget, but if someone embeds it into th...

maybe we set it in mkQApp? that's usually where I set the Application level settings usually which I don't want conflicting w/ other people's usage of the library.

fervent vale Feb 21, 2025, 9:33 AM

#

So for embedded devices, specifically RPI5 and RPI4, they support OpenGL ES 3.1, which is close to OpenGL 4.3

#

On RPI5 , the best result is obtained by using the ES profile

#

So if you were to set it in mkQApp, the advice for macOS users who don't want to use mkQApp would be for them to manually request a 4.1 profile

rough furnace Feb 21, 2025, 9:37 AM

#

I think that's a perfectly viable thing to do put in the documentation.

#

Hell, I can even put it inside a fancy "warning" box 😄

fervent vale Feb 21, 2025, 9:38 AM

#

Compute shaders actually works on the RPI5

#

That's a requirement of OpenGL 4.3 and OpenGL ES 3.1

#

Furthermore, OpenGL 4.2 is required to be compatible to OpenGL ES 3.0

#

(See how macOS just missed the mark...)

rough furnace Feb 21, 2025, 9:41 AM

#

...yeah macOS's support of only 4.1 is a real bummer for sure

#

Keep wondering if this would be suitable for an eventual replacement: https://doc.qt.io/qt-6//qrhiwidget.html

QRhiWidget Class | Qt Widgets 6.8.2

The QRhiWidget class is a widget for rendering 3D graphics via an accelerated grapics API, such as Vulkan, Metal, or Direct 3D.

fervent vale Feb 21, 2025, 9:42 AM

#

On Linux desktop systems, requesting an OpenGL ES context also works

rough furnace Feb 21, 2025, 9:42 AM

#

(granted couldn't do that until we kill off Qt5 support)

#

of course not sure if that works w/ our own shaders...

#

https://doc.qt.io/qt-6//qshader.html#details

could use that (again, won't work w/ Qt5)

QShader Class | Qt GUI 6.8.2

Contains multiple versions of a shader translated to multiple shading languages, together with reflection metadata.

fervent vale Feb 21, 2025, 9:48 AM

#

So what I'm going to do is to move the shader code out of shaders.py and into their respective GLGraphicsItem source files

#

Except for the GLMeshItem shaders, not sure what to do with them yet

#

Then each GLGraphicsItem needs 2 versions of the shader code

#

1 legacy for OpenGL 2.x

rough furnace Feb 21, 2025, 9:49 AM

#

are the differences between versions substantial?

fervent vale Feb 21, 2025, 9:49 AM

#

1 for Core / OpenGL ES 3

#

You can see the difference in texture3d and texture3d-es3 shaders

#

So texture3d-es3 will be changed to be used by Core and ES3

#

Just need to prepend the correct version string depending on whether we are Core or ES3

fervent vale Feb 21, 2025, 11:19 AM

#

I have just pushed the first implementations for GLLinePlotItem and GLScatterPlotItem

rough furnace Feb 21, 2025, 12:04 PM

#

alright, got my son setup with an Ori and the Will of the Wisps randomizer... should have some time to merge some PRs 😂

#

I assume you're still working on pyqtgraph/pyqtgraph#3246 ?

supple leafBOT Feb 21, 2025, 12:06 PM

#

GitHub

PROpen [pyqtgraph] #3246 create a default VAO if using Core profile

fervent vale Feb 21, 2025, 12:22 PM

#

I need to test out the new changes

rough furnace Feb 21, 2025, 12:22 PM

#

let me know if you need me to test on macOS

fervent vale Feb 21, 2025, 12:23 PM

#

Well, I could make the various examples default to OpenGL 4.1 Core on macOS.

rough furnace Feb 21, 2025, 12:23 PM

#

the last mac that didn't support OpenGL 4.1 Core is absolutely ancient...

#

Mid-2010 models started supporting 4.1. Those versions of macOS don't support anything newer than macOS 10.13... which has been unsupported for some time now. tl;dr on macOS I think we can safely assume 4.1 (although if we wanted to support older devices, we would have to support 3.3)

fervent vale Feb 21, 2025, 12:40 PM

#

Okay you should be able to try out some examples

#

Actually, we are not using any features newer than OpenGL 3.2 Core

rough furnace Feb 21, 2025, 12:43 PM

#

not sure if it's your branch/PR but I got this error when testing the 3d image example:

#

/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py:260: RuntimeWarning:
Traceback (most recent call last):
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/examples/GLImageItem.py", line 60, in <module>
pg.exec()
...
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py", line 221, in paintGL
self.paint(region=region, viewport=region)
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py", line 233, in paint
self.drawItemTree(useItemNames=useItemNames)
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py", line 265, in drawItemTree
self.drawItemTree(i, useItemNames=useItemNames)
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py", line 260, in drawItemTree
debug.printExc()
--- exception caught here ---
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/GLViewWidget.py", line 257, in drawItemTree
i.paint()
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/items/GLImageItem.py", line 134, in paint
program = self.getShaderProgram()
^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/ogi/Developer/pyqtgraph/pyqtgraph/opengl/items/GLImageItem.py", line 113, in getShaderProgram
compiled = [shaders.compileShader([glsl_version, v], k) for k, v in sources.items()]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/ogi/.pyenv/versions/pyqtgraph-pyqt_607-py312/lib/python3.12/site-packages/OpenGL/GL/shaders.py", line 235, in compileShader
raise ShaderCompilationError(

#

OpenGL.GL.shaders.ShaderCompilationError: ('Shader compile failure (0): b"ERROR: 0:11: Invalid call of undeclared identifier \'texture2D\'\\n"', [b'#version 140\n', b'\n        #ifdef GL_ES\n        precision mediump float;\n        #endif\n        uniform sampler2D u_texture;\n        in vec2 v_texcoord;\n        out vec4 fragColor;\n        void main()\n        {\n            fragColor = texture2D(u_texture, v_texcoord);\n        }\n    '], GL_FRAGMENT_SHADER)
  debug.printExc()

#

confirmed 3D Image works on master

fervent vale Feb 21, 2025, 12:57 PM

#

Oops. Cut and pasted too much

#

It's the GLImage2D example

#

It actually worked on my system without error

rough furnace Feb 21, 2025, 1:01 PM

#

confirming it works

fervent vale Feb 21, 2025, 1:01 PM

#

Only means my driver was not strict

rough furnace Feb 21, 2025, 1:01 PM

#

#

ok, let me check videospeedtest and such

fervent vale Feb 21, 2025, 1:02 PM

#

VideoSpeedTest is not affected

rough furnace Feb 21, 2025, 1:03 PM

#

huh macos failing in CI, let me see if I can reproduce locally...

fervent vale Feb 21, 2025, 1:27 PM

#

The error has to do with line width

#

Core is stricter on what line widths it accepts

#

Core is only obliged to support line width of 1

#

We could remove the 4.1 profile for the GLLinePlotItem example for now

rough furnace Feb 21, 2025, 1:34 PM

#

I'm good with that

fervent vale Feb 21, 2025, 2:46 PM

#

I fixed it the "right" way by querying the device line width limits

fervent vale Feb 21, 2025, 3:04 PM

#

Okay I have tested it out on OpenGL ES 3 too...

rough furnace Feb 21, 2025, 3:34 PM

#

the codeql error about buf not being used, I assume that's a false positive as while we don't use the variable we still need to select the buffer?

fervent vale Feb 21, 2025, 3:39 PM

#

I am not sure if it's needed. But I didn't want to change the original code

#

That code path is not much exercised

#

It actually still works on nvidia drivers

#

Even with the code upgraded to use shaders

#

That path is most certainly not a Core feature

rough furnace Feb 21, 2025, 3:46 PM

#

fair enough... i'll merge momentarily

rough furnace Feb 21, 2025, 4:36 PM

#

noticed I'm getting a segfault on exit of ConsoleWidget.py example

fervent vale Feb 21, 2025, 5:05 PM

#

Could you try the GLMeshItem and GLSurfacePlot examples?

#

Oops. Didn't realised you had merged

rough furnace Feb 21, 2025, 5:44 PM

#

no problem, i'll give those a try on 3256 shortly

#

both of those examples run w/o error

rough furnace Feb 21, 2025, 6:38 PM

#

should I merge 3256 or you plan to add more to it?

fervent vale Feb 21, 2025, 8:54 PM

#

Can merge.

fervent vale Feb 22, 2025, 12:37 AM

#

So macOS got OpenGL 4.1 Core in 2013. So all supported systems have it. The issue that I mentioned is that you likely can't mix and match widgets using different context versions

#

So it probably wouldn't work to instantiate an application using pyqtgraph in one widget and vispy in another

#

Or less far fetched, would PlotCurveItem opengl path currently even work with 4.1 Core?

#

Probably can just leave this question to be solved for the future

#

I think the library is in a good spot now that GLViewWidget can run on both OpenGL 2.1 and 4.1 Core

#

It sure took a long time to get there!

rough furnace Feb 22, 2025, 6:04 AM

#

wouldn't have happened w/o your effort pijyoi, thank you!!!!

fervent vale Feb 22, 2025, 7:22 AM

#

Ok, I have also fixed up PlotCurveItem and PColorMeshItem

rough furnace Feb 22, 2025, 7:36 AM

#

I'll test on macOS right now

#

hm... PColorMeshItem ... I need to change the example to set useOpenGL= True and enableExperimental = True right?

#

hmm...with those flags set to true, on macOS I'm getting slower FPS than w/o setting them, is that expected?

fervent vale Feb 22, 2025, 7:45 AM

#

I think the library example dimensions may need to be bumped up for OpenGL to be useful

#

I had an example before with much larger dimensions

rough furnace Feb 22, 2025, 7:47 AM

#

With useOpenGL=True and enableExperimental=True I'm getting 60 fps, without it I'm getting ~80 fps

fervent vale Feb 22, 2025, 7:47 AM

#

60 would be the vsync

rough furnace Feb 22, 2025, 7:48 AM

#

i'll dial up the density

#

thanks for including that parameter 😄

#

oh my god...

fervent vale Feb 22, 2025, 7:49 AM

#

The non-OpenGL will choke

rough furnace Feb 22, 2025, 7:49 AM

#

dialed density to 5, on CPU it came to a grinding halt, opengl still at 60 fps

#

wow...

fervent vale Feb 22, 2025, 7:54 AM

#

To test out on Core, you would need to add the surface format code

#

MouseSelection example is set up to use the opengl path

rough furnace Feb 22, 2025, 8:04 AM

#

if 'darwin' in sys.platform:
    fmt = QtGui.QSurfaceFormat()
    fmt.setRenderableType(fmt.RenderableType.OpenGL)
    fmt.setProfile(fmt.OpenGLContextProfile.CoreProfile)
    fmt.setVersion(4, 1)
    QtGui.QSurfaceFormat.setDefaultFormat(fmt)

set this on the PColorMeshItem.py example before mkQApp?

fervent vale Feb 22, 2025, 8:07 AM

#

Yes. You could even set it for all platforms

rough furnace Feb 22, 2025, 8:07 AM

#

ok, with that blurb, I'm getting significantly better performance. With density=20 not forcing Core 4.1 I'm getting 18 FPS. I'm getting 35 FPS with forcing the 4.1 Core

#

(this is on a Macbook Pro M2 Max)

fervent vale Feb 22, 2025, 8:08 AM

#

I heard that macOS has 2 different codebases. One for OpenGL 2.1 and the other for OpenGL 4.1 Core

rough furnace Feb 22, 2025, 8:08 AM

#

man, pijyoi, I imagine your train of thought is something like "let me add hardware acceleration for PColorMeshItem and improve performance ~1000x ...oh wait let me improve it another 2x" 🤣

rough furnace Feb 22, 2025, 8:09 AM

#

fervent vale I heard that macOS has 2 different codebases. One for OpenGL 2.1 and the other f...

you know, I may actually be able to find this out ... xcode let's me look at library code ... might not to get to it today tho

fervent vale Feb 22, 2025, 8:10 AM

#

Oh wait...

#

PColorMeshItem does have a different codepath for newer OpenGL

#

The flat shading thing

#

A lot less vertices to upload

#

macOS was the only one not using that path by default

#

So at least there's a real benefit now to supporting 4.1 Core

#

There was an example that I made in the github comments that lets you display a large image using PCMI and then waves it around

#

That should be faster on macOS now

fervent vale Feb 22, 2025, 8:34 AM

#

The example is in PR3090

rough furnace Feb 22, 2025, 8:51 AM

#

i'll try it now... (sorry had to help w/ a kid project)

#

#

on master it blows up (expectedly)

QOpenGLShader::compile(Vertex): ERROR: 0:1: '' :  #version required and missing.
ERROR: 0:2: 'attribute' : syntax error: syntax error

*** Problematic Vertex shader source code ***
#define lowp
#define mediump
#define highp
#line 1
attribute vec4 a_pos; void main() { gl_Position = a_pos; }

***
QOpenGLShader::compile(Fragment): ERROR: 0:1: '' :  #version required and missing.

*** Problematic Fragment shader source code ***
#ifdef GL_KHR_blend_equation_advanced
#extension GL_ARB_fragment_coord_conventions : enable
#extension GL_KHR_blend_equation_advanced : enable
#endif
#define lowp
#define mediump
#define highp
#line 1
void main() { gl_FragColor = vec4(1.0); }

#

hmm...probably need a bigger image commenting out the code forcing the Core 4.1 profile doesn't show a difference in performance

rough furnace Feb 22, 2025, 9:19 AM

#

ok, found some larger images to use, sure enough, forcing the Core 4.1 profile yields > 2x improvement.

fervent vale Feb 23, 2025, 1:46 AM

#

I think it should be safe to enable 4.1 Core Profile for macOS in mkQApp

fervent vale Feb 23, 2025, 2:03 AM

#

Actually macOS requires 4.1 Core Profile with Forward Compatibility bit enabled. Not setting DeprecatedFunctions in Qt is equivalent to enabling Forward Compatibility bit

#

Enabling Forward Compatibility means only line widths of 1.0 are supported

fervent vale Feb 23, 2025, 2:25 AM

#

There is an unfiltered call to glLineWidth in PlotCurveItem. That would fail for widths other than 1 if 4.1 Core was enabled for macOS

#

Not sure what kind of failure, since the call is made through QOpenGLFunctions (as opposed to PyOpenGL)

#

PyOpenGL raises an exception for when a GL calls returns an error code

#

I tested it out on the MouseSelection example on Windows, 4.1 Core with no DeprecatedFunctions enabled. So it doesn’t blow up.

#

The user visible effect is that clicking on the curves does not make them thick

rough furnace Feb 23, 2025, 5:53 AM

#

fervent vale Enabling Forward Compatibility means only line widths of 1.0 are supported

Booooooooo

rough furnace Feb 23, 2025, 9:02 AM

#

FINALLY FOUND IT!!!!

Came across this website a little over a year ago but I couldn’t find it each time I searched. It talks about drawing thick lines in webgl but has wonderful images/animations explaining all the issues:

https://sudonull.com/post/69577-Drawing-thick-lines-in-WebGL

SudoNull

Sudo Null - Latest IT News

Programming news, technology, and just useful information

#

Wait this isn’t the site I remembered

#

This is still good…

#

https://blog.scottlogic.com/2019/11/18/drawing-lines-with-webgl.html

Scott Logic

Drawing Lines with WebGL

Connecting points on a chart with a line is one of the most basic uses of a charting library. When drawing that line in WebGL, however, it's not as straightforward as you might think.

#

This is the website: https://wwwtyro.net/2021/10/01/instanced-lines-part-2.html

#

That’s the website I remembered 🤣 (I’m in a waiting area and decided to try and find this)

fervent vale Feb 23, 2025, 9:41 AM

#

Incidentally just last week I managed to draw my first triangle in webgl2

#

Probably might be the lowest barrier to playing with opengl

#

I read that all 3 major browsers support webgl2 since a few years ago

rough furnace Feb 23, 2025, 9:58 AM

#

Yeah

rough furnace Feb 23, 2025, 10:30 AM

#

Yeah webgl2 I would think is safe to assume at this point. Nice thing about browsers is they have a tendency to be updated fairly quickly.

fervent vale Feb 23, 2025, 10:35 AM

#

Would have been a good target: #version 300 es

rough furnace Feb 23, 2025, 10:53 AM

#

Is there a platform we want to support that can’t do ES 3.0?

fervent vale Feb 23, 2025, 10:59 AM

#

macOS

#

If you have OpenGL 4.2, you will have the extension ARB_ES3_compatibility

#

So you can just write es3 shaders

#

It's possible to have the extension while being less than 4.2, but that's not the case for macOS

#

RPI5 is only OpenGL desktop 3.1 but it has OpenGL ES 3.1

#

So while in GLES3 context, it is feature-wise very close to Desktop 4.3

#

Which includes compute shaders

rough furnace Feb 23, 2025, 11:10 AM

#

Ugh was afraid macOS would be the answer

rough furnace Feb 23, 2025, 5:43 PM

#

pyqtgraph/pyqtgraph#3257 ready for merging?

supple leafBOT Feb 23, 2025, 5:43 PM

#

GitHub

PROpen [pyqtgraph] #3257 Make PlotCurveItem and PColorMeshItem work in Core Profile

fervent vale Feb 23, 2025, 7:09 PM

#

rough furnace pyqtgraph/pyqtgraph#3257 ready for merging?

Yes

rough furnace Feb 23, 2025, 7:16 PM

#

back under 50 open PRs 😅

rough furnace Feb 23, 2025, 7:51 PM

#

@torn coyote do you have an issue with this PR?

https://github.com/pyqtgraph/pyqtgraph/pull/3240

these context menu options, they're all the rage it would seem

GitHub

Plot option to subtract mean value from all samples by wetzeld · Pu...

Hello Team,
this PR introduces a new plot transformation that subtracts the mean value of a dataset from all samples.
This makes plotting / analyzing two signals with different offsets much easier ...

fervent vale Feb 24, 2025, 2:51 AM

#

Managed to trigger a new bug by setting pxMode=True in the GLGraphItem example. The parent-child relationship in GLGraphicItem has always been problematic

#

It was papered over in the original implementation of GLGraphItem by making GLScatterPlotItem "know" that it could be used as a child

#

That bit of seemingly innocuous workaround code got removed in the recent PRs.

#

Will fix it properly (hopefully!) later

rough furnace Feb 24, 2025, 3:24 AM

#

decided to brush the dust off off of my branch to numpydoc-ify some of our docs (I was working on this ~8 months ago) ...picking up something that you set down so long ago, you instantly remember stuff you hated about it. In this case, the extensive use of kwargs ... I finally figured out the best way to have rendered output when you want to document a whole bunch of kwargs but ...wow, I hate it 🙃

rough furnace Feb 24, 2025, 3:44 AM

#

to get a decent list of supported args and their explanation, I have to embed a table into the kwargs parameters. The rendered output...isn't great, but it was the best I could do.

fervent vale Feb 24, 2025, 10:37 AM

#

Finally figured out a random-ness issue: GLGraphicsItem stores its children in an unordered set

#

If you run the GLViewWidget example multiple times, the rendered output can be different from run to run

rough furnace Feb 24, 2025, 11:49 AM

#

oof, that must have been annoying to debug

#

get AxisItem in docs to build w/o errors/warning, run numpylint, another 58 issues 🙃

fervent vale Feb 24, 2025, 12:00 PM

#

Ok, the fixes are in

#

More like puzzling

#

I noticed it many months ago

#

But just brushed it off as a driver issue

rough furnace Feb 24, 2025, 2:18 PM

#

fervent vale Ok, the fixes are in

love the usage of the walrus operator

fervent vale Feb 24, 2025, 2:22 PM

#

The GLGraphItem example looks more 3D now that node size is scaled to distance

#

GLScatterItem and GLVolumeItem are the 2 items that need some camera parameters from GLViewWidget

#

fov and width sounds like something that could have been extracted from the modelView transform

rough furnace Feb 24, 2025, 2:59 PM

#

yeah I would think it does too

rough furnace Feb 24, 2025, 3:51 PM

#

Just saw this library referenced on reddit, author in the comments says he sees his library as a successor to pyqtgraph (shots fired!) https://fastplotlib.org/ver/dev/index.html

rough furnace Feb 24, 2025, 8:23 PM

#

"it's done" /meme https://github.com/pyqtgraph/pyqtgraph/pull/3261

GitHub

Convert PlotItem and AxisItem to use numpydoc Docstring Style by j9...

Convert the docstrings of PlotItem.py and AxisItem.py to conform to numpydoc codestyle.
Due to the extensive use of *args and **kwargs, the best way I could figure out to render the supported optio...

#

if I could never edit a docstring again, that would be wonderful.

#

oh wait, I blew up the test suite through one of my "minor" changes 🤦‍♂️

fervent vale Feb 25, 2025, 1:34 AM

#

Ugh, messed up the commits. A change for the set-to-list for GLGraphicsItem went to the wrong commit

rough furnace Feb 25, 2025, 3:56 AM

#

Saw that, it’s fine

#

If it’s going to bother you, I can drop those commits from master

fervent vale Feb 25, 2025, 4:36 AM

#

Got to stick to the no-rewriting-public-git-history rule

rough furnace Feb 25, 2025, 6:48 AM

#

new dependency support timeline just dropped: https://scientific-python.org/specs/spec-0000/ (and by just, I mean I just noticed it, not sure how long ago it was released)

SPEC 0 — Minimum Supported Dependencies

Description# This SPEC recommends that all projects across the Scientific Python ecosystem adopt a common time-based policy for dropping dependencies. From the perspective of this SPEC, the dependencies in question are core packages as well as older Python versions.
All versions refer to feature releases (i.e., Python 3.8.0, NumPy 1.19.0; not Py...

#

per this spec, we should drop support for Python 3.10 🎊

fervent vale Feb 25, 2025, 7:24 AM

#

We don't even have a version released for 3.10

rough furnace Feb 25, 2025, 7:26 AM

#

fervent vale We don't even have a version released for 3.10

last release was 3.10+ ?

#

If we go to 3.11+ I think it means we drop support for PySide2?

fervent vale Feb 25, 2025, 7:29 AM

#

0.13.7 was for py39

rough furnace Feb 25, 2025, 7:30 AM

#

was it? per the tag on the repo, the README shows 3.10+ https://github.com/pyqtgraph/pyqtgraph/tree/pyqtgraph-0.13.7

fervent vale Feb 25, 2025, 7:30 AM

#

I am looking at the git history

#

Immediately after the 0.13.7 tag, pyproject.toml was bumped to 310

rough furnace Feb 25, 2025, 7:32 AM

#

huh, must have jumped the gun w/ the README or something, doesn't really matter, I think the library worked ...but I am excited to drop support for PySide2

fervent vale Feb 25, 2025, 7:32 AM

#

There was some churn between 0.13.4 and 0.13.7

rough furnace Feb 25, 2025, 7:32 AM

#

yeah some regressions made it in, wonder who caused those 🙃

#

certainly appreciate the diplomatic phrasing!

fervent vale Feb 25, 2025, 7:52 AM

#

If you still follow NEP 29, one last release for 3.10 can be made by 2025-04

#

Then you switch policy to SPEC 0

rough furnace Feb 25, 2025, 7:53 AM

#

I'm fine w/ that, are you stuck on 3.10 for some reason? ... I remember you having to support a much older version of python for a project

fervent vale Feb 25, 2025, 7:53 AM

#

Not really

#

Ubuntu 22.04's system python is 3.10

#

Debian bookworm system python is 3.11

#

SPEC 0 wants to drop 3.11 on 2025-Q4