Strange Behavior with PyDMImageView and PVA Image

[slactjohnson]

Describe the bug Images accessed using PVA appear to get transposed somehow as compared to CA images. See below with our older CA image viewer on the right, and a small test panel, with the image in question on the right hand side.

I was unable to set the width manually, and setting the readingOrder appeared to have no effect.

More details are in the Slack thread: https://slac.slack.com/archives/C6B8GRCPQ/p1716918965522069

Expected behavior Ideally the PVA and CA images would be identical, with the widget picking up the necessary metadata from the PVA image directly

My Platform rhel7-x86_64 pcds-envs 5.8.4

[nstelter-slac]

Hi,

So it seems like Pydm is interpreting the image dims in a different order then the request expects.

We are creating the image based on this incoming data (output below is from pvget)

    dimension_t[] dimension
        dimension_t 
            int size 640
            int offset 8
            int fullSize 640
            int binning 2
            boolean reverse false
        dimension_t 
            int size 480
            int offset 3
            int fullSize 480
            int binning 2
            boolean reverse false

We do this by making a numpy shape array 'shape = [dimension[0], dimension[1]]', and then this ordering gets interpreted in numpy as the image having 'height = shape[0], width = shape[1]' (https://github.com/slaclab/pydm/blob/master/pydm/data_plugins/epics_plugins/pva_codec.py)

This is why doing a 'img = img.reshape(480, 640)' on the image-data in Pydm displays the correct image.

Question is: in these requests will first element of dimension_t[] always be 'width' and the second 'height'? (can't find any reference for this in the docs here: https://github.com/epics-base/normativeTypesCPP/wiki/Normative+Types+Specification#ntndarray) If so this can be fixed on Pydm's side.

[slactjohnson]

@nstelter-slac and I discussed this on Slack, but wanted to document it here that we think the way to proceed is to make the dimension ordering a configurable parameter so that the widget can be adapted to accommodate either case.