draw/test_transform.py, draw/test_visibility.py test fail on local Debian install

[arikrupnik]

Tests are known to pass for sure at least on Ubuntu (on CI), Fedora (for the packaging), Gentoo and Arch (on my computers). We can try on a fresh Debian installation in a container.

If you’re interested in finding a solution, could you please open a separate issue? It would be useful to share the failing results in tests/draw/results folder.

Originally posted by @liZe in #2553

On a fresh pull, 8 unit tests are failing. I am running

Debian testing
Python 3.13.7
GPL Ghostscript 10.06.0 (2025-09-09)

================================================ short test summary info =================================================
FAILED tests/draw/test_column.py::test_column_rule_normal - Failed: Pixel (0, 2) in column_rule_normal: expected rgba(255, 255, 255), got rgba(0, 0, 254)
FAILED tests/draw/test_transform.py::test_2d_transform_1 - Failed: Pixel (2, 1) in 2d_transform_1: expected rgba(255, 255, 255), got rgba(0, 0, 255)
FAILED tests/draw/test_transform.py::test_2d_transform_2 - Failed: Pixel (5, 1) in 2d_transform_2: expected rgba(255, 255, 255), got rgba(0, 0, 255)
FAILED tests/draw/test_transform.py::test_2d_transform_3 - Failed: Pixel (2, 4) in 2d_transform_3: expected rgba(255, 255, 255), got rgba(0, 0, 255)
FAILED tests/draw/test_transform.py::test_2d_transform_4 - Failed: Pixel (2, 4) in 2d_transform_4: expected rgba(255, 255, 255), got rgba(0, 0, 255)
FAILED tests/draw/test_visibility.py::test_visibility_1 - Failed: Pixel (6, 1) in visibility_1: expected rgba(255, 0, 0), got rgba(238, 0, 17)
FAILED tests/draw/test_visibility.py::test_visibility_3 - Failed: Pixel (6, 1) in visibility_3: expected rgba(255, 0, 0), got rgba(238, 0, 17)
FAILED tests/draw/test_visibility.py::test_visibility_4 - Failed: Pixel (6, 1) in visibility_4: expected rgba(255, 0, 0), got rgba(238, 0, 17)
====================================== 8 failed, 3783 passed, 39 xfailed in 55.79s =======================================

Images from tests/draw/results/:

[liZe]

Thanks for the report.

Could you please share them in an archive? It’ll be easier with filenames. 😄

[arikrupnik]

With pleasure
draw-results.zip

[liZe]

Hi!

I launched tests on a fresh Debian stable (13) ISO, there was no error.

Do you know what could be different in testing?

[arikrupnik]

I am at a loss. Venv is supposed to make sure all the libraries are at the versions you specify, and GS is at the right version. I am at a loss.

My environment is not a fresh install. I've been continuously updating the unstable for some years now. Occasionally, things get out of whack on a long-running installation like that. But even so, venv is supposed to to take care of that.

[liZe]

Venv is supposed to make sure all the libraries are at the versions you specify

Yes, it takes care of Python libraries. But other libraries and tools (Pango, Harfbuzz, Fontconfig, Ghostscript) may differ.

The problem to reproduce tests is usually fonts, but the tests failing for you look more related to rounding differences in Ghostscript, not fonts. I’ll take a look and try to find what’s wrong, but I’m not sure to find an answer.

[liZe]

@arikrupnik I get the same errors as you with a Debian testing ISO, I’ll be able to check what’s wrong. 😄

[liZe]

The PDF generated on the Debian ISO and on my computer are the same, so the difference is in Ghostscript. But what is this difference?…

[liZe]

I tried my best to understand what’s different, but didn’t find. I even recompiled my Ghostscript with Debian patches…

To be honest, I don’t think that I’ll spend more time on this, I spent too many hours already. 😄

You can try to install older versions of Ghostscript and see if it changes something. I suspect that the Debian-only patches cause this problem. You can also ask the Ghostscript package maintainer: something definitely changed between stable and testing, and the Debian devs will probably be happy not to break the package behavior.

Here’s an example:

tmpcuprnj0d.pdf

With this command:

gs -q -sDEVICE=png16m -dTextAlphaBits=1 -dGraphicsAlphaBits=1 -dBATCH -dNOPAUSE -dPDFSTOPONERROR -r720.0 -dUsePDFX3Profile -sOutputFile=out.png tmpcuprnj0d.pdf

We’re supposed to get exactly this image (at least we do on Debian stable and everywhere else):

-dGraphicsAlphaBits=1 is supposed to avoid antialiasing for images, but the result on Debian testing seems to be antialiased.

[liZe]

The problem is not in Ghostscript, but more probably in other choices in the default configuration of recent versions of Debian. We’re working on a way to fix that.

[arikrupnik]

Thank you for continuing to work on this.

[liZe]

#2696 is ready for testing! It works for me on a fresh Debian Testing ISO.