Added trusted PyPI publishing #7616

radarhere · 2023-12-14T08:08:21Z

Helps #7390

Removes "Wheels Successful" job. These jobs were added as part of Add Mergify config #4931 for mergify, but this workflow isn't referenced in https://github.com/python-pillow/Pillow/blob/main/.github/mergify.yml
Adds "Upload release to PyPI" step, as per https://github.com/marketplace/actions/pypi-publish. This still requires the steps from https://docs.pypi.org/trusted-publishers/adding-a-publisher/ to be taken before it will work.

.github/workflows/wheels.yml

hugovk · 2023-12-14T12:35:42Z

Very nice!

Looks like:

I've been using trusted publishing in my other projects (although not with environments).

I recommend we also set up trusted publishing to TestPyPI, so we know the whole pipeline is working before the big release day.

I've already registered Pillow at https://test.pypi.org/project/pillow/ and have set up trusted publishers on that side:

One sticking point:

You can't publish a file to [Test]PyPI that has already been published there, unless the version number changes. But we have a hardcoded value in src/PIL/_version.py.

One way is to use https://github.com/pypa/setuptools_scm/ to generate a version number based on the "distance"/number of commits since the last tag. We'd also need local_scheme = "no-local-version", it removes a +abc123 suffix that PyPI doesn't like, and also need some juggling to get the version to the C code. This could be a separate PR.

Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>

radarhere · 2023-12-15T04:23:31Z

You can't publish a file to [Test]PyPI that has already been published there, unless the version number changes. But we have a hardcoded value in src/PIL/_version.py.

To make sure I'm following - this is only a problem with test.pypi.org, because we may want to submit the same version more than once, yes? You're not suggesting that anything about the normal version mechanism needs to be changed.

hugovk · 2023-12-15T10:09:07Z

You can't publish a file to [Test]PyPI that has already been published there, unless the version number changes. But we have a hardcoded value in src/PIL/_version.py.

To make sure I'm following - this is only a problem with test.pypi.org, because we may want to submit the same version more than once, yes?

Yes and no :)

No, both TestPyPI and PyPI behave the same in this regard.

The restriction really is that filenames cannot be re-used. So we can re-use version numbers, but need to

For example, we have a Pillow-10.1.0-cp312-cp312-macosx_10_10_x86_64.whl at https://pypi.org/project/Pillow/10.1.0/#files. If there was something wrong with that file and we needed to upload a new one, we can't upload a file with the same name, PyPI will reject it with an error. That's why we've needed to use wheel build tags in the past, so we have a unique file we can upload:

And so yes, we may submit the same version more than once, but need to make it unique in other ways.

You're not suggesting that anything about the normal version mechanism needs to be changed.

Correct.

For normal releases, we still have the same version numbers.

This would give us new version numbers for testing uploads of intra-release builds to TestPyPI.

If we look at https://github.com/hugovk/norwegianblue (now uses Hatchling+hatch-vcs, but used to use setuptools+setuptools_scm and both work in the same way (I think hatch-vcs uses setuptools_scm under the hood).

PyPI has "normal" release numbers for proper releases: https://pypi.org/project/norwegianblue/#history
TestPyPI has releases appended with .devX: 0.16.1.dev4, 0.16.1.dev6: https://test.pypi.org/project/norwegianblue/#history

You're not suggesting that anything about the normal version mechanism needs to be changed.

Although there would be one change:

Right now, after we release 10.2.0, we bump the hardcoded version to 10.3.0.dev0.

With setuptools_scm, we'd still be getting a version like 10.3.0.dev1, 10.3.0.dev200, etc, because it's based on the last tag.

I think that's fine, it's an internal change.

This does however bring to mind a potential blocker: we do compile the version into the C layer, and error if they're different. With setuptools_scm, the devX is incremented for every commit, which could result in mismatches or needed to rebuild C more often.. This could be a blocker, and we may need another solution to give a unique filename for TestPyPI...

radarhere · 2023-12-15T10:52:34Z

we do compile the version into the C layer, and error if they're different. With setuptools_scm, the devX is incremented for every commit, which could result in mismatches

We could relax the condition to allow 10.3.0.dev1 or 10.3.0.dev200, by changing

Pillow/src/PIL/Image.py

Line 84 in e5ee034

if __version__ != getattr(core, "PILLOW_VERSION", None):

to

    if not hasattr(core, "PILLOW_VERSION") or re.sub(
        "dev[0-9]+", "dev0", __version__
    ) != re.sub("dev[0-9]+", "dev0", core.PILLOW_VERSION):

hugovk · 2023-12-15T11:05:17Z

Yep, that could work.

radarhere · 2023-12-18T08:17:33Z

I've just registered for test.pypi.org with the same username. Would you mind adding me as a maintainer to Pillow?

hugovk · 2023-12-18T08:25:35Z

Invited!

radarhere · 2023-12-18T09:39:06Z

Thanks.

I'm wondering if the use of setuptools-scm to generate unique version numbers each time is overkill? Once we have it working, isn't it unlikely to break in the future? Meaning that for whatever infrequent testing we need to do, we could just branch off from main and manually increment the version to dev1?

hugovk · 2023-12-18T10:34:56Z

If possible, I'd prefer to automate testing where possible rather than do it manually. Doing it manually means we have to foresee a change might cause problems in the release pipeline. And as said, we can consider that in a separate PR, and can consider other approaches too.

hugovk · 2023-12-18T10:43:28Z

.github/workflows/wheels.yml

+      - uses: actions/download-artifact@v4
+      - name: Merge artifacts
+        run: |
+          for artifact in macOS_x86_64 \
+                          macOS_arm64 \
+                          manylinux2014_musllinux \
+                          manylinux_2_28 \
+                          windows_x86 \
+                          windows_x64 \
+                          windows_arm64 \
+                          sdist; do
+            mv $artifact/* .
+            rmdir $artifact
+          done


Let's try something like scikit-build/cmake-python-distributions#438 / scientific-python/cookie#335 to avoid repeating the names:

Suggested change

- uses: actions/download-artifact@v4

- name: Merge artifacts

run: |

for artifact in macOS_x86_64 \

macOS_arm64 \

manylinux2014_musllinux \

manylinux_2_28 \

windows_x86 \

windows_x64 \

windows_arm64 \

sdist; do

mv $artifact/* .

rmdir $artifact

done

- uses: actions/download-artifact@v4

with:

path: all

- name: Merge files

run: |

mkdir dist

mv all/*/* dist

Ok, I've pushed something similar.

.github/workflows/wheels.yml

hugovk · 2023-12-18T10:54:51Z

.github/workflows/wheels.yml

        with:
-          name: dist
+          name: ${{ matrix.artifact_name }}


Can we re-use existing matrix variables instead of introducing a new artifact_name?

The tricky bit might be differentiating between these two:

- name: "manylinux2014 and musllinux x86_64" os: ubuntu-latest archs: x86_64 artifact_name: "manylinux2014_musllinux" - name: "manylinux_2_28 x86_64" os: ubuntu-latest archs: x86_64 build: "*manylinux*" manylinux: "manylinux_2_28" artifact_name: "manylinux_2_28"

If we do wheel-${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}, does that error or do we get something like this?

wheel-ubuntu-latest-x86_64

wheel-ubuntu-latest-x86_64-manylinux_2_28

I've pushed a commit to use ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}.

We get

macos-latest-x86_64-

macos-latest-arm64-

ubuntu-latest-x86_64-

ubuntu-latest-x86_64-manylinux_2_28

Thanks!

Since my last suggestion, and based on lots of feedback about the breaking @v4 changes, GitHub have just added pattern: and merge-multiple: inputs:

actions/upload-artifact#472 (comment)

Shall we try that? We'll need a common prefix/pattern for the sdist and all the wheel artifacts.

Thanks for finding that. I've updated the PR.

.github/workflows/wheels.yml

webknjaz · 2023-12-19T14:43:51Z

Removes "Wheels Successful" job. These jobs were added as part of Add Mergify config #4931 for mergify, but this workflow isn't referenced in https://github.com/python-pillow/Pillow/blob/main/.github/mergify.yml

Moreover, the "gate" was implemented incorrectly: https://github.com/marketplace/actions/alls-green#why.

webknjaz · 2023-12-19T14:50:28Z

.github/workflows/wheels.yml

        with:
-          name: fribidi
+          name: fribidi_${{ matrix.arch }}
          path: fribidi\*

  sdist:


May I suggest building wheels from sdist? This is what pip does — it doesn't build wheels straight from Git source, but from the tarballs. And this is what pypa/build does if you don't pass any args. Doing so will ensure that sdists are actually usable, as a smoke test.

Here's an example: https://github.com/aio-libs/yarl/actions/runs/7258064074/workflow#L124.

P.S. cibuildwheel accepts a path to the sdist tarball natively (in place of a directory path) so integrating this into the wheel build jobs is as easy as downloading the artifact and feeding the sdist from it into the cibuildwheel action.

Oh, and that repo also features the wheel build jobs moved into a reusable workflow and called with different params.

May I suggest building wheels from sdist?

We're currently using the .ci directory in the build process, which isn't included in the sdist

Pillow/.github/workflows/wheels.yml

Lines 65 to 67 in 41e45b5

- name: Build wheels

run: |

python3 -m pip install -r .ci/requirements-cibw.txt

Sure, we could checkout the repository AND use sdist - but that doesn't sound very neat to me.

I know you said that cibuildwheel can simply accept an sdist path, but we're also currently adding in extra test images, which wouldn't be quite so simple when using a compressed sdist.

Pillow/.github/workflows/wheels.yml

Lines 99 to 103 in 41e45b5

- name: Checkout extra test images

uses: actions/checkout@v4

with:

repository: python-pillow/test-images

path: Tests\test-images

Oh, and that repo also features the wheel build jobs moved into a reusable workflow

If wheels.yml gets much longer, I could see that being nice, but at the moment, I don't think it's problematic.

Also, neither of these ideas depend on this PR. If you would like to see them become part of Pillow, you could create a separate PR.

We're currently using the .ci directory in the build process, which isn't included in the sdist

And this is a reason to include it. Downstreams might want to run tests using the sdist as the source.

If wheels.yml gets much longer, I could see that being nice, but at the moment, I don't think it's problematic.

For me, it's a structural thing that helps layer components. Not necessarily related to how big the YAML files are (although, that's also important).

Also, neither of these ideas depend on this PR. If you would like to see them become part of Pillow, you could create a separate PR.

Sure. These are mostly drive-by comments I made having seen some familiar faces and tech involved.

nulano · 2023-12-19T14:53:50Z

Moreover, the "gate" was implemented incorrectly: https://github.com/marketplace/actions/alls-green#why.

The argument being that a "skipped" status is incorrectly accepted, yes?

This does not appear to be the case for Mergify, which is the only use of these gates here. A recent PR, #7608, only updated documentation and so most of the gates were skipped. However, the Mergify report did not accept these skips as a success: https://github.com/python-pillow/Pillow/runs/19444375622

webknjaz · 2023-12-19T14:59:10Z

Yeah, you're right. Mergify treats these statuses better. But overall, on the GH platform level, it might still be reasonable to implement them correctly so that somebody doesn't trip over this implementation quirk in the future.

nulano · 2023-12-21T02:00:30Z

.github/workflows/wheels.yml

        with:
-          name: dist
+          name: ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}


Suggested change

name: ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}

name: ${{ matrix.os }}-${{ matrix.archs }}${{ matrix.manylinux && '-' }}${{ matrix.manylinux }}

Would get rid of the extra trailing -; not sure if it's worth sacrificing a bit of readability to make the output names cleaner?

Could also be something like

Suggested change

name: ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}

name: >-

${{

format(

'{0}-{1}{2}{3}',

matrix.os,

matrix.archs,

(matrix.manylinux && '-' || ''),

matrix.manylinux

)

}}

(not tested)

My preference would be in the middle, with

${{ matrix.os }}-${{ matrix.archs }}${{ matrix.manylinux && format('-{0}', matrix.manylinux) }}

but I have a feeling that would make no one happy.

My preference would be in the middle, with [...] but I have a feeling that would make no one happy.

Actually, that looks like the best option to me also.

Cool, thanks, I've pushed a commit for that.

Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>

hugovk · 2023-12-25T22:44:21Z

Notice: Attempting to perform trusted publishing exchange to retrieve a temporary short-lived API token for authentication against https://test.pypi.org/legacy/ due to __token__ username with no supplied password field
Error: Trusted publishing exchange failure: 
OpenID Connect token retrieval failed: GitHub: missing or insufficient OIDC token permissions, the ACTIONS_ID_TOKEN_REQUEST_TOKEN environment variable was unset

This generally indicates a workflow configuration error, such as insufficient
permissions. Make sure that your workflow has `id-token: write` configured
at the job level, e.g.:

permissions:
  id-token: write

Learn more at https://docs.github.com/en/actions/deployment/security-hardening-your-deployments/about-security-hardening-with-openid-connect#adding-permissions-settings.

https://github.com/python-pillow/Pillow/actions/runs/7324548352/job/19948625585

But we have that at the job level. Maybe it's because we have an env configured at TestPyPI. I'll remove that, and restart the job.

hugovk · 2023-12-25T22:45:48Z

Added the second one, and left the first one:

And restarted...

hugovk · 2023-12-25T22:46:52Z

Hmm, same thing. Removed the first one and restarted...

hugovk · 2023-12-25T22:48:25Z

Nope, same thing. Perhaps it's as @webknjaz said in one thread, you can't publish from a PR.

nulano · 2023-12-25T23:56:32Z

PRs don't have access to most secrets (otherwise they could be trivially extracted by a malicious PR). From https://docs.github.com/en/actions/security-guides/using-secrets-in-github-actions#using-secrets-in-a-workflow:

With the exception of GITHUB_TOKEN, secrets are not passed to the runner when a workflow is triggered from a forked repository.

And GITHUB_TOKEN cannot have write permissions by default. From https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#changing-the-permissions-in-a-forked-repository:

You can use the permissions key to add and remove read permissions for forked repositories, but typically you can't grant write access. The exception to this behavior is where an admin user has selected the Send write tokens to workflows from pull requests option in the GitHub Actions settings. For more information, see "Managing GitHub Actions settings for a repository."

If I understand this correctly, the required setting can only be enabled for private repositories (where only members of the organization can create PRs).

webknjaz · 2023-12-26T00:00:22Z

Nope, same thing. Perhaps it's as @webknjaz said in one thread, you can't publish from a PR.

Yes, PR privileges are lower. At least, when the PRs are coming from forks. Some of them can be addressed through pull_request_target. But that's its own can of worms.

webknjaz · 2023-12-26T00:02:01Z

Although the publish action uses GITHUB_TOKEN, that token cannot have write permissions by default.

No, my action doesn't access that token. GHA sets two special env vars for OIDC when id: write is set and is allowed in the given context.

webknjaz · 2023-12-26T00:04:34Z

Hmm, same thing. Removed the first one and restarted...

I recommend keeping the environment. Just make sure it matches what you set up on the PyPI side. I suggest using pypi for PyPI and testpypi for TestPyPI in my guides and tutorials. These represent the targets rather accurately.

webknjaz · 2023-12-26T00:06:32Z

Perhaps we should spell it out in the error message that GH does allow OIDC in untrusted event contexts, like pull_request.

cc @woodruffw

nulano · 2023-12-26T00:22:22Z

No, my action doesn't access that token. GHA sets two special env vars for OIDC when id: write is set and is allowed in the given context.

That's what I thought, but I got confused by the GHA documentation.
Not that it matters, but I guess those envvars are automatically extracted(?) from the GITHUB_TOKEN by the runner?

woodruffw · 2023-12-26T00:34:36Z

Just to provide some disambiguation:

Trusted publishing allows publishing from a pull_request event. More generally: as long as a workflow has id-token: write, trusted publishing will work from it.
GitHub itself does not give id-token: write to pull_request events run from forks, even if the workflow explicitly adds that permission. This isn't super well documented anywhere, but makes sense from a security perspective (I believe this is the same for all other write permissions, to prevent 3p forks from making destructive changes to the "target" repo by default).

So TL;DR: you can publish to PyPI from a pull_request event, but only if the pull_request event is coming from a first-party branch (not from a fork). This limitation persists even if the fork is controlled or pushed to by an owner of the upstream repository.

Edit: For some additional context on this GitHub behavior, see sigstore/sigstore-conformance#55 -- it's a different project, but the outcome (can't get OIDC tokens from 3p PRs) is the same.

woodruffw · 2023-12-26T00:39:06Z

Perhaps we should spell it out in the error message that GH does allow OIDC in untrusted event contexts, like pull_request.

IMO this is likely to cause some confusion, although I think there's a variant we can chip out per #7616 (comment): if the publish is running in a pull_request event and we detect that it's a branch from a fork, then we can produce a more useful error message, like:

This error indicates that you're attempting to obtain an OpenID Connect token from a fork of a repository, which GitHub does not permit.

<link to another troubleshooting section>

hugovk · 2023-12-26T10:20:20Z

Thanks for clearing that up! That's all fine, we don't want to deploy from PRs in normal use, it was for testing.

So I think we can merge this pretty much as-is, test from main, then cleanup TestPyPI -> PyPI later.

Another option: @radarhere, if you like, we could try things out from your fork. Feel free to add a new publisher for that at: https://test.pypi.org/manage/project/pillow/settings/publishing/

Our next quarterly release is in just a week, 2nd January, so I'd prefer to just get something merged and working first to minimise risk. This will also be the first release with cibuildwheel, plus replacing setup.cfg with pyproject.toml, so we have a few big changes.

radarhere · 2023-12-26T13:42:31Z

Ok, sure. I added a publisher, and merged this to my main.

There was already a partial release for 10.2.0.dev0, so I updated the version to 10.2.0.dev1 and it worked - https://github.com/radarhere/Pillow/deployments / https://test.pypi.org/manage/project/pillow/release/10.2.0.dev1/

hugovk · 2023-12-26T15:53:45Z

Great! Did you test environments?

Let's merge this, and can you arrange with a repo admin to set up what is needed in this repo, and on PyPI? Or for them to give you and me access to do it.

.github/workflows/wheels.yml

Co-authored-by: Ondrej Baranovič <nulano@nulano.eu>

radarhere · 2023-12-26T21:07:57Z

I tested it with an environment, yes.

https://github.com/radarhere/Pillow/blob/9078f85a55090a7c4bca6c304b4b0d480de72f13/.github/workflows/wheels.yml#L187-L207

  pypi-publish:
    # TEMP for testing
    # if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags')
    needs: [build, windows, sdist]
    runs-on: ubuntu-latest
    name: Upload release to PyPI
    permissions:
      id-token: write
    environment:
      name: radarhere-test-pypi
      url: https://test.pypi.org/p/Pillow
    steps:
      - uses: actions/download-artifact@v4
        with:
          pattern: dist-*
          path: dist
          merge-multiple: true
      - name: Publish to PyPI
        uses: pypa/gh-action-pypi-publish@release/v1
        with:
          repository-url: https://test.pypi.org/legacy/

This reverts commit fc1cf9f.

webknjaz · 2023-12-31T00:32:44Z

@woodruffw that sounds reasonable!

This specializes the token retrieval error handling, providing an alternative error message when the error cause is something that we know can't possibly work due to GitHub's own restrictions on PRs from forks. PR #203 Closes #202 Ref python-pillow/Pillow#7616 Co-authored-by: Sviatoslav Sydorenko <webknjaz@redhat.com>

radarhere added 2 commits December 14, 2023 18:57

Removed "Success" job, as this is not used in mergify

cbab3a5

Added trusted PyPI publishing

6693046

hugovk reviewed Dec 14, 2023

View reviewed changes

.github/workflows/wheels.yml Outdated Show resolved Hide resolved

.github/workflows/wheels.yml Outdated Show resolved Hide resolved

hugovk reviewed Dec 14, 2023

View reviewed changes

.github/workflows/wheels.yml Show resolved Hide resolved

hugovk added the Release label Dec 14, 2023

Updated step name

c81beb5

Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>

radarhere mentioned this pull request Dec 15, 2023

Update actions/upload-artifact action to v4 #7618

Closed

1 task

radarhere force-pushed the pypi branch 2 times, most recently from 50d2c74 to 2c62533 Compare December 15, 2023 08:26

hugovk reviewed Dec 18, 2023

View reviewed changes

radarhere force-pushed the pypi branch from 3f61efe to 17a8859 Compare December 19, 2023 10:32

hugovk reviewed Dec 19, 2023

View reviewed changes

.github/workflows/wheels.yml Show resolved Hide resolved

hugovk mentioned this pull request Dec 19, 2023

Use setuptools_scm for dynamic versioning #7627

Draft

webknjaz reviewed Dec 19, 2023

View reviewed changes

nulano reviewed Dec 21, 2023

View reviewed changes

radarhere and others added 2 commits December 24, 2023 13:37

Updated environment name

808f312

Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>

Published to Test PyPI

fc1cf9f

Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>

Merge branch 'main' into pypi

4de1765

radarhere force-pushed the pypi branch from 0052fa4 to 55904ac Compare December 26, 2023 01:38

woodruffw mentioned this pull request Dec 26, 2023

Provide a better troubleshooting message when used from a 3P PR pypa/gh-action-pypi-publish#202

Closed

radarhere force-pushed the pypi branch from 55904ac to 4de1765 Compare December 26, 2023 11:29

nulano reviewed Dec 26, 2023

View reviewed changes

.github/workflows/wheels.yml Outdated Show resolved Hide resolved

Updated Windows artifact name

a66fbb9

Co-authored-by: Ondrej Baranovič <nulano@nulano.eu>

Revert "Published to Test PyPI"

043e13d

This reverts commit fc1cf9f.

radarhere merged commit 265ee32 into python-pillow:main Dec 26, 2023
15 of 16 checks passed

radarhere deleted the pypi branch December 26, 2023 22:37

This was referenced Dec 26, 2023

Automate build and release to PyPI #7390

Closed

Updated RELEASING now that Trusted PyPI publishing is in use #7652

Merged

woodruffw mentioned this pull request Dec 31, 2023

oidc-exchange: specialize error on PRs from forks pypa/gh-action-pypi-publish#203

Merged

	- name: Build wheels
	run: \|
	python3 -m pip install -r .ci/requirements-cibw.txt

	- name: Checkout extra test images
	uses: actions/checkout@v4
	with:
	repository: python-pillow/test-images
	path: Tests\test-images

	name: ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}
	name: ${{ matrix.os }}-${{ matrix.archs }}${{ matrix.manylinux && '-' }}${{ matrix.manylinux }}

-          name: ${{ matrix.os }}-${{ matrix.archs }}-${{ matrix.manylinux }}
+          name: >-
+            ${{
+              format(
+                '{0}-{1}{2}{3}',
+                matrix.os,
+                matrix.archs,
+                (matrix.manylinux && '-' || ''),
+                matrix.manylinux
+              )
+            }}

Added trusted PyPI publishing #7616

Added trusted PyPI publishing #7616

Conversation

radarhere commented Dec 14, 2023

hugovk commented Dec 14, 2023

radarhere commented Dec 15, 2023

hugovk commented Dec 15, 2023

radarhere commented Dec 15, 2023

hugovk commented Dec 15, 2023

radarhere commented Dec 18, 2023

hugovk commented Dec 18, 2023

radarhere commented Dec 18, 2023

hugovk commented Dec 18, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

webknjaz commented Dec 19, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nulano commented Dec 19, 2023

webknjaz commented Dec 19, 2023 • edited by radarhere

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nulano Dec 24, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hugovk commented Dec 25, 2023

hugovk commented Dec 25, 2023

hugovk commented Dec 25, 2023

hugovk commented Dec 25, 2023

nulano commented Dec 25, 2023 • edited

webknjaz commented Dec 26, 2023

webknjaz commented Dec 26, 2023

webknjaz commented Dec 26, 2023

webknjaz commented Dec 26, 2023

nulano commented Dec 26, 2023

woodruffw commented Dec 26, 2023 • edited

woodruffw commented Dec 26, 2023

hugovk commented Dec 26, 2023

radarhere commented Dec 26, 2023 • edited

hugovk commented Dec 26, 2023

radarhere commented Dec 26, 2023

webknjaz commented Dec 31, 2023

webknjaz commented Dec 19, 2023 •

edited by radarhere

nulano Dec 24, 2023 •

edited

nulano commented Dec 25, 2023 •

edited

woodruffw commented Dec 26, 2023 •

edited

radarhere commented Dec 26, 2023 •

edited