Fetch dependencies using Ivy

[tresf]

Removes dependant libraries from the source repository and fetches them from Maven Central using ivy.

Closes #1387 #1253

[tresf]

Once mainline has received the new Headless support, we can move to Ivy for good.

This has progressed since opening this PR: #1225 (comment). Since "Headless" requires Java 21 or higher, this PR must be merged with #1222 in order to test the headless libraries.

[tresf]

This PR is in surprisingly good shape... One outstanding issue is that the JavaFX version that we bundle is currently broken for vector (rasterize: false) prints. We're tracking this with the FX team in SUPQZ-20. We'll need a patch to JavaFX before we can continue testing.

tray/ant/ivy-deps.properties

Line 16 in f0cbadf

javafx.version=24-headless+0-2024-12-02-101029

[tresf]

We're tracking this with the FX team in SUPQZ-20. We'll need a patch to JavaFX before we can continue testing.

The FX team has provided a patch for vector HTML prints, bumping the version and re-testing.

[tresf]

We're tracking this with the FX team in SUPQZ-20. We'll need a patch to JavaFX before we can continue testing.

The FX team has provided a patch for vector HTML prints, bumping the version and re-testing.

So vector prints are fixed, but I've found a bug with raster prints. From what I can tell this issue existed in the last FX snapshot version and I missed it. @akberenz and I spent an hour or so on it, but something with setZoom seems to cause the height on our rasterized-fitted tests to render way too tall. Our tests didn't catch this because the snapshot is actually succeeding, just at an unusual height. Something with the page height calculation is returning values that appear to be using zoom levels from previous snapshots.

[tresf]

We're tracking this with the FX team in SUPQZ-20. We'll need a patch to JavaFX before we can continue testing.

The FX team has provided a patch for vector HTML prints, bumping the version and re-testing.

So vector prints are fixed, but I've found a bug with raster prints. From what I can tell this issue existed in the last FX snapshot version and I missed it. @akberenz and I spent an hour or so on it, but something with setZoom seems to cause the height on our rasterized-fitted tests to render way too tall. Our tests didn't catch this because the snapshot is actually succeeding, just at an unusual height. Something with the page height calculation is returning values that appear to be using zoom levels from previous snapshots.

I was able to reproduce this bug on master branch, so this is no longer a blocker; testing is ready to continue.

[tresf]

Tested macOS, Windows 11, Ubuntu 24.04.

No regressions found. Now we wait for a JDK21 JavaFX "Headless" backport.

[tresf]

Tested macOS, Windows 11, Ubuntu 24.04.

No regressions found. Now we wait for a JDK21 JavaFX "Headless" backport.

Update from the JavaFX team:

Hi Tres, the Headless platform is now merged, and it will be part of JavaFX 26. It goes a bit against the spirit of LTS releases to backport this to 17/21, but there will be little to no technical issues to backport this to 17/21, and even 11 should be doable, I believe. Might be better that we do this in specific versions though. Almost all code for the headless platform is new code in a separate package, so there won't be merge conflicts. The only changes that need to be applied when backporting are due to API changes in parent or related classes.

... so now we await an LTS release.

[tresf]

Slight update... JDK25 LTS has been released! This means we can update to JavaFX 26-ea. Sadly, testing is failing for vector HTML prints (using the maven mirrors). I've reached out to Gluon for help.

[tresf]

Slight update... JDK25 LTS has been released! This means we can update to JavaFX 26-ea. Sadly, testing is failing for vector HTML prints (using the maven mirrors). I've reached out to Gluon for help.

Filed upstream bug report SUPQZ-23. Meanwhile, I've bumped this branch to use the last-working-snapshot version so we can proceed with this branch.

[tresf]

Slight update... JDK25 LTS has been released! This means we can update to JavaFX 26-ea. Sadly, testing is failing for vector HTML prints (using the maven mirrors). I've reached out to Gluon for help.

Filed upstream bug report SUPQZ-23. Meanwhile, I've bumped this branch to use the last-working-snapshot version so we can proceed with this branch.

Upstream says this is fixed, will bump and retest.

The latest releases of JavaFX 26+ea are passing. Looks like ivy branch will soon see the light of day. 🤞

[tresf]

@Vzor- wrote us some unit tests using TestNG that will fail if we have as much as a single-pixel difference between two raw tests by comparing PDF, image and HTML printing to some baselines. These changes are now live on master branch. I've cherry-picked these to the ivy branch and noticed the following:

• ✅ Image tests pass
• ✅ PDF tests pass
• 🚫 HTML tests fails

Further inspection shows some subtle differences in the way fonts and sub-pixel edges are rendered between JavaFX 19 and JavaFX 26 (pictured below). These are sane. I'm updating the baselines here to match those of JavaFX 26, which will fix the CI.

Before	After	Diff

[tresf]

Windows tests are still failing... Weird.

[tresf]

Hmm... CI is failing but from what I can tell, there's something wrong with the JDK version in the runner. I've opened a dedicated issue with the setup-java@v5 repo here: actions/setup-java#997