8368467: [lworld] Add new flag generation for jimage to support preview mode #1621

david-beaumont · 2025-09-23T19:25:22Z

Adds support for writing preview related flags into jimage files.

Preview mode is complex. It's not nearly as simple as "does something in /modules/xxx/... have an entry in /modules/xxx/META-INF/preview/...".

Specific issues include:

Supporting preview-only resources without forcing a double lookup on everything.
Changing the set of entries in /packages/xxx directories to account for preview only packages in some modules.
Minimising the work done during image reader initialization to only need to process the small number of preview resources (rather than scanning the whole file to look for them).
The new flags added by this code address these issues, but calculating them correctly with only minor adjustments to the existing code was not feasible, it just became a lot larger and very complex.

To address this, a new type (ModuleReference) is introduced to track and then merge information about packages seen in each module. This allows a much simpler inner loop for processing resource paths when building the node tree, combined with a subsequent merging stage to produce the final package information for each module.

Not that since ModuleReference is needed during jimage reading, that class is already present in the previous PR on which this is based, but it starts to be used to calculate the module flags in this PR.

This PR can also adds the ImageReader unit tests for preview mode, which rely on being able to generate jimage files with preview mode flags in.

Compare and review this against #1619.

Progress

Change must not contain extraneous whitespace

Integration blocker

⚠️ Dependency #1619 must be integrated first

Issue

JDK-8368467: [lworld] Add new flag generation for jimage to support preview mode (Sub-task - P3)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/valhalla.git pull/1621/head:pull/1621
$ git checkout pull/1621

Update a local copy of the PR:
$ git checkout pull/1621
$ git pull https://git.openjdk.org/valhalla.git pull/1621/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 1621

View PR using the GUI difftool:
$ git pr show -t 1621

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/valhalla/pull/1621.diff

Using Webrev

Link to Webrev Comment

bridgekeeper · 2025-09-23T19:27:32Z

👋 Welcome back david-beaumont! A progress list of the required criteria for merging this PR into pr/1619 will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

openjdk · 2025-09-23T19:29:18Z

❗ This change is not yet ready to be integrated.
See the Progress checklist in the description for automated requirements.

mlbridge · 2025-09-23T19:32:20Z

Webrevs

david-beaumont · 2025-09-23T20:16:18Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageFileCreator.java

        return resultResources;
    }

+    private static ResourcePool getResourcePool(


Pulled out so the variable holding it is effectively final and can be used in a lambda expression.

david-beaumont · 2025-09-23T20:17:14Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageLocationWriter.java

-               .addAttribute(ATTRIBUTE_OFFSET, contentOffset)
-               .addAttribute(ATTRIBUTE_COMPRESSED, compressedSize)
-               .addAttribute(ATTRIBUTE_UNCOMPRESSED, uncompressedSize);
+                .addAttribute(ATTRIBUTE_MODULE, moduleName)


This got reformatted correctly to 8 indent because of the new entry.

david-beaumont · 2025-09-23T20:18:48Z

test/jdk/tools/jlink/whitebox/jdk.jlink/jdk/tools/jlink/internal/ImageResourcesTreeTest.java

+        assertEquals(PKG_FLAG_IS_PREVIEW_ONLY, emptyRef.flags());
+    }
+
+    @Test


This test needs updating. The order of the entries has changed. Sorry for not spotting sooner.

Actually I had fixed this locally and somehow it hadn't got patched into this PR. Updated now and the test compiles and passes.

david-beaumont · 2025-09-23T20:25:49Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageResourcesTree.java

+            // later be merged with a non-empty reference for the same package.
+            ModuleReference emptyRef = ModuleReference.forEmptyPackage(modName, isPreviewPath);
+
+            // Work down through empty packages to final resource.


This loop and the section after replaces the logic between L191 and L238 in the original code.

Instead of two maps (which would be 4 if we tried to add preview support that way) with near-duplicated logic, there's now just "walk down the path creating empty package nodes until the last segment is reached, then exit and create the resource node in a non-empty package".

The empty/non-empty/preview nature of these packages is all handled when the module references are merged later, but for now they are independent.

david-beaumont · 2025-09-23T20:29:02Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageResourcesTree.java

+            packageToModules.keySet().removeIf(p -> p.isEmpty() || p.equals("META-INF") || p.startsWith("META-INF."));
+            packageToModules.forEach((pkgName, modRefs) -> {
+                // Merge multiple refs for the same module.
+                List<ModuleReference> pkgModules = modRefs.stream()


This is where module references are merged so each package has the right set of flags according to all the resources visited in the module (empty, non-empty, preview, preview-only...). This replaces the use of the two maps (moduleToPackage and packageToModule) and a bunch of other logic.

david-beaumont · 2025-09-23T20:30:14Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageResourcesTree.java

+                List<ModuleReference> refs = pkgNode.getModuleReferences();
+                ByteBuffer byteBuffer = ByteBuffer.allocate(8 * refs.size());
+                byteBuffer.order(writer.getByteOrder());
+                ModuleReference.write(refs, byteBuffer.asIntBuffer(), writer::addString);


Call into the ModuleReference code added in the previous PR to write the package flags and offsets.

david-beaumont · 2025-09-23T20:34:28Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageResourcesTree.java

+     * <p>While there are 8 combinations of these 3 flags, some will never
+     * occur (e.g. {@code HAS_NORMAL_CONTENT + IS_PREVIEW_ONLY}).
+     *
+     * <p>Package node entries are sorted by name, with the exception that (if


This is out of date (along with the test). It's now sorted "preview first" and then "by name". The reason for the change is that I had made an incorrect assumption that I didn't need to process empty packages from preview resources, but I do (they still appear in /packages/xxx directories).

* Remove TODOs now jimage version is bumped * jimage writer changes to support preview mode.

openjdk · 2025-09-23T23:04:15Z

@david-beaumont Please do not rebase or force-push to an active PR as it invalidates existing review comments. Note for future reference, the bots always squash all changes into a single commit automatically as part of the integration. See OpenJDK Developers’ Guide for more information.

RogerRiggs · 2025-09-26T01:40:47Z

src/java.base/share/classes/jdk/internal/jimage/ImageLocation.java

-                // TODO: Uncomment the FLAGS_IS_PACKAGE_ROOT test below.
-                // return (getFlags() & FLAGS_IS_PACKAGE_ROOT) != 0
-                return getBase().charAt(1) == 'p'
+                return (getFlags() & FLAGS_IS_PACKAGE_ROOT) != 0


Introducing a flag bit that is carried around in lots of entries but applies to only one is overkill.
The simple test above for the first letter is cleaner and very localized.

RogerRiggs · 2025-09-26T19:30:11Z

src/java.base/share/classes/jdk/internal/jimage/PreviewMode.java

+            if (!ENABLE_PREVIEW_MODE) {
+                return false;
+            }


As commented in the 1618 PR, the extra subclasses caused by the overrides can be replaced by the resolve method switching on ordinal().

RogerRiggs · 2025-09-26T19:35:07Z

src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImageLocationWriter.java

-            ImageStringsWriter strings,
-            long contentOffset, long compressedSize, long uncompressedSize) {
+                                           ImageStringsWriter strings,
+                                           long contentOffset, long compressedSize, long uncompressedSize, int previewFlags) {


Wrap or untab to avoid long lines.

RogerRiggs · 2025-09-26T20:26:52Z

test/jdk/jdk/internal/jimage/ImageReaderTest.java

            "/modules/modfoo/com/foo/bar"})
    public void testModuleDirectories_expected(String name) throws IOException {
-        try (ImageReader reader = ImageReader.open(image)) {
+        try (ImageReader reader = ImageReader.open(image, PreviewMode.DISABLED)) {


The image variable might be better documented in upper case to make it easier to see its a pre-computed value.

david-beaumont changed the base branch from lworld to pr/1619 September 23, 2025 19:25

david-beaumont mentioned this pull request Sep 23, 2025

8368475: [lworld] Add preview classes to jimage at make time #1622

Open

1 task

openjdk bot added the rfr Pull request is ready for review label Sep 23, 2025

david-beaumont commented Sep 23, 2025

View reviewed changes

david-beaumont added 4 commits September 24, 2025 00:51

Manually deleting ImageReaderFactory (it returned somehow)

936b22d

jimage writer changes to support preview mode.

8858134

* Remove TODOs now jimage version is bumped * jimage writer changes to support preview mode.

add system property guard to preview mode

d009eaa

Restoring lost changes and updating some comments.

b192c84

david-beaumont force-pushed the jdk_8368467_reader branch from 0c0ad6a to b192c84 Compare September 23, 2025 23:02

RogerRiggs reviewed Sep 29, 2025

View reviewed changes

8368467: [lworld] Add new flag generation for jimage to support preview mode #1621

Are you sure you want to change the base?

8368467: [lworld] Add new flag generation for jimage to support preview mode #1621

Uh oh!

Conversation

david-beaumont commented Sep 23, 2025 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Integration blocker

Issue

Reviewing

Uh oh!

bridgekeeper bot commented Sep 23, 2025

Uh oh!

openjdk bot commented Sep 23, 2025

Uh oh!

mlbridge bot commented Sep 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

openjdk bot commented Sep 23, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

david-beaumont commented Sep 23, 2025 •

edited by openjdk bot

Loading

mlbridge bot commented Sep 23, 2025 •

edited

Loading