OpenWebStart/Java 17 migration

[Don-vip]

Things are starting to take shape with what comes after Java WebStart (see #16047):

​https://openwebstart.com/

Original 2019 plan:

On the current roadmap, the first version will be released end of October. A macro planning for a Java 8 => Java 11 transitions for all JOSM users would roughly look like this:

• September 2019: we start testing OpenWebStart on all platforms. Likely we'll found a lot of bugs
• November 2019 : first OpenWebStart version. Unlikely to fix all bugs we'll find
• Somewhere in 2020: OpenWebStart version without any bug impacting us, we start asking everyone to switch
• End of 2020: End of Java WebStart support by Oracle for Java 8. We force everyone to switch
• Somewhere in 2021: Enough JOSM users have switched to OpenWebStart so we can consider moving the codebase to Java 11.

New 2021 plan:

• 2021-03-28: ask Oracle Java WebStart users to switch to OpenWebStart => r17679
• 2021-08-22: new Windows package that includes Java 16 => r18151:18155
• 2021-08-22: include JavaFX 16 in macOS and Windows packages => r18161
• 2021-08-22: update JNLP files to request Azul JVM from OpenWebStart as it includes JavaFX => r18158:18159
• 2021-08-22: update Debian/Ubuntu launch script to depend on openjfx => r18160
• 2021-09-15: Java 17 is released. Switch macOS / Windows packages to Java 17 and JavaFX 17, update Debian/Ubuntu launch script to prefer 17 over 11 and 8 => r18225
• 2022-04-21: Ubuntu 22.04 LTS is released and ships Java 17 (note: default-jre is Java 11)
• 2022-11-25: OpenWebStart ​adds Java 17 to their ​JVM list => ​done
• somewhere in 2023/2024: Enough JOSM users are now using Java 17+ so we can consider moving the codebase to Java 17.
• 2024-04-xx: Ubuntu 24.04 LTS is released and ships Java 17+ as default; Debian Bookworm already ships Java 17 as default.

Java 21 followup: #23564

[Don-vip]

First problem: #17632

[Don-vip]

First alpha version (0.2.0) is available for download: ​https://openwebstart.com/download/

[mdk]

Version 1.1.1 is released (at 16.12.2019)

[Don-vip]

Second problem: #18390 => ​https://github.com/AdoptOpenJDK/IcedTea-Web/pull/564 => ​https://github.com/AdoptOpenJDK/IcedTea-Web/releases/tag/icedtea-web-2.0.0-alpha9 => ​https://github.com/karakun/OpenWebStart/releases/tag/v1.1.4

[Don-vip]

AdoptOpenJDK API now available through https://josm.openstreetmap.de/remote/adoptopenjdk-api/ (#18723)

[Don-vip]

​https://wiki.ubuntu.com/BionicBeaver/ReleaseNotes/

As of 18.04.4, OpenJDK 11 is the default in 18.04.

OpenJDK 8 has moved to universe and will remain available there for the life of 18.04, to provide migration time for packages, custom applications, or scripts that can't be build with OpenJDK 11. OpenJDK 8 will be updated in 18.04 until Ubuntu 16.04 LTS reaches EOL in April 2021.

[Don-vip]

Another issue: #18737 => ​https://github.com/AdoptOpenJDK/IcedTea-Web/issues/595

[Don-vip]

AdoptOpenJDK officially refused to ship OpenJFX in their binary distributions:
​https://github.com/AdoptOpenJDK/TSC/issues/27#issuecomment-607663428

So the best distribution for us would be ​Azul Zulu and ​Bellsoft's Liberica.

[taylor.smock]

Another issue: #19044 (Azul Zulu 11.0.6/AdoptOpenJDK 11.0.6). It looks like jdk.swing.interop.SwingInterOpUtils doesn't exist, for whatever reason. Oddly enough, it is never directly called by us (its called by JFXPanel, which is in the java distribution). I haven't filed a bug upstream, since I haven't run with a "standard" version of Azul Zulu yet.

EDIT: Not reproducible with Azul Zulu or AdoptOpenJDK from CLI.

Possibly related to ​https://github.com/AdoptOpenJDK/IcedTea-Web/issues/595 .

[mdk]

It looks like ​​https://github.com/AdoptOpenJDK/IcedTea-Web/issues/595 is fixed in the actual release 1.1.7

[jBeata]

Do you have a more clear deadline for moving the code-base to Java11? It will happen in the first half of 2021 or the second?

[stoecker]

Replying to jBeata:

Do you have a more clear deadline for moving the code-base to Java11? It will happen in the first half of 2021 or the second?

Usually when less than 5% of our users use older versions. Currently 70% of our users use Java 8. About 20% use Java 11.

[Don-vip]

Oracle changed again their plans concerning Java 8. It was previously expected they stop releasing public updates of Java 8 on java.com (the version used by nearly all our Windows users) at the end of 2020.

Now ​the plan is:

Oracle will continue to provide free public updates and auto updates of Java SE 8 indefinitely for Personal, Development and other Users via java.com. Oracle will provide at least 18 months notice on this page and other communication channels if an end of availability date is set. [...]
Oracle does not plan to migrate desktops from Java SE 8 to later versions via the auto update feature. This includes the Java Plugin and Java Web Start. [...]

So we won't see any major decrease of Java 8 usage until we force this change by promoting OpenWebStart and Java 11 to our users. I didn't start to work on this yet.

Given on our past experience, it takes months between the time we start to deprecate a version of Java and the time enough people abandoned it, so I wouldn't see a Java 11 migration before Q2 2021 at best.

[stoecker]

Replying to Don-vip:

Q2 2021 at best.

Still very optimistic. Q4 2021 or later seems more realistic ;-) Even Linux has still about 30% Java 8.

P.S. Added --os option to our checkjosm tool :-)

[Don-vip]

I meant S2, with Q4 in mind actually :)

[jBeata]

Thanks for the clarification. We also use Java 8 since the minimum version for running JOSM is Java 8. But we plan to switch as soon as JOSM main version is updated to 11.

[Don-vip]

Current numbers:

Java Main Version --> 8 (7174, 67.6%) 9 (21,  0.2%) 10 (24,  0.2%) 11 (2130, 20.1%) 12 (167,  1.6%) 13 (248,  2.3%) 14 (299,  2.8%) 15 (532,  5.0%) 16 (10,  0.1%) 17 (7,  0.1%)

I'm really happy to see that OpenWebStart really kicks in: several versions released, 5 companies seem to sponsor Karakun on front page, nearly 200 developers have starred the project on GitHub. Software is now available on Windows, mac and Linux. It's a complete and viable replacement to Oracle WebStart.

I'll see how to detect Oracle WebStart and suggest users to switch to OpenWebStart. At the same time they do that, they'll switch from Oracle JRE 8 to Azul JRE 11.

@Dirk would it be possible to update checkjosm to get vendor stats? I'd like to see the percentage of users running an Oracle runtime versus those who do not.

[stoecker]

That information isn't available in the User-Agent, so I can't display it :-)

[Don-vip]

Replying to stoecker:

That information isn't available in the User-Agent, so I can't display it :-)

Ah, I thought it was sent. I'll check only Java 8 then. Should be the same thing.

[Don-vip]

I just fixed #18737 which was the only blocker to OpenWebStart migration. Let's get rid of Oracle Java WebStart.

[Don-vip]

Vote for your preferred rocket icon! Which one looks the best? 1, 2, 3, 4?

[simon04]

I vote for 4️⃣ :-)

[simon04]

Awesome, successfully tested (ticket #18737) on macOS! On Windows and macOS the status report lacks any indication that JOSM is running via webstart:

Relative:URL: ^/trunk
Repository:UUID: 0c6e7542-c601-0410-84e7-c038aed88b3b
Last:Changed Date: 2021-03-26 22:41:31 +0100 (Fri, 26 Mar 2021)
Revision:17674
Build-Date:2021-03-27 02:30:56
URL:https://josm.openstreetmap.de/svn/trunk

Identification: JOSM/1.5 (17674 en) Mac OS X 10.16
OS Build number: macOS 11.2.3 (20D91)
Memory Usage: 410 MB / 2048 MB (171 MB allocated, but free)
Java version: 16+36, Azul Systems, Inc., OpenJDK 64-Bit Server VM
Look and Feel: com.formdev.flatlaf.FlatLightLaf
Screen: Display 1 1440×900 (scaling 2,00×2,00) Display 2 3008×1692 (scaling 2,00×2,00)
Maximum Screen Size: 3008×1692
Best cursor sizes: 16×16→16×16, 32×32→32×32
VM arguments: [--add-modules=java.scripting,java.sql, --add-exports=java.desktop/com.apple.eawt=ALL-UNNAMED, --add-exports=java.desktop/com.sun.imageio.spi=ALL-UNNAMED, --add-exports=java.desktop/com.sun.imageio.plugins.jpeg=ALL-UNNAMED, --add-exports=javafx.graphics/com.sun.javafx.application=ALL-UNNAMED, --add-exports=jdk.deploy/com.sun.deploy.config=ALL-UNNAMED, --add-opens=java.base/java.lang=ALL-UNNAMED, --add-opens=java.base/java.nio=ALL-UNNAMED, --add-opens=java.base/jdk.internal.loader=ALL-UNNAMED, --add-opens=java.base/jdk.internal.ref=ALL-UNNAMED, --add-opens=java.desktop/javax.imageio.spi=ALL-UNNAMED, --add-opens=java.desktop/javax.swing.text.html=ALL-UNNAMED, --add-opens=java.prefs/java.util.prefs=ALL-UNNAMED, -Djava.util.Arrays.useLegacyMergeSort=true, --add-exports=java.base/sun.net.www.protocol.jar=ALL-UNNAMED,java.desktop, --add-exports=java.base/jdk.internal.util.jar=ALL-UNNAMED,java.desktop, --add-exports=java.base/com.sun.net.ssl.internal.ssl=ALL-UNNAMED,java.desktop, --add-reads=java.naming=ALL-UNNAMED,java.desktop, --add-exports=java.desktop/sun.awt.X11=ALL-UNNAMED,java.desktop, --add-exports=java.desktop/sun.applet=ALL-UNNAMED,java.desktop,jdk.jsobject, --add-exports=java.base/sun.security.action=ALL-UNNAMED,java.desktop, --add-reads=java.base=ALL-UNNAMED,java.desktop, --add-exports=java.base/sun.net.www.protocol.http=ALL-UNNAMED,java.desktop, --add-exports=java.naming/com.sun.jndi.toolkit.url=ALL-UNNAMED,java.desktop, --add-exports=java.base/sun.security.util=ALL-UNNAMED,java.desktop,ALL-UNNAMED, --add-reads=java.desktop=ALL-UNNAMED,java.naming, --add-exports=java.desktop/sun.awt=ALL-UNNAMED,java.desktop, --add-exports=java.base/sun.security.x509=ALL-UNNAMED,java.desktop,ALL-UNNAMED, --add-exports=java.desktop/javax.jnlp=ALL-UNNAMED,java.desktop, --add-exports=java.base/sun.security.provider=ALL-UNNAMED,java.desktop, --add-exports=java.base/sun.security.validator=ALL-UNNAMED,java.desktop]

Plugins:
+ ImportImagePlugin (35567)
+ apache-commons (35524)
+ ejml (35458)
+ flatlaf (35703)
+ geotools (35458)
+ jts (35458)
+ log4j (35458)

This statement might help us to identify OpenWebStart:

Class.forName("com.openwebstart.launcher.OpenWebStartLauncher")

[skyper]

Replying to Don-vip:

Vote for your preferred rocket icon! Which one looks the best? 1, 2, 3, 4?

I like 1 or 4. Maybe, the dark brown on the outside wings could be a bit lighter. With the dark mode icons need to respect dark and light background and often contrasting outlines like 1 and 2 show are helpful.

[simon04]

When starting, OpenWebStart shows this information message:

App name on macos screen menu is Boot · Issue 325 · karakun/OpenWebStart · ​https://github.com/karakun/OpenWebStart/issues/325

[Don-vip]

Replying to simon04:

This statement might help us to identify OpenWebStart:

Class.forName("com.openwebstart.launcher.OpenWebStartLauncher")

I plan to use this (not tested yet):

    /**
     * Determines whether JOSM has been started via Web Start (JNLP).
     * @return true if JOSM has been started via Web Start (JNLP)
     * @since xxx
     */
    public static boolean isRunningWebStart() {
        try {
            // See http://stackoverflow.com/a/16200769/2257172
            return Class.forName("javax.jnlp.ServiceManager") != null;
        } catch (ClassNotFoundException e) {
            return false;
        }
    }

    /**
     * Determines whether JOSM has been started via Oracle Java Web Start.
     * @return true if JOSM has been started via Oracle Java Web Start
     * @since 15740
     */
    public static boolean isRunningJavaWebStart() {
        return isRunningWebStart() && Package.getPackage("com.sun.javaws") != null;
    }

    /**
     * Determines whether JOSM has been started via Open Web Start (IcedTea-Web).
     * @return true if JOSM has been started via Open Web Start (IcedTea-Web)
     * @since xxx
     */
    public static boolean isRunningOpenWebStart() {
        return isRunningWebStart() && Package.getPackage("net.adoptopenjdk.icedteaweb") != null;
    }

[Don-vip]

This is the upgrade text I came with. Comments?

For the icon, I've started a ​poll on Twitter as well, I hope a clear winner comes out :D

[Don-vip]

Replying to Don-vip:

    public static boolean isRunningOpenWebStart() {
        return isRunningWebStart() && Package.getPackage("net.adoptopenjdk.icedteaweb") != null;
    }

And we should also be aware of upcoming changes that will probably affect package names for ITW ​https://blog.adoptium.net/2021/03/eclipse-adoptium-announcement/

[simon04]

Replying to Don-vip:

This is the upgrade text I came with. Comments?

Two small suggestions:

-Oracle implementation (2x)
+an Oracle implementation

-as a new product: OpenWebStart.
+as a new product: OpenWebStart

What is the difference between "OK" and "Download OpenWebStart"?

You you plan to add this dialog for the 21.03 release (I'm referring to i18n)?

[Don-vip]

Replying to simon04:

Two small suggestions:

-Oracle implementation (2x)
+an Oracle implementation

-as a new product: OpenWebStart.
+as a new product: OpenWebStart

Thanks.

What is the difference between "OK" and "Download OpenWebStart"?

OK does nothing, it just closes the dialog. It's the same behaviour as current MainApplication.askUpdateJava dialog.

You you plan to add this dialog for the 21.03 release (I'm referring to i18n)?

Yes, tomorrow evening.

[stoecker]

Replying to Don-vip:

This is the upgrade text I came with. Comments?

Remove the remaining sentence after "Java 11" and join the two sentences with "but".

I'd vote for icon 4 (icon 1 is second place).

[Don-vip]

New version taking into account Simon and Dirk feedback:

[simon04]

Looks good! 👍🏿

[stoecker]

Actually I meant to join the first two sentences, but this is even better ;-)

[Don-vip]

Good :) Thank you :)

[Don-vip]

In 17679/josm:

see #17858 - ask Oracle Java WebStart users to switch to OpenWebStart

Rocket icon from ​https://github.com/twitter/twemoji/blob/v13.0.2/assets/svg/1f680.svg

[simon04]

Java Warnings (Package.getPackage is deprecated since Java 9):

/Users/simon/src/josm/src/org/openstreetmap/josm/tools/Utils.java:1742: warning: [deprecation] getPackage(String) in Package has been deprecated
        return isRunningWebStart() && Package.getPackage("com.sun.javaws") != null;
                                             ^
/Users/simon/src/josm/src/org/openstreetmap/josm/tools/Utils.java:1751: warning: [deprecation] getPackage(String) in Package has been deprecated
        return isRunningWebStart() && Package.getPackage("net.adoptopenjdk.icedteaweb") != null;
                                             ^

[Don-vip]

In 17692/josm:

see #17858 - fix deprecation warnings

[simon04]

Replying to simon04:

When starting, OpenWebStart shows this information message:

"The proxy 'System proxy' does not support 'Passive FTP Mode (PASV)'"

Only affects macOS, see ​https://github.com/karakun/OpenWebStart/blob/26c8990af7f2214b290d1056cd27a7f78f4477a6/openwebstart/src/main/java/com/openwebstart/proxy/mac/MacProxyProvider.java#L36-L38

Fixed by selecting "No Proxy" in the OpenWebStart Settings:

[gaben]

On Windows, the WebStart dialogue pops up and not steals the focus, sometimes resulting in a JOSM 'loading' indefinitely. The same applies to Java update notification windows. The only way to resolve is to click somewhere on the splash screen.

URL:https://josm.openstreetmap.de/svn/trunk
Repository:UUID: 0c6e7542-c601-0410-84e7-c038aed88b3b
Last:Changed Date: 2021-04-01 23:17:01 +0200 (Thu, 01 Apr 2021)
Build-Date:2021-04-01 21:46:03
Revision:17702
Relative:URL: ^/trunk

Identification: JOSM/1.5 (17702 hu) Windows 10 64-Bit
OS Build number: Windows 10 Pro for Workstations 2009 (19042)
Memory Usage: 912 MB / 1820 MB (719 MB allocated, but free)
Java version: 1.8.0_281-b09, Oracle Corporation, Java HotSpot(TM) 64-Bit Server VM
Look and Feel: com.sun.java.swing.plaf.windows.WindowsLookAndFeel
Screen: \Display0 1920×1200 (scaling 1,00×1,00)
Maximum Screen Size: 1920×1200
Best cursor sizes: 16×16→32×32, 32×32→32×32
System property file.encoding: Cp1250
System property sun.jnu.encoding: Cp1250

[nkamapper]

I have an old iMac running High Sierra, which is not supported by OpenWebStart. The iMac cannot be upgraded to a newer OS. Is there a way to get an earlier version of josn.jnlp, which would work?

[skyper]

Not on the official side, I guess, but note, *.jnlp are text files which can be modified and created with any text editor.

Another solution is to use .jar files directly. Starting should work from file manager or with java -jar path/filename from the console or with a simple start script including the java command. See Download.

[Don-vip]

We will only support OpenWebStart for JNLP. However in your case you can use the macOS build directly instead of WebStart. There is also the brew cask package on macOS.

[taylor.smock]

Somewhere in 2021: Enough JOSM users have switched to OpenWebStart so we can consider moving the codebase to Java 11.

Replying to Don-vip:

Current numbers:

Java Main Version --> 8 (7174, 67.6%) 9 (21,  0.2%) 10 (24,  0.2%) 11 (2130, 20.1%) 12 (167,  1.6%) 13 (248,  2.3%) 14 (299,  2.8%) 15 (532,  5.0%) 16 (10,  0.1%) 17 (7,  0.1%)

Have these numbers changed significantly? As in, will we be able to move to Java 11 somewhere in 2021, or is it going to be somewhere in 2022?

[stoecker]

Looks still similar (I stripped the absolute values, only left the %):

Java Main Version --> 8 (66.4%) 9 (0.1%) 10 (0.2%) 11 (21.1%) 12 (1.0%) 13 (1.7%) 14 (2.3%) 15 (0.9%) 16 (5.6%) 17 (0.5%) 18 (0.3%)

Values of 15 now moved to 16, but the others are mainly unmoved.

[taylor.smock]

OK. We probably won't be moving to Java 11 this year then.
Thanks for looking.

[stoecker]

It seems the majority of newer Java versions comes from newer Linux systems.

[Don-vip]

Good news: most of Linux and mac users now use Java >= 11.
Bad news: almost 90% of Windows users (Windows users being about two thirds of the JOSM user base) are still using Java 8.

But it's not surprising as Oracle JRE will never prompt Java 8 users to update. And apart of the Java WebStart > OpenWebstart migration message, I still haven't make anything to ask "regular" JRE-based users to transition from an Oracle JRE to something else, so now would be a good time.

From ​https://rafael.codes/openjdk/ I think we should promote these two OpenJDK distributions:

• Liberica JDK from Bell: ​https://bell-sw.com/
• Zulu from Azul: ​https://www.azul.com/downloads/?package=jdk

They both ship OpenJFX, which could be handy for future long-term JOSM developments.

[taylor.smock]

Replying to Don-vip:

Bad news: almost 90% of Windows users (Windows users being about two thirds of the JOSM user base) are still using Java 8.

I think I remember something somewhere talking about Minecraft and Java 16 being the baseline. Hopefully that pushes most of the users on Windows to Java 16 (and 16 > 11 :) ).

Do we want to use jpackage for the Windows installer as well (we are already using it for Mac)? This may also help us push to Java 11+. But it does increase size (15 mb -> 38mb).

They both ship OpenJFX, which could be handy for future long-term JOSM developments.

Good to know -- Mapillary was asking me (again) if I could reuse their Javascript based image viewer, which requires either (1) a JS engine that supports newer features (no native Java ones) or (b) embedding a web browser and distributing that for all the platforms JOSM supports.

[stoecker]

Replying to taylor.smock:

Good to know -- Mapillary was asking me (again) if I could reuse their Javascript based image viewer, which requires either (1) a JS engine that supports newer features (no native Java ones) or (b) embedding a web browser and distributing that for all the platforms JOSM supports.

We did that in the past with a tool called webkit-image, which rendered a URL and saved the resulting image. This was necessary to follow the Yahoo guidelines for their service. Was an ugly workaround, but it worked. Don't know if JOSM still has the code for this or if it was removed.

[Don-vip]

Replying to stoecker:

Don't know if JOSM still has the code for this or if it was removed.

Removed long ago ;) r11578:11581

[Don-vip]

Replying to taylor.smock:

Do we want to use jpackage for the Windows installer as well (we are already using it for Mac)? This may also help us push to Java 11+. But it does increase size (15 mb -> 38mb).

Yes, can be an option too.

[gaben]

I still use Java 8 on Windows because in later versions they changed how scaling works and it is unusable for me. On Linux there is no difference :)
Will try again and report back if something improved in the meantime, which potentially means minus one Java 8 user.

[taylor.smock]

Replying to Don-vip:

Replying to stoecker:

Don't know if JOSM still has the code for this or if it was removed.

Removed long ago ;) r11578:11581

author of removal: stoecker :)

And yes, it looks like an ugly workaround. AKA not something I want to do. I kind of want to try to avoid workarounds, if at all possible. They tend to break.

But if JavaFX is an option in the future, I believe MS Streetside has a 360 image viewer, which would probably fix the problem anyway (ignoring the perennial "2048px is too little for 360" comments).

[taylor.smock]

Replying to gaben:

I still use Java 8 on Windows because in later versions they changed how scaling works and it is unusable for me. On Linux there is no difference :)
Will try again and report back if something improved in the meantime, which potentially means minus one Java 8 user.

Stupid question: Have you reported a bug to the JDK maintainers? Or is the scaling issue intentional?

[stoecker]

Replying to taylor.smock:

Don't know if JOSM still has the code for this or if it was removed.

Removed long ago ;) r11578:11581

author of removal: stoecker :)

Ah no. That's unfair. It was broken before.

[anonymous]

Replying to taylor.smock:

Stupid question: Have you reported a bug to the JDK maintainers? Or is the scaling issue intentional?

It was reported years ago and as far as I know it's an intentional fix for another scaling issue.
If you are interested I can collect the JDK ticket(s) and SO questions and a possible solution if there is any (should be).

Edit: the login cookie expired, it was me (gaben).

[Don-vip]

Ah great, we can give an hint to OpenWebStart about the vendor we prefer:
​https://openwebstart.com/docs/OWSGuide.html#_specify_a_specific_vendor_in_the_jnlp_file

So if we set "Azul" or "BellSoft", OpenWebStart will download this JVM instead of AdoptOpenJDK (which doesn't include JavaFX).

It seems we can only specify a single vendor, so which one should we use between the two? Both seem OK.

[Don-vip]

In 18158/josm:

see #17858 - add custom WebStart DTD with OpenWebStart vendor extension, as it is not (yet?) provided

• ​https://openwebstart.com/docs/OWSGuide.html#_specify_a_specific_vendor_in_the_jnlp_file
• ​https://stackoverflow.com/a/62633477/2257172
• ​http://java.sun.com/dtd/JNLP-6.0.dtd

[Don-vip]

In 18159/josm:

see #17858 - request Azul JVM from OpenWebStart and add following JavaFX modules: controls,media,swing,web

[Don-vip]

In 18160/josm:

see #17858 - make Debian/Ubuntu package depend on openjfx and add following JavaFX modules: controls,media,swing,web

[Don-vip]

In 18161/josm:

see #17083, see #17858 - include JavaFX 16 in macOS/Windows packages

[Don-vip]

I think we have now a good chance to switch to Java 17 somewhere in 2022. This migration will have been absolutely horrible, the worst in my Java development experience.

[Don-vip]

In 18166/josm:

see #17858 - promote Azul and BellSoft distributions over Oracle one

[Don-vip]

In 35805/osm:

see #17858 - do no longer ship Java FX through giant openjfx plugin

[Don-vip]

In 35806/osm:

see #17858 - do no longer ship Java FX through giant openjfx plugin (dist)

[Don-vip]

In 18167/josm:

see #17858 - deprecate native javafx plugins, no longer needed, would rather cause harm as loading JavaFX through classpath is no longer supported anayway

​https://github.com/openjdk/jfx/blob/master/doc-files/release-notes-16.md#javafx-runtime-logs-a-warning-if-javafx-modules-are-loaded-from-the-classpath

[taylor.smock]

Add automatic module name -- #15229 seems to indicate we'll want a lot of subpackages, but it may be useful to say "hey, we will be using this base module name"

[taylor.smock]

Replying to Don-vip:

In 18167/josm:

see #17858 - deprecate native javafx plugins, no longer needed, would rather cause harm as loading JavaFX through classpath is no longer supported anayway

​https://github.com/openjdk/jfx/blob/master/doc-files/release-notes-16.md#javafx-runtime-logs-a-warning-if-javafx-modules-are-loaded-from-the-classpath

For this deprecation, I should probably update the MS Streetside plugin to no longer depend upon JavaFX, right? Or should I wait for the September release to avoid breaking stuff right now?

[Don-vip]

No need to wait for next release if you bump the min JOSM version of the plugin.

[taylor.smock]

Replying to Don-vip:

No need to wait for next release if you bump the min JOSM version of the plugin.

Fair enough. Except I don't think the current JOSM packages for Windows are built that way. So I've got to wait for the stable release in order to set the min version. I can technically set the min version to the latest version, but I probably ought to wait for a release.

Anyway, I was thinking:

• Move the JavaFX 360 viewer from MS Streetside into JOSM core (in a try-catch block, just so that people without JavaFX can still use JOSM), this will fix #16472.
• Modify the ImageEntry class to have a isPano or is360 method (probably a default method returning false)
• Modify the ImageEntry class to have a getNext and getPrevious method (maybe also getFirst and getLast as well)
• Modify the ImageEntry class to have an additionalPainters method so that detections can be drawn on the image (Mapillary/KartaView).

But that should be in a different ticket. :)

[Don-vip]

Replying to taylor.smock:

• Move the JavaFX 360 viewer from MS Streetside into JOSM core (in a try-catch block, just so that people without JavaFX can still use JOSM), this will fix #16472.

I plan to finally take a look at it, move it to the openjfx plugin, with necessary changes in core if needed (like the MP3 player) so that 360 pictures can be viewed with just the openjfx plugin (indeed to fix #16472). Making sure JavaFX was available everywhere was the first step.

Let's discuss it in #16472.

[stoecker]

Replying to taylor.smock:

I can technically set the min version to the latest version, but I probably ought to wait for a release.

That's not necessary. The minimum version is the needed minimum version for the plugin to work. The "stable" release is a rather artificial construct for josm to satisfy users (and developers) who expect such a thing. For the plugin mechanics it makes no difference.

[Don-vip]

In 35807/osm:

see #17858 - fix javafx plugin

[Don-vip]

In 35808/osm:

see #17858 - fix javafx plugin (dist)

[taylor.smock]

Replying to stoecker:

That's not necessary. The minimum version is the needed minimum version for the plugin to work. The "stable" release is a rather artificial construct for josm to satisfy users (and developers) who expect such a thing. For the plugin mechanics it makes no difference.

The reason why I was inclined to wait for release is just in case someone's JNLP file doesn't self update -- I'm presuming there will be a note in StartupPage about the new JavaFX dependency, so they would at least be able to know about the change.

EDIT: I'm going to have to troubleshoot why I cannot seem to build MS Streetside (I've tried with openjfx installed).

See r35809/osm.

[Don-vip]

In fact, why did you remove the dependency? I only deprecated the native plugins, but the openjfx plugin still exists, and I'd like to make it the only JOSM plugin to provide a 360° image viewer, if it appears JavaFX is really mandatory and we can't create a Swing-based one in core.

[Don-vip]

In 18171/josm:

see #17858 - add javafx to module-path in Linux launcher

[taylor.smock]

Replying to Don-vip:

In fact, why did you remove the dependency? I only deprecated the native plugins, but the openjfx plugin still exists, and I'd like to make it the only JOSM plugin to provide a 360° image viewer, if it appears JavaFX is really mandatory and we can't create a Swing-based one in core.

My bad. I read "deprecated" as move off ASAP.

[Don-vip]

No problem. The JavaFX handling is not really simple :D

[Don-vip]

New numbers:

Java Main Version --> 8 (63.2%) 9 (0.1%) 10 (0.2%) 11 (20.4%) 12 (0.7%) 13 (1.3%) 14 (1.7%) 15 (0.8%) 16 (9.2%) 17 (1.3%) 18 (1.1%)

[skyper]

Problem with open webstart on windows: #21354

[skyper]

There are problems with kendzi3d plugin and java 16/17, see #21348.

[Don-vip]

Looks like the impact of Windows installer is massive, Java 8 dropped a lot more than I expected:

Java Main Version --> 8 (39.3%) 9 (0.1%) 10 (0.1%) 11 (24.2%) 12 (0.4%) 13 (1.0%) 14 (1.2%) 15 (0.4%) 16 (17.9%) 17 (8.3%) 18 (7.2%)

Among Java 8 users, the OS market share is:

Linux (18.6%) Mac (8.1%) Windows (73.4%)

[taylor.smock]

Replying to skyper:

There are problems with kendzi3d plugin and java 16/17, see #21348.

Good news: I've got a kendzi3d-dev plugin that runs on Mac OSX under Java 17. There are a couple of bugs, but it doesn't crash.

[taylor.smock]

As a heads up, some of the tools we use depend upon Java 11+ now.

Specifically,

• checkstyle (as of checkstyle 10, see ​https://checkstyle.sourceforge.io/releasenotes.html#Release_10.0 )
• error_prone (as of error_prone 2.11, see ​https://github.com/google/error-prone/releases/tag/v2.11.0 )
  • UnicodeDirectionalityCharacters
  • AlreadyChecked (looks for if (a == b) {} else if (a == b) {} type issues)

[stoecker]

StartupPage now warns for Java <= 11. Maybe that helps ;-)

[taylor.smock]

Good to know (also, Ubuntu 22.04 was released today).

It has been awhile since the last update of Java stats (see comment:87), but I would presume we are at least 60%+ Java 11 or later. And hopefully more like 80%+.

Anyway, neither checkstyle nor error_prone are required updates to get Java-EarlyAccess-JOSM working again (I think I could update jacoco from 0.8.7 to 0.8.8, but I don't know if something else will cause it to fail).

[stoecker]

Current stats: 8 (36.1%) 11 (22.1%) 17 (34.6%)

[anonymous]

Hello, i still run into issues in OWS. It keeps saying that "no suitable jvm is found" whenever I launch from the file.

[anonymous]

Replying to anonymous:

Hello, i still run into issues in OWS. It keeps saying that "no suitable jvm is found" whenever I launch from the file.

worth noting, I can't launch from the terminal either.

[anonymous]

Used OWS in general without problems. Had a break of some weeks and now on MacOS 12.3.1 OWS complains about a missing JVM and it does not download it automaticalle. Is there a way to install the needed JVM (azul?) manually?

[ajf3934221jos]

Same problem … NO JVM

[jBeata]

Do you have a more clear deadline for discontinuing Java 8 support?

[ajf3934221jos]

Replying to ajf3934221jos:

Same problem … NO JVM

... solved by itself

[taylor.smock]

Replying to anonymous, anonymous, anonymous, ajf3934221jos, ajf3934221jos:

Hello, i still run into issues in OWS. It keeps saying that "no suitable jvm is found" whenever I launch from the file.

This appears to have been an OpenWebStart Bug. See ​https://github.com/karakun/OpenWebStart/issues/514 for more information. It should be fixed now.

Replying to jBeata:

Do you have a more clear deadline for discontinuing Java 8 support?

I do not. I don't know about the rest of the team. As noted by stoecker, we still have ~1/3 users on Java 8. Previous practice has been to increase Java versions when a relatively small portion of JOSM users are on the deprecated version (from comment:14, <5%). With that said, getting users to migrate off of Java 8 is somewhat of a nightmare (specifically for Mac/Windows users), since the ​java.com releases are Java 8. I've updated documentation to point to other Java distributions (specifically Azul with JavaFX), and I encourage users to either use OpenWebStart or the installers.

[taylor.smock]

Stupid question: Do we want to update the link to java.com on the JOSM home page? (**JOSM** is an extensible editor for [osmwww: OpenStreetMap] (OSM) for [https://www.java.com Java 8+].)

I.e., **JOSM** is an extensible editor for [osmwww: OpenStreetMap] (OSM) for [https://www.java.com Java 8+], but we recommend [https://www.azul.com/downloads/?version=java-17-lts&package=jre-fx#download-openjdk Java 17+].

JOSM is an extensible editor for ​OpenStreetMap (OSM) for ​Java 8+, but we recommend ​Java 17+.

[stoecker]

Wha, when you change it, then simply replace the link. ;-)

[Don-vip]

current stats:

8-10 (31.2%) 11-16 (27.3%) 17+ (41.5%)

these numbers include everyone, meaning also people running a very old version of JOSM and unlikely to be affected by our decision to migrate to Java 11/17 as they don't update their JOSM at all. If we only look at people running at least JOSM 18360 (21.12), numbers are:

8-10 (19.6%) 11-16 (28.4%) 17+ (52.0%)

Among these users, the Java 8 users distribute as follow:

OS > 1%:
Linux (28.7%) Mac (5.0%) Windows (66.0%)

Languages > 5%:
en (51.5%) de (15.5%) en_GB (6.2%) fr (5.4%)

[sebastic]

The Debian package had to drop support for openjdk-8 when it switch to building with openjdk-11 because it not backwards compatible, would openjdk-11 support need to be dropped when JOSM is built with openjdk-17 or does that do preserve backward compatibility?

[taylor.smock]

Replying to sebastic:

The Debian package had to drop support for openjdk-8 when it switch to building with openjdk-11 because it not backwards compatible, would openjdk-11 support need to be dropped when JOSM is built with openjdk-17 or does that do preserve backward compatibility?

We still build and develop JOSM with Java 8. If we were to build it with Java 11, then yes, there might be some incompatibilities (there are some classes that override a method in Java 11 that did not in Java 8, and if we compile with Java 11 it looks for those overridden methods in the implementing classes). I do not know if that is the case for Java 17 -> Java 11, but a good general recommendation it to compile it with the lowest version of Java possible, since that will (almost always) work on later versions of Java.

So the answer is "I don't know". There is a --release switch that can be used by javac to output the appropriate class files, with the caveat that some JDK classes may now override a method, which has caused issues in the past. For your sanity (and ours), just use the "minimum" java version the debian packagers want to support that JOSM also supports.

I don't know how the Debian package decides which compiler to use (default-jdk?), but that is really the limiting factor.

With all that said, some plugins are now depending upon Java 11+, so it isn't a horrible thing that the Debian package requires Java 11+.

[sebastic]

Java 11 isn't the problem, Java 17 is. Specifically these bits:

• 2021-09-15: Java 17 is released. Switch macOS / Windows packages to Java 17 and JavaFX 17, update Debian/Ubuntu launch script to prefer 17 over 11 and 8 => r18225
• 2022-04-21: Ubuntu 22.04 LTS is released and ships Java 17 (note: default-jre is Java 11)
• somewhere in 2022: Enough JOSM users are now using Java 17+ so we can consider moving the codebase to Java 17.

Debian bullseye is in the same boat as Ubuntu jammy, openjdk-11 is still the default-jdk and default-jre.

Requiring Java 17 to build JOSM will mean using a non-default JDK on Debian stable and Ubuntu LTS.

openjdk-17 is available in backports for Debian stable, java-common in experimental even made it the default, but that change isn't going to find its way into stable and LTS releases.

[taylor.smock]

Here is a hopefully stupid question:

Do we have infrastructure in place to inform users that they must update to a newer version of Java? I didn't see anything, but I might have missed it.

Pseudocode of what I was looking for:

    private static void javaVersionCheck() {
        if (Utils.getJavaVersion() < 11) {
            if (Utils.isRunningJavaWebStart()) {
                // Show user link to OpenWebStart, then exit. Maybe two months from release before a force exit though
                // (aka, we start compiling most classes for Java 11).
            } else {
                // Our installers all include Java 17+, so everyone here must be running the jar file or be on Unix/Linux.
                if (PlatformManager.isPlatformOsx()) {
                    // Link to OSX installer and Java 17, maybe OpenWebStart?
                } else if (PlatformManager.isPlatformWindows()) {
                    // Link to Windows installer and Java 17, maybe OpenWebStart?
                } else if (PlatformManager.isPlatformUnixoid()) {
                    // Ask user to install Java 11/17 and use that instead. Packagers might be a problem.
                } else {
                    // We probably won't ever hit this, but just tell the user that they need to use Java 11+ and link to Java 17 (maybe Java 17 source?)
                }
            }
        }
    }

I was kind of thinking that we could continue compiling a few classes with Java 8 (MainApplication, PlatformManager, Utils, OpenBrowser) and give the user more concrete steps for their situation while we move to Java 11/17 for all other classes.

EDIT: We could probably modify the Platform#startupHook for this.

[taylor.smock]

Warn on Java < 11 when not run under OpenWebStart, update URL to point at azul and pre-fill fields to decrease user confusion, Utils.getJavaLatestVersion uses Java 11 (WebStart) and Java 17 for latest Java versions

[taylor.smock]

I'd appreciate it if someone with Windows (x64) could test attachment:17858.patch​ against a 32 bit Java 8 install. It should detect that the OS is x64 and set the appropriate download URL. I'll see if I can get my hands on a Windows machine to test, but I'd like a second opinion in any case.

[Klumbumbus]

With 64 bit Windows and 32 bit Java 8 and applied patch I was asked to Update Java and was linked to:

​https://www.azul.com/downloads/?version=java-17-lts&os=windows&architecture=x86-64-bit&package=jre-fx

(already scrolled down to Download Azul Zulu Builds of OpenJDK)

Build-Date:2022-10-20 21:25:38
Revision:18579
Is-Local-Build:true

Identification: JOSM/1.5 (18579 SVN de) Windows 10 64-Bit
OS Build number: Windows 10 Enterprise 2004 (19041)
Memory Usage: 137 MB / 247 MB (44 MB allocated, but free)
Java version: 1.8.0_351-b10, Oracle Corporation, Java HotSpot(TM) Client VM
Look and Feel: com.sun.java.swing.plaf.windows.WindowsLookAndFeel
Screen: \Display0 2160×1440 (scaling 1.00×1.00)
Maximum Screen Size: 2160×1440
Best cursor sizes: 16×16→32×32, 32×32→32×32
System property file.encoding: Cp1252
System property sun.jnu.encoding: Cp1252
Locale info: de_DE
Numbers with default locale: 1234567890 -> 1234567890

[taylor.smock]

Replying to Klumbumbus:

With 64 bit Windows and 32 bit Java 8 and applied patch I was asked to Update Java and was linked to:

​https://www.azul.com/downloads/?version=java-17-lts&os=windows&architecture=x86-64-bit&package=jre-fx
(already scrolled down to Download Azul Zulu Builds of OpenJDK)

Thanks. :)

I thought it was going to work (based off of documentation), but I wanted to make certain. This will hopefully get rid of the tickets where someone installed 32bit Java on 64bit Windows, and wonders why they run out of memory easily.

I'll go ahead and apply the patch Monday (I want to give other contributors a chance to look at the patch).

I don't think anything is controversial in the patch, but I want to make certain people don't have a problem with us pointing at Java 17 LTS when I think we are going to be jumping to Java 11 LTS next (because of default-jdk/default-jre in Debian, and OpenWebStart not currently having any Java 17 JREs available).

[gaben]

Before moving from Java 8, please take a look at the reported hidpi issues affecting every Java version 9 and above on Windows.

I tried to improve the situation in JOSM, but no success :(

[taylor.smock]

Replying to gaben:

Before moving from Java 8, please take a look at the reported hidpi issues affecting every Java version 9 and above on Windows.

I tried to improve the situation in JOSM, but no success :(

This most likely needs to be fixed in the upstream JDK. As far as I know, we don't do anything with respect to the standard swing component text rendering.

Have anyone filed a Java ticket for the behavior? I know I haven't, as I usually only file tickets when I've diagnosed the problem and can (at least) provide a suggestion on how to fix it. For all I know, this might be fixed if we start compiling against Java 9+ (the build.xml currently just compiles against Java 8, even if the compiler is Java 17, IIRC).

Rather unfortunately, I do not have a windows machine with HiDPI available for debugging.

[Don-vip]

Part of the difficulty of the HiDPI issues is that it's difficuult to support both Java 8 and later versions. Switching to Java 17 might reduce the difficulty of tackling these issues.

[taylor.smock]

Replying to gaben:

Before moving from Java 8, please take a look at the reported hidpi issues affecting every Java version 9 and above on Windows.

I tried to improve the situation in JOSM, but no success :(

For reference, we are working around the following bugs in Java 8:

• ​https://bugs.openjdk.java.net/browse/JDK-8146450 (#18588)
• ​https://bugs.openjdk.java.net/browse/JDK-6292739 (#6345)
• ​https://bugs.openjdk.java.net/browse/JDK-8179014 (#14821)
• ​https://bugs.openjdk.java.net/browse/JDK-8180379 (#14821)
• ​https://bugs.openjdk.java.net/browse/JDK-8159956 (#20621)

In addition, we have special handling for Java 8 in:

• ImageProvider#read (#9984/#16047)
• PlatformHookOsx#startupHook (looks like it is mostly for java -jar users)
• PlatformHookUnixoid#preStartupHook (#12022/#16666)

With all that, I would not be surprised if there was special handling for Java 8 bytecode in the Swing code paths (AKA, if (bytecode < 53) { /* Assume application will look horrible if hidpi paths are followed */ }). But this is probably an interaction between the scaling done by Windows and the scaling done by the JVM, which will require stepping through code on a windows machine.

You could also try running JOSM with -Dsun.java2d.uiScale=1.0 and see if that "fixes" the issue. Definitely not ideal, as it will effectively disable the HiDPI detection code.

Possible upstream bugs:

• ​https://bugs.openjdk.org/browse/JDK-8240568 (josm-found -- TIL we have our own label in the JDK bug tracker)
• ​https://bugs.openjdk.org/browse/JDK-8162529
• ​https://bugs.openjdk.org/browse/JDK-8272707

Replying to Don-vip:

Part of the difficulty of the HiDPI issues is that it's difficuult to support both Java 8 and later versions. Switching to Java 17 might reduce the difficulty of tackling these issues.

As much as I'd like to, I don't think we'll be able to move to Java 17 until OpenWebStart adds Java 17 JVMs to their ​jvm list.

[Don-vip]

Replying to taylor.smock:

josm-found -- TIL we have our own label in the JDK bug tracker

Yep you just have to state to add this label when you report a bug.

As much as I'd like to, I don't think we'll be able to move to Java 17 until OpenWebStart adds Java 17 JVMs to their ​jvm list.

Well if it takes too long we can still update to Java 11 as I planned originally :)

[taylor.smock]

In 18580/josm:

See #17858: start linking to Java 17 for Java updates.

The link for the Java download page now goes to azul.com, and attempts to pre-fill as
much of the download form as possible.

This also adds a method to warn users running Java 10 or earlier that their version
of Java will soon be unsupported by JOSM. This only affects users using Java 10 or
earlier if they are not running JOSM using WebStart.

[tuxayo]

Replying to sebastic:

Java 11 isn't the problem

As of now, aren't there too many people on Java <11? Hopefully by prompting them to upgrade a large part will do. But for a lot of Linux cases, it might the distro install that need updating and many people aren't tech savvy enough to do that. Maybe and having an AppImage with JRE bundled with JOSM would help to up JRE version quicker without loosing much people?

Debian bullseye is in the same boat as Ubuntu jammy, openjdk-11 is still the default-jdk and default-jre.

Requiring Java 17 to build JOSM will mean using a non-default JDK on Debian stable and Ubuntu LTS.

That might be naive but are many people on Debian and Ubuntu using JOSM from their repos? If that's the case, then the package can require openjdk-17 and no issue with default-jre being 11, would that work?

«using JOSM from their repos» Ah crap they would have to enable backports of add an external repo for JOSM so likely most people download the .jar or use WebStart.

.

openjdk-17 is available in backports for Debian stable

Isn't it in the normal repos? ​https://tracker.debian.org/pkg/openjdk-17 still not the default though.

[tuxayo]

Replying to taylor.smock:

As much as I'd like to, I don't think we'll be able to move to Java 17 until OpenWebStart adds Java 17 JVMs to their ​jvm list.

How does it work even for Java 11? IIUC OpenWebStart ships by default Java 8, and the JVM Manager has be used to switch to 11. Is that easy enough so we can count on the vast majority of WebStart users to do that? If not, how to guide them to have a better success rate?

[taylor.smock]

Replying to tuxayo:

As of now, aren't there too many people on Java <11? Hopefully by prompting them to upgrade a large part will do. But for a lot of Linux cases, it might the distro install that need updating and many people aren't tech savvy enough to do that. Maybe and having an AppImage with JRE bundled with JOSM would help to up JRE version quicker without loosing much people?

There is a snap (that really needs to be updated for current "best" practices) and an unofficial flatpak (I've contributed to the flatpak). I don't think it is worth it to add an appimage to our download options.

How does it work even for Java 11? IIUC OpenWebStart ships by default Java 8, and the JVM Manager has be used to switch to 11. Is that easy enough so we can count on the vast majority of WebStart users to do that? If not, how to guide them to have a better success rate?

We have the JNLP set up to use Java 1.8+, which OWS interprets as preferring newer versions of Java. So it will (by default) download the Java 11 JRE for the user. The user can fiddle with some settings so that doesn't happen, but at that point, it is the user's problem.

[taylor.smock]

Replying to Don-vip:

[...]
8-10 (19.6%) 11-16 (28.4%) 17+ (52.0%)

Among these users, the Java 8 users distribute as follow:

OS > 1%:
Linux (28.7%) Mac (5.0%) Windows (66.0%)
[...]

What are the Java 8 numbers looking like now? I'm mostly interested in the Mac/Windows numbers, since I expect Linux users to be using the version from their distro's repository.

I'm thinking about changing the build.xml file to compile everything for Java 11 starting in January (assuming Java 11+ uptake looks good) except:

• MainApplication
• PlatformHook
• Logging
• Anything else needed to get to the point where we ask users to update their Java version

[stoecker]

Java 8: Windows (32%) Mac (14%) Linux (18%) Total (27%)
Java 8 for JOSM >= 18583: Windows (17%) Mac: (22%) Linux (13%) Total (17%)

[taylor.smock]

Thanks. I was really hoping for more movement. It appears that the total hasn't really moved from comment:103 (19.6% to 17%). Only major change is that Mac now accounts for a greater "share" of the Java 8 users (42%, up from 5%). I wonder what is going on there? Maybe some plugins aren't playing nice with Java 17 (see ​MagicWand as an example)?

I'll go ahead and get a patch put together for the build.xml, but I don't think we are at the point where we can switch.

[stoecker]

For the Mac I'd rather say it's a low user base. Mac is used by only 8% of all users. So a few users may change the result a lot.

Also note that our stats are rather low quality. +/- a few per cent are normal.

We'll probably need to choose 10% as goal for switching. 5% seems not realistic for a long time.

[skyper]

Replying to taylor.smock:

I wonder what is going on there? Maybe some plugins aren't playing nice with Java 17 (see ​MagicWand as an example)?

Yes plugins could be one reason but you example is wrong as MagicWand is Java 11+, only.

[jBeata]

Is there any deadline for when Java 8 will not be supported anymore? Currently, also the ImproveOSM, KartaView, and Geohash plugins are running on Java 8, but we can upgrade to Java 17 as soon as JOSM is upgraded.

[stoecker]

Even thought there is a warning now there is not much improvement. So if we switch to java > 8 we'll loose a lot of users.

Thought plugins can already switch earlier. Simply set "Plugin-Minimum-Java-Version" in manifest :-)

Maybe that helps convince some users?

[taylor.smock]

Replying to stoecker:

[Though] plugins can already switch earlier. Simply set "Plugin-Minimum-Java-Version" in manifest :-)

Don't tempt me. :)

Geotools 29.x requires Java 11 or later, and updating that one will make the following plugins effectively require Java 11+:

• cadastre-fr
• ImportImagePlugin
• opendata
• matsim
• Tracer-testing

And then there is the need to change the build-common.xml file for it (as it is part of the ordered_plugins list).

@jBeata: from comment:103, you are looking at ~20% of users not running Java 11 or later. We've had the warning up for about a year now (see comment:90) and JOSM has been linking to a Java 17 distribution for 6 months when asking users to update Java (see comment:116).

Quite frankly, I would encourage plugins to start updating to Java 11 or later at this point, and I'm really close to advocating a drop-dead date for Java 8 of 2024-05-01 (2 year lead time from when we started indicating that we were planning to drop Java 8 support).
Even then, I think we'll probably try to compile a select few classes with Java 8 specifically to give holdouts instructions on how/where to update their java version for another year or so.

[stoecker]

Current stats:

Java Versions with at least 1%: 8 (23.7%) 11 (19.3%) 16 (1.0%) 17 (50.8%) 18 (1.1%) 19 (2.6%)

Values per OS. The second number is cumulated count, so 80% for 11 means 80% of users use >= Java 11.

FreeBSD 17(20%, 60%) 11(20%, 80%)  8(20%,100%)
Linux   17(15%, 23%) 11(57%, 83%)  8(17%,100%)
Mac     17(67%, 71%) 11(16%, 90%)  8( 9%,100%)
Windows 17(65%, 67%) 11( 3%, 71%)  8(28%,100%)

[stoecker]

I see no issue with updating plugins to Java >= 11 when this allows to go forward with e.g. geotools.

With JOSM core OTOH I'm not so sure. I don't see the big killer feature in the core where we really need Java 8 which justifies to loose a quarter of our users. Or do I overlook something?

[taylor.smock]

Replying to stoecker:

With JOSM core OTOH I'm not so sure. I don't see the big killer feature in the core where we really need Java 8 which justifies to loose a quarter of our users. Or do I overlook something?

• Desktop.moveToTrash (Java 9, instead of using Files.delete; this would give users a chance to revert an accidental deletion)
  • I think this would be the most "critical" reason to move to Java 11+, but we could probably work around it by having a Consumer<File> be set and then a Utils.delete method that calls the consumer.
• sealed classes (Java 15, restrict implementors of OsmPrimitive to Node, Way, Relation)
• record classes (Java 15, use for stuff like LatLon, mostly to make it easier to move to a value class when Java supports it)
• HTTP/2 support (currently done via plugin; there are some subtle bugs due to having a differing implementations, mostly related to the default headers)

[stoecker]

Replying to taylor.smock:

Replying to stoecker:

With JOSM core OTOH I'm not so sure. I don't see the big killer feature in the core where we really need Java 8 which justifies to loose a quarter of our users. Or do I overlook something?

• Desktop.moveToTrash (Java 9, instead of using Files.delete; this would give users a chance to revert an accidental deletion)
  • I think this would be the most "critical" reason to move to Java 11+, but we could probably work around it by having a Consumer<File> be set and then a Utils.delete method that calls the consumer.
• sealed classes (Java 15, restrict implementors of OsmPrimitive to Node, Way, Relation)
• record classes (Java 15, use for stuff like LatLon, mostly to make it easier to move to a value class when Java supports it)
• HTTP/2 support (currently done via plugin; there are some subtle bugs due to having a differing implementations, mostly related to the default headers)

Nice, but not really killer features. I think currently we should still go the way of increasing the pressure (i.e. with switching plugins to higher java) and not yet abandon Java 8 for the core (and reevaluate in summer :-).

[Don-vip]

record is a killer feature if we manage to decrease the memory consumption of LatLon substantially. Did anyone start to try that to see how much memory we could gain?

[taylor.smock]

Replying to Don-vip:

record is a killer feature if we manage to decrease the memory consumption of LatLon substantially. Did anyone start to try that to see how much memory we could gain?

I don't think record by itself will be a significant memory reduction. I'll check and see, but I believe that ​value classes will be where we start seeing significant memory improvements. Big things (from our perspective):

• Implicitly final (like a record class)
• No super call
• No synchronized methods (since it doesn't have an identity)
• new Value(0.0, 0.0) == new Value(0, 0) (new Value(0.0, 0.0).equals(new Value(0, 0)) is not necessary)
• An array of LatLon can be stored as [lat1, lon1, lat2, lon2, lat3, lon3] instead of [Pointer1[lat1, lon1], Pointer2[lat2, lon2], Pointer3[lat3, lon3]] which would be a 33% memory reduction (naive, assumes all latlon values are in arrays). TBH, making Node a value class would probably have higher absolute memory reduction returns, since it stores lat/lon in its own fields. But making LatLon a value class might reduce the number of new LatLon calls (so instead of latLon1.interpolate(latLon2, 0.5).lat() creating a new LatLon in interpolate, it just does the calculation for lat()).

When I was looking at it, there were going to be several different hurdles to overcome (namely converting Coordinate to an interface), but otherwise, LatLon is already very similar to a record (only final fields), it just need to become final.

With that said, there might be some memory optimizations that the JVM can do with record classes.

[taylor.smock]

In 18790/josm:

Fix #23103, see #17858: Notify users that a plugin requires a newer Java version

We were previously only logging a warning if a plugin required a newer Java
version. We additionally needed to update the download link generation for Azul
and sync the next minimum Java version with that used by the JOSM wiki check.

There was also a help topic that linked to a dead page (which was also
unavailable in the internet archive).

[gaben]

Replying to taylor.smock:

Why JDK instead of JRE?

Edit: Okay, because JDK always seems to have an installer, while JRE just most of the time.

[taylor.smock]

Edit: Okay, on linux it has an installer.

As noted in the comments, the JDK's have installers. For whatever reason, Azul doesn't provide installers for the JRE packages. At least on Mac/Windows. I didn't check Linux -- most Linux users should just use whatever their distribution provides.

It is a bit more space used on the user's machine, but it should "just work" for most people.

[taylor.smock]

As a heads up, it looks like wiremock will not be providing explicit support for Java 21 in their 2.x series, and won't be supporting Java 8 in their (as yet unreleased) 3.x series. See ​https://github.com/orgs/wiremock/projects/5 for details.

I don't think we'll have issues using their 2.x series with Java 21, but I wanted to bring it up for when we add Java 21 to the build system. Just in case it consistently fails.

Anyway, other features in newer Java versions that I discovered when reviewing a PR on GitHub:

• Predicate.not (Java 11)
• Math.clamp (Java 21, not Java 17, so we still wouldn't be able to use it; we an the equivalent in Utils)

Beyond that, it looks like I might have to correct myself.
In comment:133, I said "I don't think record by itself will be a significant memory reduction", but a ​JDK developer did some benchmarks which show (minimally) a CPU speed increase. I would anticipate that this is due to "constant folding of instance fields", but the JVM might be doing some other optimizations due to being able to "trust" the record classes.

As a side note, Java 21 is being released in 5 days.
Do we want to move all of our installers to Java 21 once that happens, or do we want to wait a month or so?

Additional notes:

• Most URL constructors are deprecated (do we want to move everything over to URI, keeping URL methods for legacy reasons?)
• SecurityManager is deprecated -- its removal will fix a whole class of bugs with respect to OpenWebStart

[taylor.smock]

When we move to Java 9+ and start writing a module-info.java file, what module name are we going to want to use?

• de.openstreetmap.josm (see attachment:17858.auto_module_name.patch​, matches our domain name)
• org.openstreetmap.josm (matches the package naming scheme we are currently using)

I'd like to add the Automatic-Module-Name to our MANIFEST.MF so that plugins can start writing their own module-info.java files.

Here is a rundown of the current state of our dependencies:

Dependencies with a set module name

• jmapviewer: org.openstreetmap.gui.jmapviewer (see #23114, needs a release and update in source:trunk/ivy.xml​)
• org.apache.commons:commons-compress: org.apache.commons.compress
• org.tukaani:xz: org.tukaani.xz
• ch.poole:OpeningHoursParser: ch.poole.openinghoursparser

Dependencies without a set module name

Unused/to be removed:

• javax.json:javax.json-api -> replaced by jakarta.json:jakarta.json-api (kept for backwards compatibility)
• org.glassfish:javax.json -> replaced by org.eclipse.parsson:parsson (kept for backwards compatibility)
• com.adobe.xmp:xmpcore: We don't actually use this in JOSM core; it is a dependency of com.drewnoakes:metadata-extractor though.
• oauth.signpost:signpost-core: I'm looking at removing this come 2024 since the OSM website considers OAuth 1.0a to be deprecated.

Requires merge and/or release and update:

• org.apache.commons:commons-jcs3-core: Should be fixed in their next release (org.apache.commons.jcs3, probably not 3.2, which came out yesterday. Caveat: their next release will probably require Java 17)
• com.drewnoakes:metadata-extractor: Probably com.drew.imaging, see ​https://github.com/drewnoakes/metadata-extractor/issues/566 (I should probably ping this again)

Deprecated/unmaintained/unlikely to have a release in the near future:

• com.google.code.findbugs:jsr305: Near as I can tell, this is deprecated (especially since it uses the javax root package). ​jspecify is (​apparently) using a fork of ​checker framework with the intent of merging it back in. There seem to be two major alternatives for nullness annotations, the aforementioned checkerframework, jspecify, and Jetbrain's annotation library.
• com.formdev:svgSalamander: It looks like either we will have to pick up the release process or move to a newer ​Java SVG library -- the author of svgSalamander apparently has ​issues releasing to maven central, and the com.formdev guys have moved to JSVG. I did just get JPMS information added to svgSalamander source code though.

Other:

• org.webjars.npm:tag2link: This is really just a "resource" jar.

[taylor.smock]

Do we have some updated Java Version usage stats? Specifically for Windows (28% on Java 8 6 months ago)? The last update had Mac at <9%, and I'm not that worried about Linux/BSD users (I expect 90%+ of both BSD and Linux users to be using a package manager, so I expect their Java 8 usage to rapidly decrease once package maintainers update their minimum Java version).

[stoecker]

Raw data (percent, cumulated percent) Timeframe approx 1 day.

Version with at least 1%:

All     21( 4%,  4%) 20( 2%,  6%) 19( 1%,  7%) 18( 1%,  8%) 17(55%, 63%) 16( 1%, 64%) 11(13%, 77%)  8(23%,100%)

Linux   21( 6%,  6%) 20( 2%,  8%) 19( 2%, 10%) 18( 1%, 12%) 17(34%, 46%) 14( 1%, 47%) 11(41%, 89%)  8(11%,100%)
Mac     21(15%, 15%) 20( 1%, 17%) 17(42%, 59%) 16( 1%, 60%) 15( 1%, 61%) 11( 4%, 65%)  8(35%,100%)
Windows 21( 1%,  1%) 20( 2%,  3%) 19( 1%,  4%) 18( 1%,  4%) 17(68%, 72%) 16( 1%, 73%) 11( 1%, 74%)  8(26%,100%)

Linux (27.7%) Mac (13.9%) Windows (58.4%)

[taylor.smock]

Windows hasn't moved much, if at all. :(

I really wish I knew why people were sticking with Java 8.

I guess I'll just keep bumping the min Java versions for plugins when I make significant modifications.

[gaben]

Replying to taylor.smock:

I really wish I knew why people were sticking with Java 8.

I'm still using Java 8 on Windows, AMA.

In short, every time I tried to move off, I ran into various UI issues across different apps. Java 8 is rock-solid.

[stoecker]

If we get trouble to support 8, then we'll migrate to 11 and users will need to adapt. And trouble is already starting. E.g. Jenkins is also Java >= 11 only.

[taylor.smock]

Also:

• error-prone only supports Java 11+ (and probably soon to be Java 17+) as the compiler
  • I've got a way to use error-prone 2.10 on Java 8 and error-prone 2.22 on Java 11+, but there are a lot of new issues I need to go through and fix. They do claim to support the --release flag, so we could modify the job to compile with Java 11 (or later) but run the tests with Java 8. But that might require changes to the ant targets -- I don't think ant is "smart" enough to avoid recompiling. Maybe a test-java-version variable or something.
• checkstyle only supports Java 11+
  • I can probably do the same thing for checkstyle that I'm doing for error-prone
• wiremock 3 dropped Java 8 support. I probably won't be able to update this test dependency going forward (until we move off of Java 8 or only run tests on Java 11+).

We can technically compile only on Java 17 (or later) and use the --release javac option to compile against Java 8 (which we already do).

[stoecker]

Replying to taylor.smock:

We can technically compile only on Java 17 (or later) and use the --release javac option to compile against Java 8 (which we already do).

That usually only works correct if you also compile against the Java 8 classes. And that means you need to install it. I don't think that reduces the trouble ;-)

[taylor.smock]

Replying to stoecker:

That usually only works correct if you also compile against the Java 8 classes. And that means you need to install it. I don't think that reduces the trouble ;-)

From ​JEP 247

automatically configures the compiler to produce class files that will link against an implementation of the given platform version

What javac --release does is set --source, --target, and --bootclasspath. The important one is the latter; it is what prevents us from accidentally using newer methods/classes. The source of that definitions is $JAVA_HOME/lib/ct.sym. It is a maintenance burden for upstream Java, which is why Java 21 gives a deprecation warning when compiling for Java 8 -- they only support N-3 LTS versions.

See also #20920.

[stoecker]

Did the behaviour change somewhen? Because last time I tried I got results which were old java byte-code but still didn't work.

[taylor.smock]

Replying to stoecker:

Did the behaviour change somewhen? Because last time I tried I got results which were old java byte-code but still didn't work.

It must have if it wasn't previously working.

Here is how I tested:

public class Test {
  public static void main(String... args) {
    System.out.println(" ".isBlank()); // isBlank -> isEmpty later, to check javac 17 compiling to java 8 bytecode
  }
}

$ java -version
openjdk version "17.0.8.1" 2023-08-24 LTS
OpenJDK Runtime Environment Zulu17.44+53-CA (build 17.0.8.1+1-LTS)
OpenJDK 64-Bit Server VM Zulu17.44+53-CA (build 17.0.8.1+1-LTS, mixed mode, sharing)
$ JAVA_HOME="$(/usr/libexec/java_home -v 1.8)" java -version

openjdk version "1.8.0_382"
OpenJDK Runtime Environment (Temurin)(build 1.8.0_382-b05)
OpenJDK 64-Bit Server VM (Temurin)(build 25.382-b05, mixed mode)

$ javac --release 8 Test.java && JAVA_HOME="$(/usr/libexec/java_home -v 1.8)" java Test
Test.java:3: error: cannot find symbol
    System.out.println(" ".isBlank());
                          ^
  symbol:   method isBlank()
  location: class String
1 error

$ javac --release 11 Test.java && java Test
true

If I then change isBlank to isEmpty and run the same command for the java 8 test

$ javac --release 8 Test.java && JAVA_HOME="$(/usr/libexec/java_home -v 1.8)" java Test
false

[stoecker]

Looking at the recent changes and the increasing usage of Java 11 components I think we'll change our approach. We'll switch to Java 11 beginning next year (so 23.12 will be the last Java 8 release)!

[skyper]

There is no java 21 available for stable Debian 12 (bookworm). Until openjdk-21 might be available in backports the plugins requiring java 21 won`t work.

[taylor.smock]

I'm looking at updating the debian packages to depend upon Java 17.

@stoecker: Do you know off the top of your head how our debian packages are built?

[stoecker]

With a Makefile I think. I'm not totally sure about the Debian/Ubuntu build process since for my own stuff I abstracted it away and create an RPM like spec file format and later forgot the details, but I think the Makefile in apt-source directory builds the packages ;-)

Changing for other Java versions probably only means to extend the control file? We could add java21 and 17 there. Assuming that | means OR that would make it future proof ;-) Please provide the correct names for that (do the packages follow the same naming or did it change like it usually does).

cat DEBIAN/control

Package: josm-latest
Version: 1.5.svn{{VERSION}}
Section: utils
Maintainer: josm developers <josm-dev@openstreetmap.org>
Homepage: https://josm.openstreetmap.de
Priority: extra
Architecture: all
Depends: openjdk-11-jre | java11-runtime | openjdk-8-jre | java8-runtime,
         proj-data, fonts-noto, openjfx
Description: Editor for OpenStreetMap (daily development snapshot)
...

[taylor.smock]

I was going to add default-jre (>= 2:1.17) -- I don't know what it is for anything earlier than Debian Buster (oldoldstable) which is 2:1.11.

I'll just manually repackage it for testing purposes, since I was running into issues with dpkg-buildpackage.

[sebastic]

You only need to specify a single jre package with the alternative dependency using the minimum runtime version, the java<N>-runtime virtual package is provided by all relevant jre packages (e.g. ones created by java-package).

If you want to get users to upgrade to openjdk-17 you can change the dependencies to:

Depends: openjdk-17-jre | java8-runtime,
         proj-data, fonts-noto, openjfx

That does not automatically change the alternatives, for that update-java-alternatives should be used.

[stoecker]

Remember this needs to work for new and old Debian and Ubuntu!

[stoecker]

Thought we probably could drop some releases (especially Ubuntu). I'm pretty sure some of them are already out of support...

currently it's:
mantic lucid natty oneiric precise quantal raring saucy trusty utopic vivid wily xenial yakkety zesty artful bionic cosmic disco eoan focal groovy hirsute impish jammy kinetic lunar noble

[taylor.smock]

I'm in the process of manually updating the file in a current archive release so I can use it for testing. And documenting how I'm doing it so someone looking at svn history knows how to do it in the future.

1. curl -O https://josm.openstreetmap.de/apt/pool/universe/j/josm-latest/josm-latest_1.5.svn18887_all.deb
2. ar x josm-latest_1.5.svn18887_all.deb control.tar.xz
3. tar xf control.tar.xz control
4. Edit control to match new control
5. xz --decompress control.tar.xz && tar uvf control.tar ./control && xz control.tar
6. ar r josm-latest_1.5.svn18887_all.deb control.tar.xz

Thought we probably could drop some releases (especially Ubuntu). I'm pretty sure some of them are already out of support...

Current versions in support:

• 23.10 Mantic (2024-07-01)
• 23.04 Lunar (2024-01-20)
• 22.04 Jammy (2027-04-01 for standard LTS, 2032-04-02 for extended LTS)
• 20.04 Focal (2025-04-02 for standard LTS, 2030-04-02 for extended LTS)
• 18.04 Bionic (2028-04-01 for extended LTS)
• 16.04 Xenial (2026-04-02 for extended LTS)
• 14.04 Trusty (2024-04-02 for extended LTS)

[sebastic]

I recommend to not support any Ubuntu release that's out of standard support: ​https://wiki.ubuntu.com/Releases

That leaves you with the two most recent LTS releases: focal & jammy

And two non-LTS releases: lunar & mantic

openjdk-17 is available since bionic: ​https://launchpad.net/ubuntu/+source/openjdk-17

[stoecker]

Hmm, are these all Ubuntu. No Debian?

[taylor.smock]

Debian has

• Buster
• Bullseye
• Bookworm
• Trixie
• Sid

I think debian users use alldist.

[stoecker]

So "alldist trusty xenial bionic focal jammy lunar mantic" should remain. sebastic would also drop first 3, but 3 more aren't so much when I drop 21 ;-)

[stoecker]

Stripped the distros.

[taylor.smock]

I've verified the control changes (as in, "does JOSM install and run") on:

• Buster (oldoldstable). Note: There was an issue during ca-certificates install which is not our problem. Retrying worked. Installs Java 11 instead of Java 17 with default-jre (>=2:1.17), probably because it also matches java8-runtime. EOL is 2024-06-30.
• Bullseye (oldstable)
• Bookworm (stable)
• Trixie (testing)
• Sid (unstable)
• 18.04 Bionic
• 20.04 Focal
• 22.04 Jammy
• 23.04 Lunar
• 23.10 Mantic

We should drop trusty -- it does not have Java 8 in their repositories.
We should also drop xenial, since it uses Java 9 for default-jre, and Java 9 has been EOL for quite some time (Java 11 is not in xenial's repositories). In any case, the version of Java 9 used by xenial seems to have issues (like no --add-modules).

Control changes: Depends: default-jre (>= 2:1.17) | java8-runtime,

Test setup: podman run -it --rm --volume $(pwd):/mnt ${IMAGE} /usr/bin/bash -c "apt update && apt install -y /mnt/josm-latest_1.5.svn18887_all.deb && josm-latest --version && josm-latest --status-report"
See comment:157 on how the deb file was generated.

[stoecker]

All done.

[taylor.smock]

In 18889/josm:

See #17858: Add Java 17 as an option for the Java version in deb packages

By depending upon default-jre with a version specification, we can avoid
hardcoding many specific Java JREs versions via |. Instead, we just indicate
that the default-jre must be >= 17 or the currently installed version must be
Java 8 compatible. This will change on 2023-12-31 when Java 8 support is dropped.
See #17858 for details.

This was verified via the following in an Ubuntu 23.04 container:

1. curl -O https://josm.openstreetmap.de/apt/pool/universe/j/josm-latest/josm-latest_1.5.svn18887_all.deb
2. ar x josm-latest_1.5.svn18887_all.deb control.tar.xz
3. tar xf control.tar.xz control
4. Edit control to match new control
5. xz --decompress control.tar.xz && tar uvf control.tar control && xz control.tar
6. ar r josm-latest_1.5.svn18887_all.deb control.tar.xz

The resulting deb package was then tested against Debian Buster, Bullseye,
Bookworm, Trixie, and Sid, along with Ubuntu 18.04, 20.04, 22.04, 23.04, and
23.10 with
podman run -it --rm --volume $(pwd):/mnt ${IMAGE} /usr/bin/bash -c "apt update && apt install -y /mnt/josm-latest_1.5.svn18887_all.deb && josm-latest --version && josm-latest --status-report"

[skyper]

Something does not work as there is no newer josm-latest package available than r18887.

[taylor.smock]

I'm going to go out on a limb and assume that whatever we use to build the deb packages didn't like the changes. They are valid (I did test the changes manually since I have no clue what we actually use to build the packages). I didn't use the standard debian build tools for testing since I didn't know which one we were using, and the first one I tried didn't like the original control file.

@stoecker: How would you feel about building the deb packages on GitHub CI? If you are worried about vendor lock-in, ​Gitea has a CI workflow that is compatible with GitHub actions.

[sebastic]

The official Debian buildds (using sbuild) don't take alternative dependencies into consideration when installing build dependencies, so when default-jre (>= 2:1.17) isn't installable the build will fail. Other resolvers like those in pbuilder/cowbuilder chroots fall back to the alternative java8-runtime virtual package as expected.

[taylor.smock]

sbuild might be the one I tried first -- it fails for various reasons:

1. No changelog (I made a local one for testing)
2. Needs Source field in control

So I highly doubt we are using sbuild.

I just tried dpkg-deb (from dpkg), and that largely worked. All I had to do was "fix" the Version field.
ant dist && cp dist/josm-custom.jar native/linux/latest/usr/share/josm-latest/josm-latest.jar && dpkg-deb --build --root-owner-group native/linux/latest/ test.deb

Again, I have no clue what we are actually using.

[stoecker]

As far as I see we're using "dpkg -b" to build the package.

The issues seems not with the build, but the repo. I simply deleted the old stuff and reprepro complained that I need to cleanup. We'll see if the next release works error-free.

How would you feel about building the deb packages on GitHub CI?

Things which work on the Linux server wont go to external servers. Only stuff which wont run on the Linux server like the Windows and Mac packages.

[taylor.smock]

The issues seems not with the build, but the repo.

Good to know. My standard assumption when I just changed something that is now failing is that it is something I did.

Things which work on the Linux server wont go to external servers. Only stuff which wont run on the Linux server like the Windows and Mac packages.

I'm still tempted to add it to the GitHub actions, mostly as a sanity check. Although we probably won't need it for several years.

[anonymous]

Replying to stoecker:

Looking at the recent changes and the increasing usage of Java 11 components I think we'll change our approach. We'll switch to Java 11 beginning next year (so 23.12 will be the last Java 8 release)!

I updated the StartupPage version check to tell Java 11 is minimum requirement.

[taylor.smock]

In 18985/josm:

Fix #23355: Sanity check JVM arguments on startup

See #17858: JOSM will no longer continue running if the user is on an unsupported
Java version (for this commit, older than Java 11; message indicates Java 17).
This does update the link for Azul from Java 17 to Java 21 as well.

In order to (hopefully) reduce confusion, the webstart and Java update nags will
also be reset in the event that JOSM will exit due to old Java versions. This is
mostly so that users will get the messages to update to OpenWebstart or the
appropriate Java link for their platform and architecture.

Additionally, this will (hopefully) reduce the number of tickets we have to close
due to missing JVM arguments by informing users of the missing arguments at startup.

[taylor.smock]

In 19018/josm:

See #17858/#23564: Update Java versions in Linux start scripts

Additions:

• Java 19-20 (EOL, added for completeness)
• Java 21 (preferred LTS)
• Java 22 (latest Java version)
• Java 23 (next Java version)

Removals:

• Java 8, 9, and 10: We no longer support Java < 11.

[skyper]

Replying to skyper:

There is no java 21 available for stable Debian 12 (bookworm). Until openjdk-21 might be available in backports the plugins requiring java 21 won`t work.

Well, openjdk-21-jre is in testing (trixie) but still not even sources for openjdk-21 are available in current stable (bookworm) which means that the next 14 month all plugins requiring java21 are not usable for Debian users. Please, keep that in mind when updating plugins. Thanks a lot.

[stoecker]

I changed server installation. Java8 is out. Tomorrow it will start building with Java 11.