icedtea-web compatible with OpenJDK11

[Jiri Vanek]

On 09/05/2018 08:45 PM, Laurent Bourgès wrote:
> Dear Jiri,
> 
> Sorry for my late response, I was working hard on fixing IcedTea-Web 1.7 hanging problem (EDT) on 
> OpenJDK11.
In last three months the usual delay was a month :) So NP.

> I reported this problem to awt-dev list and wrote a simple reproducer:
> http://mail.openjdk.java.net/pipermail/awt-dev/2018-September/014264.html
> http://mail.openjdk.java.net/pipermail/awt-dev/2018-September/014269.html
> 
Great!

> Today I managed fixing the AWT deadlock by using a workaround:
> use a specific thread pool within the Main thread group (not JNLP's AppContext) to relay the EDT 
> invocations (invokeLater / invokeAndWait) as I figured out that SplashScreen & DownloadIndicator 
> dialogs were using the JNLP AppContext
> => 2 concurrent AppContexts were used and 2 event dispatcher threads led to deadlocks in AWT 
> SequencedEvent processing (Window Gain / Lost focus).
> With that workaround, all EDT actions from ITW are delivered to the Main AppContext and it works well !
> Of course, the application is still launched using its own separate AppContext, so the bug is still 
> critical if any ITW dialog (permission, file check ?) can happen while the application is running.
> My apps use allPermissions so I can not test. Do you have any example app ?

I do not have direct reproducer around.
There is list of applications to play with: https://icedtea.classpath.org/wiki/IcedTea-Web-Tests 
They should help.

For the EDT pathc, I'm unable to decide if it could affect security. ITW have some 1800 integration 
tests, to check sandbox. I was not running that since Firefox removed NPAPI, but I think it is 
worthy to reincarnate them again. I will definitely require myself [read I, jvanek will do it] to 
run the wholes suite on both head and 1.7. Ideally both before and after pushes
> 
> Moreover, I added the well known ThreadCheckingRepaintManager & TracingEventQueue to detect and fix 
> any EDT violation: it is OK, so I can remove these classes.

This decision is up to you. I do not have preferences on theirs existence in repo.  Thery can be handy.
> 
> Now it is time to clean up the code and submit patches !
> 
> Please see my comments in the text:
> 
> Le mer. 29 août 2018 à 16:37, Jiri Vanek <jvanek at redhat.com <mailto:jvanek at redhat.com>> a écrit :
> 
>     On 08/29/2018 03:37 PM, Laurent Bourgès wrote:
>      > Hi,
>      >
>      > I am performing tests with icedtea-web-1.7 (cloned) + OpenJDK 11 EA 28 (oracle build) to
>     evaluate if
>      > it can provide a suitable alternative to Oracle JDK on major platforms (win/mac/linux).
> 
> 
>     Thank you *very* much for doing this!
> 
> 
> Thank you all for all the good work on ITW. It made huge progresses in 10 years (fedora 11).
> By the way, the control panel is great and it would be wonderful to open it via 'javaws -viewer' 
> (like O... jdk) or any other option.

Should be doable. Will try it.
added: https://icedtea.classpath.org/wiki/IcedTea-Web#IcedTea-Web_1.8
> 
> Another issue I could do: JNLPClassLoader does many useless class/resouce download queries if many 
> child loaders are in action... let's see later.

Is it really so bad? Hmm. Worthy of reproducer.
> 
> 
>      >
>      > FYI I compiled icedtea-web-1.7 without plugin and modified the generated shell scripts (sh /
>     bat) to
>     Have you disabled both native and java parts of plugin, or only native?
> 
>      > successfully run my javaws applications on jdk11 (win/linux).
>      > For example: http://jmmc.fr/~bourgesl/Aspro2/ <http://jmmc.fr/%7Ebourgesl/Aspro2/>
>     <http://jmmc.fr/%7Ebourgesl/Aspro2/>
>     Great! I'm really glad to hear this. when I was improving ITW launcher to survive jigsaw, usually
>     ITW itself run well, but nearly all client applications were broken because of illegal access.
> 
> 
> Yes our apps work almost on jdk11 (few third-party libs are causing issues).
> I am very happy ITW works so well on jdk11 (linux), but jigsaw args are missing in windows scripts, 
> so I will rewrite them.
> 
> PS: how to set jigsaw CLI args in JNLP ? Did you try ? Does it change the java command line ?

Never try that. Still both javaws and jnlp's xml allows to pass -J arguments, so it should work. but 
never tested.  And of course via jnlp it will cause fork.
> 
> 
>     Thank you very much for improving and verifying the sh/bats.
> 
> 
> I tested linux shell scripts and only windows command line (derived from linux CLI). It works also 
> on mac (latest update) with ojdk11-28.

Wou. Never tried mac before! That is making some complications to the native launchers. No mac around:(
> 
> 
>     I'm looking forward to see changes you mad!
> 
>     For future (1.8) ITW sh/bat launchers are being rewritten to rust, to avoid duplicated bat/sh code.
>     The rust launchers are based 100% on current sh/bat, with improve in prebuild, downloadable
>     packages
>     for both windows and linux (currently linux is distribution only, but windows is downloadable)
> 
>      >
>      > I had many times security dialogs hanging (XAWT lock issue), so I started diagnosing deadlocks
>      > (jvisualvm) and fixed several EDT violations.
> 
>     This is exceptional. I'm aware of those locks, but never had solid reproducer nor time to debug it
>     properly.
> 
> 
> Thanks. It is definitely 'done' for javaws, maybe some issue remains in control panel ...

They are not performance not mission critical. Also ControlPanel is not so heavily multi threaded.
> 
> 
>     This patch will be heavily appreciated, and will be backported to 1.7 and will make candidate for
>     another 1.7 release.
> 
> 
> As I only worked on 1.7 code base, I could use webrev or just patches.
> Do you prefer me to rebase on the bleeding edge repo ?

Webrev or patch do not meter. Both are good.
Yes please. It will be important to push and test first to head.
But I do nto mind to finish review on 1.7 and adapt afterwards.
> 
>      >
>      > If I use -headless, -Xnofork or -Xtrustall, I have no hanging issue.
>      > - Do you have any idea why separate AppContext are absolutely needed that maybe cause XAWT lock
>      > deadlocks ?
> 
>     ITW and sandboxed application should be as separate as possible. Isnt Aseparate appcontext part of
>     it? Maybe not. When I inherrited ITW, this was already there, so I keel in the habit.
> 
>      > - Running jdeps -jdkinternals (from jdk10) on the netx library reports lots of internal jdk
>     usages,
>      > do you plan fixing them ? I could help
> 
>     This should be fixed for future ITW. Any patch is welcomed. This is currently moreover one (busy)
>     man show. So only small steps forward with development, a bit bigger with bugfixes,  but much
>     smaller in cleanups.
> 
> 
> As it is not blocking, I agree it can stay as it is. However some future class removals in jdk could 
> hurt.

Yes. It will need to be fixed, workaround or the feature removed.  I usually deal with those once 
thy hit.
> 
> 
>      >
>      > 1. Are you interested by some patch in netx code ?
> 
>     Definitely.
> 
> What form ? As webrev on cr.ojdk... ?

Both are ok:) Even plain email is.
> 
> 
>      > what is your contribution process (I am an openjdk commiter) ?
> 
>     ITW have different contribution process, as it is not hg.openjdk.net <http://hg.openjdk.net>,
>     but classpath.org <http://classpath.org>.
>     Looking to the quality of your research and fact that you are OpenJDK commiter, mean that you will
>     post patch or two, I review them and will commit them on your behalf, and then I will nominate you
>     for "classpath hacker" (poetic name:). Doing so, you will get push access to classpath repos.
> 
> 
>      >
>      > 2. In latest icedtea-web repo, the shell scripts were removed. So what is your plan for
>     windows /
>      > mac compatibility ?
> 
>     As written above - they are replaced by rust launchers. They are not removed, jsut moved, and
>     marekd
>     as deprecated if you built only them (configure switches). Still I intend to maintain them at least
>     for itw 1.8 and maybe more further, as on Linux they are exceptionally useful. But in some time,
>     they are likely to die, yes:(
> 
>     I have never run ITW on mac, nor have any feedback on its usage. Likely with oracle removing
>     javaws,
>     it can change. I have some machines rented from redhat to test ITW deeply on linux, and well, alsoa
>     biton windows, but I'm far from being  expereinced window one. Still some effort is done to
>     stabilise it. (many silently known issues for ITW on windows, but no more show stoppers). Main
>     kudos
>     for win port goes to Jacob, Alex and Michal and few students I had misused for it.
> 
> 
> Great. I confirm my tests worked on mac too. Having an openjdk with installer on mac would be 
> awesome. My scientific users are mainly linux/mac addicts.

Argh. Mac again.  That is area someone else will have to investigate.  I have never touched that, 
nor do have any experience in support.
Mac can not live with untared binary?

> 
> 
>      >
>      > Finally IcedTea-Web seems a promising alternative on windows for OpenJDK 11, so I think openjdk
>      > distributions should adopt it when OpenJDK 11 will be released.
>      >
> 
>     Thank you for kind words, and an exceptional work you did on launchers, jdk11 testing and EDT
>     violations.  I'm looking forward to see the patches, and I hope I will make theirs upstreaming as
>     simple comfortable as possible to return the value.
> 
> 
> Thank you for your very positive feedback, I will propose asap patches on:
> 1. Netx code (EDT)
> 2. shell scripts (linux then win)
Please note the small patch I sent to Head. It can affect your work.
I guess you noticed that batfiles are in really poor state. And I'mnot sure if it si worthy to fix. 
But if you do, thank you.

Once those few fixes are in, I think you will wish 1.7.2 release. I can definitely do it, and there 
landed few more good patches in past, and 1.8 is still far from release.

Thanx a lot!

   J.

-- 
Jiri Vanek
Senior QE engineer, OpenJDK QE lead, Mgr.
Red Hat Czech
jvanek at redhat.com    M: +420775390109