Bugzilla – Full Text Bug Listing
| Summary: | [console] IOConsoleInputStream.available() should not return -1 | ||
|---|---|---|---|
| Product: | [Eclipse Project] Platform | Reporter: | Paul Pazderski <paul-eclipse> |
| Component: | Debug | Assignee: | Paul Pazderski <paul-eclipse> |
| Status: | VERIFIED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | P3 | CC: | loskutov, sarika.sinha |
| Version: | 4.12 | ||
| Target Milestone: | 4.12 M3 | ||
| Hardware: | All | ||
| OS: | All | ||
| See Also: |
https://bugs.eclipse.org/bugs/show_bug.cgi?id=307309 https://git.eclipse.org/r/141011 https://git.eclipse.org/r/141013 https://git.eclipse.org/c/platform/eclipse.platform.debug.git/commit/?id=e464979869b087a1af7c21fa4b54136db563bda4 https://git.eclipse.org/c/platform/eclipse.platform.debug.git/commit/?id=6fd317d247b94e2e4184ff79f665c7eb848ceb6e |
||
| Whiteboard: | |||
| Bug Depends on: | |||
| Bug Blocks: | 545769, 552030 | ||
|
Description
Paul Pazderski
(In reply to Andrey Loskutov from bug 307309 comment #19) > Regression is: I see that some ant build jobs never terminate now, because > of not closed input streams, so they are logged as: > > 3eclipse.buildId=4.12.0.I20190421-1800 > java.version=11.0.2 > java.vendor=Oracle Corporation > BootLoader constants: OS=win32, ARCH=x86_64, WS=win32, NL=de_DE > Command-line arguments: -os win32 -ws win32 -arch x86_64 -data > C:\workspaces\platform > > org.eclipse.core.jobs > Warning > Mon Apr 22 22:55:58 CEST 2019 > Job found still running after platform shutdown. Jobs should be canceled by > the plugin that scheduled them during shutdown: > org.eclipse.debug.internal.ui.views.console.ProcessConsole$InputReadJob Unfortunately I could not reproduce this warning in a normal way. Neither with running an ant target in eclipse. I already considered bug 307309 comment #10 in my initial patch and also saw the problems in git history. The warning says the job is still running. The job does little more than reading from input stream in a loop until the read returns -1 (or the stream is null but this should never happen before the other condition). The ProcessConsole either reads from a FileInputStream (irrelevant since I changed nothing here) or from IOConsoleInputStream. The read() either returns the available data (the loop continues) or it blocks to wait to wait for more data until it is closed and no more data is available, than read() returns -1 and the loop ends. I see no flaws at this point. Neither with or without my patch. So the only situation the job is 'still running' I can imagine is when the stream is not closed and the read never returns. The ProcessConsole close the stream if the process already terminates before console initialization or if the terminate event is received or if the console is disposed. The warning is emitted by the JobManager shutdown (I found this message nowhere else). On shutdown, my eclipse for testing first stops the DebugUIPlugin which disposes the process consoles which again stops any pending InputReadJobs. The JobManager is stopped afterwards and all InputReadJobs already terminated. My only success to reproduce this warning was to trigger a race condition in ProcessConsole init so that the process can terminate but the console miss the termination (open a new bug for this?) and after that manually shutdown the JobManager (which is discouraged but possible). Actually the JobManager tries to stop the job and produce the warning after this attempt failed. So maybe the problem is that InputReadJob is unstoppable. The warning says the plugin should cancel the job but DebugUIPlugin does cancel the job (or better dispose the consoles) and I found no problems there nor how my change could interfere it. New Gerrit change created: https://git.eclipse.org/r/141011 New Gerrit change created: https://git.eclipse.org/r/141013 Long story short I could not reproduce this problem and you may retry my change https://git.eclipse.org/r/#/c/141011/ in combination with this new change https://git.eclipse.org/r/141013 which makes the InputReadJob cancelable. (In reply to Paul Pazderski from comment #4) > Long story short I could not reproduce this problem and you may retry my > change > https://git.eclipse.org/r/#/c/141011/ > in combination with this new change > https://git.eclipse.org/r/141013 > which makes the InputReadJob cancelable. I will check this later. I have lot of SDK projects in my workspace, some of them are using external ant builders, and the effect was that after a workspace rebuild (I update the platform often) some of those builders output left "hanging" in the Console view and I was not able to "remove" them and saw those warnings on shutdown. I had no chance to verify yet if the revert fixed this or not, but looking on the recent platform changes only the available() change was related. Gerrit change https://git.eclipse.org/r/141013 was merged to [master]. Commit: http://git.eclipse.org/c/platform/eclipse.platform.debug.git/commit/?id=e464979869b087a1af7c21fa4b54136db563bda4 This is the script that sometimes produces the "hanging" input job: rt.equinox.p2/bundles/org.eclipse.equinox.p2.publisher.eclipse/scripts/buildExtraJAR.xml The script is configured to run on project build, interestingly I do not always see the console opening, probably because it is terminating faster as we can show the console. The many short processes may in fact had triggered bug 546710 which was, for my test, a precondition to produce the warning. I'll try to reproduce it again with the new information. Gerrit change https://git.eclipse.org/r/141011 was merged to [master]. Commit: http://git.eclipse.org/c/platform/eclipse.platform.debug.git/commit/?id=6fd317d247b94e2e4184ff79f665c7eb848ceb6e The termination problems came out to be unrelated with this fix. @Paul, Please verify this bug with the new I build. Verified with Build id: I20190520-1805 |