SnapshotException: SnapshotImpl_Error_ObjectNotFound

[ralph-bergmann]

I got the following error and was asked to report it here(?):

05-17 12:00:57.099  17748-18395/me.foo.debug D/LeakCanary﹕ In me.foo.debug:1.0.0-debug (Git SHA: ce19c3b):1001000004.
    * FAILURE:
    org.eclipse.mat.SnapshotException: SnapshotImpl_Error_ObjectNotFound
            at org.eclipse.mat.parser.internal.SnapshotImpl.mapAddressToId(SnapshotImpl.java:1194)
            at org.eclipse.mat.snapshot.model.ObjectReference.getObjectId(ObjectReference.java:38)
            at org.eclipse.mat.snapshot.model.ObjectReference.getObject(ObjectReference.java:42)
            at org.eclipse.mat.parser.model.AbstractObjectImpl.resolveValue(AbstractObjectImpl.java:149)
            at com.squareup.leakcanary.HeapAnalyzer.findLeakingReference(HeapAnalyzer.java:161)
            at com.squareup.leakcanary.HeapAnalyzer.checkForLeak(HeapAnalyzer.java:95)
            at com.squareup.leakcanary.internal.HeapAnalyzerService.onHandleIntent(HeapAnalyzerService.java:57)
            at android.app.IntentService$ServiceHandler.handleMessage(IntentService.java:65)
            at android.os.Handler.dispatchMessage(Handler.java:102)
            at android.os.Looper.loop(Looper.java:135)
            at android.os.HandlerThread.run(HandlerThread.java:61)
    * Reference Key: ee5aac58-f9ad-4b50-86b7-0d0ce10a1f1e
    * Device: asus google Nexus 7 razor
    * Android Version: 5.1.1 API: 22 LeakCanary: 1.3.1
    * Durations: watch=5010ms, gc=125ms, heap dump=2440ms, analysis=20306ms

[pyricau]

Can you upload the heap dump and share the link here?

[ralph-bergmann]

Sorry the heap dump for this crash is in /dev/null... but the next one is yours :-)

[ralph-bergmann]

oooh I found it...
https://drive.google.com/file/d/0Bz_DGujWHPo2Z0FNM2xFYjktN0k/view?usp=sharing

[pyricau]

Seems like a heap dump parsing bug. The FacebookActivity instance is the culprit. HAHA don't see any instance of that, although it does have a memory address for the keyed weak reference, it then can't resolve it to an object id.

Interestingly, MAT can do the job just find. Seems like a bug in the old MAT implementation.

[pyricau]

Here's how I reproduce it: c92a778

I need to look into this more, this will require digging into the core MAT mechanics.

[andkulikov]

i reproduce it too

* FAILURE:
org.eclipse.mat.SnapshotException: SnapshotImpl_Error_ObjectNotFound
at org.eclipse.mat.parser.internal.SnapshotImpl.mapAddressToId(SnapshotImpl.java:1194)
at org.eclipse.mat.snapshot.model.ObjectReference.getObjectId(ObjectReference.java:38)
at org.eclipse.mat.snapshot.model.ObjectReference.getObject(ObjectReference.java:42)
at org.eclipse.mat.parser.model.AbstractObjectImpl.resolveValue(AbstractObjectImpl.java:149)
at com.squareup.leakcanary.HeapAnalyzer.findLeakingReference(HeapAnalyzer.java:161)
at com.squareup.leakcanary.HeapAnalyzer.checkForLeak(HeapAnalyzer.java:95)
at com.squareup.leakcanary.internal.HeapAnalyzerService.onHandleIntent(HeapAnalyzerService.java:57)
at android.app.IntentService$ServiceHandler.handleMessage(IntentService.java:65)
at android.os.Handler.dispatchMessage(Handler.java:102)
at android.os.Looper.loop(Looper.java:211)
at android.os.HandlerThread.run(HandlerThread.java:61)

* Reference Key: 46c532a6-ef64-4566-93de-ff5ea15927be
* Device: Sony Sony D5803 D5803
* Android Version: 5.0.2 API: 21 LeakCanary: 1.3.1
* Durations: watch=5124ms, gc=152ms, heap dump=2092ms, analysis=30342ms

link to dump
https://docs.google.com/file/d/0Bx6U6K76tqX-RFQycUpWYTJ0cDA/edit

[ninja-]

same here

[ninja-]

it's a race condition when the weak reference is still there but the 'object' has been properly gc'd before the heap dump

[ninja-]

#157 fixes this

[pyricau]

@ninja- actually, nope. This isn't a race condition AFAIK. As pointed out above, the heapdump can be parsed fine in MAT and the reference does exists, but it can't be read in HAHA. See c92a778

[SecUpwN]

Thanks for pointing the issue, should be an easy fix (regardless of the number of GC). Cool!

Hoping to see a fix for this soon so that we can re-implement leakcanary into our app again.

[pyricau]

@SecUpwN As I pointed out in the follow up comment, it's actually not easy. You're more than welcome to checkout the heap dump parsing code and fix this bug.

[pyricau]

We'll very likely need to change our heap dump parser anyway, see #160.

[pyricau]

Will be fixed by #219

[xiaolongyuan]

+1

[LeoDZhao]

I also met this issue

[pyricau]

Please update.