okhttp fails sporadically on android 4.x mediatek devices [70909581]

Fixed

Bug

device specific JIT bugs affecting D8

Status Update

No update yet.

Description

xi...@gmail.com

created issue #1

Dec 21, 2017 03:54AM

we found this unreducable crashes when we enable d8..

Caused by: java.lang.ClassCastException: int[] cannot be cast to java.lang.String[]
at uo.doInBackground(PromotionTask.java:36)
at android.os.AsyncTask$2.call(AsyncTask.java:288)
at java.util.concurrent.FutureTask.run(FutureTask.java:237)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1112)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:587)
at java.lang.Thread.run(Thread.java:841)

at android.os.MessageQueue.nativePollOnce(Native Method)
at android.os.MessageQueue.next(MessageQueue.java:138)
at android.os.Looper.loop(Looper.java:125)
at android.app.ActivityThread.main(ActivityThread.java:5503)
at java.lang.reflect.Method.invokeNative(Native Method)
at java.lang.reflect.Method.invoke(Method.java:515)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:976)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:792)

设备：
LEPHONE W7
剩余内存：
513MB
CPU用量：
44%
设备剩余空间：
1502MB
CPU型号：
sc8830
CPU指令集：
ARMV7
是否ROOT：
已ROOT
剩余电量：
59%
系统版本：
Android 4.4.4

GPS：
关闭
应用UI朝向：
竖屏
运营商：
中国联通
接入方式：
WIFI

and this only show in android 4.4.4 and 4.1.2

Comments

ar...@google.com <ar...@google.com> #2Dec 21, 2017 05:04AM

Status: Won't Fix (Infeasible)

This website is for developer issues with the AOSP Android source code and the developer toolset, not user support.
We have absolutely no control or visibility over other OEM's source code or updates. Please reach out to specific OEM for such issues.

xi...@gmail.com <xi...@gmail.com> #3Dec 21, 2017 06:43AM

oop...Still thank you.

as your words

d8 may be hard for expend in chinese developers..

we hope it stable

xi...@gmail.com <xi...@gmail.com> #4Dec 21, 2017 06:49AM

whatever, I will still put some details here

we found it in the use of

"Variable length parameter" method..

we put "1" "2" .... into params

but we define it as
public void test(String ...){
}

then we got classCastException: int[] cannot be cast to java.lang.String[]

maybe it is not a bug of d8 ..

xi...@gmail.com <xi...@gmail.com> #5Dec 22, 2017 01:55AM

We had proved that.. The crashes has direct association with d8.
we remove it and crashed gone..

Maybe, You need pay more attention with this...

ar...@google.com <ar...@google.com> #6Dec 22, 2017 04:59AM

Assigned to ar...@google.com.

For us to further investigate this issue, please provide the following additional information:

Steps to reproduce
Please provide sample project or apk to reproduce the issue.
Also kindly mention the steps to be followed for reproducing the issue with the given sample project or apk.

Frequency
How frequently does this issue occur? (e.g 100% of the time, 10% of the time)

Expected output
Could you please explain the expected behavior.

Current output
Could you please explain the current behavior.

to...@gmail.com <to...@gmail.com> #7Dec 22, 2017 05:38PM

Here is sample code:

https://github.com/square/okhttp/issues/3641
It looks that it happens only on some Android 4.4
I had the same problem, with android.enableD8=false problems are gone.
Definitely a huge problem.

ar...@google.com <ar...@google.com> #8Dec 27, 2017 09:01AM

We are not able to reproduce the issue on Nexus 5 device with Android 4.4.2 build.
Attached the sample project we have used for reproducing the issue.

Can you use this sample project and check whether you are able to reproduce the issue on other devices with Android 4.4 builds?
If you are able to reproduce the issue, please let us know the device and build in which issue was observed.
Also capture a bugreport and share it here.

Below are the steps we followed for reproducing the issue:
1. Install the attached HttpSample.apk in the device.
2. Launch HttpSample application.
3. Tap on "Asynchronous Get" button and observe the behavior.
4. Tap on "Synchronous Get" button and observe the behavior.

We are able to see the responses after step 3 and step 4.

HttpSample.apk

1.6 MB

Download

HttpSample.zip

1 MB

Download

ar...@google.com <ar...@google.com> #9Jan 3, 2018 06:42AM

Please provide the information requested in

comment #8

to investigate further.

to...@gmail.com <to...@gmail.com> #10Jan 7, 2018 11:17PM

I'm not able to reproduce it on any of my devices. I just know that when I released my app with D8 enabled it started to happen to some users and when I turn it off it disappears.

Here are fabric reports with it:

http://crashes.to/s/127ace272e7

http://crashes.to/s/53996523eed

It happens on some Lenovo and Huawei devices and only on Android 4.X.

I think that you (I mean Google) should have also some similar crashes from own apps.

ar...@google.com <ar...@google.com> Jan 10, 2018 09:25AM

Status: New

ag...@google.com <ag...@google.com> #11Jan 18, 2018 10:13AM

Assigned to ag...@google.com.

We have been seeing issues with okhttp on certain Android 4.4 devices. We pushed a fix out for those issues for Android Studio 3.1 where D8 is the default. Looks like you are using Android Studio 3.0? Can you try this with an Android Studio 3.1 canary build to see if it stll reproduces?

I will try out the sample code from

comment #7

as well.

ag...@google.com <ag...@google.com> #12Feb 6, 2018 11:28AM

We have failed to reproduce this issue with many different 4.X devices and emulators. If someone manages to create a reproducible case that they can provide us with we will be happy to investigate. At this point we are dead in the water since the code generated by D8 works on all devices we have tried.

ag...@google.com <ag...@google.com> #13Mar 6, 2018 09:17AM

Reassigned to sg...@google.com.

We finally managed to reproduce this. The Dalvik JIT on these devices has been modified and has bugs in the implementation of long arithmetic and long comparisons. Luckily, the bugs can be worked around by choosing different dex registers. We are working on that change now.

ap...@google.com <ap...@google.com> #14Mar 6, 2018 11:21AM

Project: r8
Branch: master

commit 818c7aa6430961a92e2ccdb1e54b58ad590d872b
Author: Søren Gjesse <sgjesse@google.com>
Date: Tue Mar 06 12:12:28 2018

Extend the check for instructions taking long operands overlapping their result

Support for not overlapping the result of cmp-long with its operands and
extend the current check for add-long, sub-long, or-long, and-long and xor-long
to disallow any overlap between operands and result registers.

All these check now only apply on all versions of dalvik.

Bug: 70909581
Bug: 74084493

Change-Id: I49a4903cc1d707f0ab92ff2a3e018778084b7c18

M src/main/java/com/android/tools/r8/ir/code/Add.java
M src/main/java/com/android/tools/r8/ir/code/And.java
M src/main/java/com/android/tools/r8/ir/code/Or.java
M src/main/java/com/android/tools/r8/ir/code/Sub.java
M src/main/java/com/android/tools/r8/ir/code/Xor.java
M src/main/java/com/android/tools/r8/ir/regalloc/LinearScanRegisterAllocator.java
M src/main/java/com/android/tools/r8/utils/InternalOptions.java
M src/test/java/com/android/tools/r8/R8RunArtTestsTest.java

https://r8-review.googlesource.com/17600

ap...@google.com <ap...@google.com> #15Mar 7, 2018 08:15AM

Project: r8
Branch: master

commit e6f27c14ba1e158988405b4f7565a9646e56043d
Author: Mads Ager <ager@google.com>
Date: Wed Mar 07 08:59:06 2018

Spill a valid register when no valid register candidate can be found.

In rare cases, when all registers candidates are used and the
last register candidate is invalid, we would fail instead of
spilling a valid register to use. This change fixes that.

R=sgjesse@google.com

Bug: 70909581
Change-Id: I5e4fb6e7b37e3c6ff0b52e9be6324e6de8df3ccd

M src/main/java/com/android/tools/r8/ir/regalloc/LinearScanRegisterAllocator.java
M src/test/java/com/android/tools/r8/R8RunArtTestsTest.java

https://r8-review.googlesource.com/17860

ag...@google.com <ag...@google.com> Mar 8, 2018 02:38PM

Marked as fixed.

js...@jackhenry.com <js...@jackhenry.com> #16Mar 28, 2018 07:30PM

Has the fix for this made it into the Android Studio 3.2 Canary yet?

h4...@gmail.com <h4...@gmail.com> #17Apr 4, 2018 07:59AM

It's not yet fixed on 3.2.0-alpha09. According to this comment, the fix is coming over the next couple of weeks

https://github.com/square/okio/issues/323#issuecomment-375208627

js...@jackhenry.com <js...@jackhenry.com> #18May 4, 2018 07:04PM

Any word on whether the fix is into the latest 3.2.0 alpha yet?

ag...@google.com <ag...@google.com> #19May 8, 2018 09:31AM

Joshua, sorry about the late reply on this. Yes, these fixes are available in the latest 3.2 canary 13. The only caveat is if you are using an old version of okhttp:

https://b.corp.google.com/issues/77996377 that fix didn't make it into canary 13 and will be available from canary 14. If you are using okhttp version 3.5 or newer, you should be good to go with the current canary 13.

ra...@gmail.com <ra...@gmail.com> #20Jun 25, 2018 01:08PM

Wich build tools should I use to fix this bug if I'm using a continous build system without an IDE?

ag...@google.com <ag...@google.com> #21Jun 25, 2018 11:03PM

#20, I guess your continuous build system is using the android studio gradle plugin to build your code? You should make sure to use either the gradle plugin for the lastest stable version of studio (3.1.3) or the latest beta of 3.2.