Please find the log and Video @ http://go/chrome-androidlogs1/6/690802

satya@ issue still repro on latest M58: 58.0.3029.51

can't do pe-cl bisect, issue doesn't repro on chrome dev.


Thanks!

battun@, can we get any sort of bisect? Per-build? If not, feel free to open this back up.

I can't reproduce this on Samsung SM-G935S Galaxy S7 Edge.

CC'ing Changwan, since this is IME-related.

ntfschr@Issue observed on M54 onwards please find bisect range, can't per-cl bisect.

Note:Issue unable to repro on Samsung Galaxy S7 / MMB29K

Regression range: https://chromium.googlesource.com/chromium/src/+log/54.0.2825.0..54.0.2826.0?pretty=fuller&n=10000

Thanks!

Good build: 54.0.2825.0
Bad build:  54.0.2826.0

Thanks! Changwan, can you investigate?

I'm lowering the priority since this regression has obviously been around for a while.

removing RVG since there isn't much google specific information.

ajith, could you take a look? Note that this is targeting M-60. Thanks!

battun@ mentioned video link is unable to open. Could you share proper link for the video again. Thank you

ajith.v@ please find attached screen record.

Thanks!

Deleted: screen_keyboard.mp4 3.0 MB

	screen_keyboard.mp4 3.0 MB View Download

I tried reproducing in Galaxy J5 Prime SM-G570F. But couldn't reproduce with same steps in the video. I will try to get Galaxy J5 device and try again.

I am using Android System WebView 58.0.3029.83
Build Number: MMB29K, Android M
Device: Samsung Galaxy J5 Prime

Issue is not reproduced. Can we close this issue as it's fixed with latest Android WebView ?

battun@ could you verify #16?

changwan@ Issue still repro on Samsung Galaxy J5(SM-J500F) / MMB29M Android version 6.0.1
I am using Android System WebView 58.0.3029.8. Sharing new screen record

Thanks!

Deleted: Screen_Samsungj5.mp4 1.8 MB

	Screen_Samsungj5.mp4 1.8 MB View Download

As I said, I have only Samsung Galaxy J5 Prime, which is very close in similarity wrt to Galaxy J5 here. So I am not able to reproduce it the issue locally.

battun@ - Could you try on any other Samsung J, A, Note, S series device for reproducing this issue ?

battun@, you mentioned that the issue doesn't repro on chrome dev as of Apr 3. Could you check if M59 / M60 fixed the issue? M59 is already rolled out, so I don't think we need to worry if it's already fixed.

changwan@ Issue observed on Samsung Galaxy J7(SM-J700F) / MMB29Kon latest M61: 61.0.3136.3
Doesn't repro on Samsung Galaxy Grand Prime 4G(SM-G531F) / LMY48B.

Thanks!

I was able to reproduce this issue on Chrome on SM-J700H.

When the text is:
... hello java [program]

When the user presses backspace key, Samsung keyboard calls:

finishComposingText()
sendKeyEvent()

... hello java progra

a few more backspace keys and we get

... hello java prog

then all of a sudden, Samsung keyboard calls

setComposingText [progr]

which appends the text at the end.

Notably, I can reproduce this without the Hebrew text. The minimum test case I found so far is backward deletion on " programm" on contenteditable, such as https://www.w3schools.com/tags/tryit.asp?filename=tryhtml5_global_contenteditable

I suspect that this is a Samsung issue, but I am not sure why this was working before ImeThread flag was set (#7). I'll investigate it further.

This CL in the bisect range looks suspicious:
https://codereview.chromium.org/2121953002

The bisect is wrong. I can reproduce the issue on Chrome Dev 44.0.2403.157.

Seems like this might be a Samsung bug as it's persisted across so many different Chrome changes, although it's curious that we haven't been able to reproduce it in an EditText widget. There must be some behavioral difference in our implementation. Perhaps there's part of the InputConnection API that we haven't implemented yet?

The bug has been passed to Samsung for investigation. If they have an update for us, it will be posted on Google-internal b/70298628.