onboard segfaults

added release-v24.06 ui-mate labels

mentioned in issue #2861 (closed)

By Oliver Smith on 2024-06-16T13:50:32

mentioned in issue #2887 (closed)

By Peter M on 2024-06-16T17:32:05

I also found this when testing edge on xfce4. Reported in #2887 (closed).

By Bryant Mairs on 2024-06-17T05:43:42

marked the checklist item edge (master branch) as completed

By Oliver Smith on 2024-06-17T05:44:13

changed the description

By Oliver Smith on 2024-06-17T05:44:37

added release-edge ui-xfce4 labels

Thanks for setting up this issue report. I hadn't time to look into this yet, too busy with other things.

There is ~~one~~ upstream issue report with not much activity yet [edit: added three more]:

And some more issue reports at Ubuntu with a bit more activity:

https://bugs.launchpad.net/ubuntu/+source/onboard/+bug/2063041
https://bugs.launchpad.net/ubuntu/+source/onboard/+bug/2064775
not sure if this one is related: https://bugs.launchpad.net/ubuntu/+source/onboard/+bug/2066241

I guess rebuilding against python 3.12 wouldn't help, as Ubuntu Noble 24.04 gets shipped with python 3.12 and has the issue as well. (Though maybe it's still worth to have a closer look, maybe something else is not up to date.)

Some of the issue reports mention a work around: Onboard Settings -> Keyboard -> tab Advanced -> Input event source: Change from "Xinput" to "GTK". For me, tried on Xfce4, this avoids onboard to crash, but causes other strange behavior instead (e.g. keys being stuck).

In the first Ubuntu link, in the first comment, there is a reference to a possible patch (link below). Further down in the Ubuntu issue the comment No. 11 says that the patch didn't solve the problem.

https://github.com/Azkali/onboard-keyman/commit/ac8633670a5467e7b7364651e77be2c94f7e3104.patch

I hope for some patch being applied to Debian or Ubuntu – but there is none so far:

By Jakko on 2024-06-18T20:06:35

I also see that onboard in Alpine was already rebuilt against python 3.12 by this commit:

https://gitlab.alpinelinux.org/alpine/aports/-/commit/71a773b0618ce60d276a3df74ed69e6bd6d13237?page=10#88255e70c3d88f06f4b276ac7e2bf77686bf9e7a

Btw. the Debian patch "debian/patches: Fix ftbfs on python3.12" ...

https://salsa.debian.org/debian-mate-team/onboard/-/commit/e1c683eda8f7e548ad486d7b5fe6bb15ad9eb4cf

... is kind of applied to Alpine as well by "community/onboard: fix build with python 3.12 ". It fixes a build issue but not the issue reported here.

https://gitlab.alpinelinux.org/alpine/aports/-/commit/9c906ae5c50a089cbe0ba6efd58501cc7207a568

By Jakko on 2024-06-18T19:57:08

mentioned in issue #1964

By Jakko on 2024-06-23T23:13:22

There is not much new activity about this in upstream onboard...

upstream bugs (go to page "Last"): https://bugs.launchpad.net/onboard
upstream code (known as unmaintained): https://code.launchpad.net/onboard

... nor in Ubuntu:

Ubuntu bugs: https://bugs.launchpad.net/ubuntu/+source/onboard
Ubuntu packaging (patches from Debian imported, see below): https://git.launchpad.net/ubuntu/+source/onboard

In Debian, some work on patches have been done (from Jun 24, 2024 to Jul 10, 2024) but as far as I can see they are not related to the issue. However, the issue was now reported to the Debian bug tracker, causing investigations on this:

Debian bugs: https://bugs.debian.org/cgi-bin/pkgreport.cgi?pkg=onboard;dist=unstable
- Bug No. 1076120: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1076120
Debian packaging: https://salsa.debian.org/debian-mate-team/onboard/-/commits/master

In Arch Linux the issue was reported, including two possible patches to fix it. However, they were not applied to the packaging yet.

Arch Linux bugs: https://gitlab.archlinux.org/archlinux/packaging/packages/onboard/-/issues
- Issue No. 2: https://gitlab.archlinux.org/archlinux/packaging/packages/onboard/-/issues/2
  - Proposed patch 1: https://github.com/void-linux/void-packages/blob/cee826ffd36d3c9f6c1e326a3b6cfd9696953cbe/srcpkgs/onboard/patches/thread-state.patch
  - Proposed patch 2: https://github.com/void-linux/void-packages/blob/cee826ffd36d3c9f6c1e326a3b6cfd9696953cbe/srcpkgs/onboard/patches/python-3.12.patch
Arch Linux packaging: https://gitlab.archlinux.org/archlinux/packaging/packages/onboard/-/commits/main

For Fedora there is a bug report at Red Hat. There is a proposed patch in comment No. 30. I'm not familiar with the package building at Fedora but it looks to me like "onboard" was removed after Fedora 39 with comment "Orphaned for 6+ weeks", likely due to unmaintained upstream onboard repository.

Fedora / Red Hat bugs: https://bugzilla.redhat.com/buglist.cgi?bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&component=onboard&product=Fedora&query_format=advanced
- Bug No. 2227390: https://bugzilla.redhat.com/show_bug.cgi?id=2227390
  - Proposed patch: https://bugzilla.redhat.com/attachment.cgi?id=2002713&action=diff
Fedora packaging (not sure this is the correct link location for Fedora packaging): https://src.fedoraproject.org/rpms/onboard/commits/main

The proposed patch at Fedora / Red Hat bug No. 2227390 seems to be the same like the one I mentioned further up, which is in the Ubuntu bug No. 2063041 in its first comment:

Ubuntu bug No. 2063041: https://bugs.launchpad.net/ubuntu/+source/onboard/+bug/2063041
- Proposed patch: https://github.com/Azkali/onboard-keyman/commit/ac8633670a5467e7b7364651e77be2c94f7e3104

Whereas the patches proposed in the Arch Linux issue No. 2 mentioned above, the first one goes into a similar direction but the second patch seems to do something else.

As the upstream onboard repository is unmaintained, we'll need to add a patch to Alpine Linux packaging.

There is no issue report on this in Alpine yet.

Unfortunately, I'm short on capacity and time since a while already.

Basically a backtrace of the crash should be done. However, some of the other issue reports already contain one. Therefore we maybe can skip this step.

Testing the different proposed patches mentioned above would be worthwhile. The one mentioned in Ubuntu / Fedora / Red Hat looks preferable. I'd say the two patches proposed in Arch Linux would be second choice. If one of those patches solves the issue, it could be added to Alpine Linux via a merge request. Maybe it would also be worth waiting to see how the issue is going to be solved in Debian, not sure how long it will take. Supporting the Debian issue by e.g. providing hints to the proposed patches might speed it up, unfortunately I'm not familiar with Debian bug reporting system und currently don't have the time to do that.

On the postmarketOS wiki I added a note to the user interfaces Xfce4, MATE and LXQt pointing to this issue report here.

There are two temporary work arounds, which both look not very satisfying to me:

As mentioned further up and in some of the issue reports: Onboard Settings -> Keyboard -> tab Advanced -> Input event source: Change from "Xinput" to "GTK". For me this caused some other issues in behavior (sticking keys). Also I guess this doesn't apply to the lock screen and (if disabling autologin) at LightDM login screen.
Moving to another virtual keyboard until the issue is solved. The easiest possibility is likely "matchbox-keyboard", though it's not as nice as "onboard". After installation, it would need to be added to the autostart applications. For lock screen and LighDM login screen it needs to be used in embedded mode matchbox-keyboard --xid. The wiki pages of Xfce4 and MATE explain in section "Lock screen" how to change the lock screen keyboard. In LXQt it's currently the same like in Xfce4. For LightDM login screen (if autologin is disabled) I'm not fully sure, it should roughly work with greeter "lightdm-gtk-greeter" setting keyboard=matchbox-keyboard --xid in "/etc/lightdm/lightdm-gtk-greeter.conf", see wiki page "Display manager" (edit: without the "keyboard-position" option, this doesn't work with "matchbox-keyboard" as far as I know).

By Jakko on 2024-07-14T15:11:34

I built manually with the first patch from Void Linux above, and it worked! Thanks for investigating this issue. Maybe I will send a patch later.

By xtex on 2024-07-14T16:01:43

Thanks for testing!

Could you also try the patch mentioned in the Ubuntu issue?

https://github.com/Azkali/onboard-keyman/commit/ac8633670a5467e7b7364651e77be2c94f7e3104.patch

I think that patch would be more preferable to submit to Alpine Linux. The one from Void Linux would be a patchset consisting of 2 patches, whereas we seem to need only the first one of it (which makes things more complicated to discusss). Also it slightly changes the structure of an error return at the beginning (with the same result, still a bit intruding). The patch proposed in the Ubuntu issue (and Fedora / Red Hat issue) looks more clear and consistent to me.

Though, as long as it works, I'm also fine with the Void Linux patch.

By Jakko on 2024-07-14T16:41:13

@jakko the mentioned patch also fixes the issue.

By xtex on 2024-07-15T00:04:25

0001-community-onboard-fix-GIL-for-obj-alloc.patch

By xtex on 2024-07-14T23:58:34

Thanks for your effort! The Alpine patch looks good to me.

By Jakko on 2024-07-15T00:07:18

mentioned in issue #3020

By Robert Eckelmann on 2024-07-15T05:20:32

i have same issue on my gtexswifi with xfce4

By 19atlas on 2024-07-15T09:36:00

Many thanks to @xtex! The MR at Alpine is up:

https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/69240

Just today in Debian bug No. 1076120 there is an entry that the bug was solved. They implemented the patch from Void Linux!

Doesn't make a big difference. Both patches do basically the same.

By Jakko on 2024-07-15T21:24:33

I now also had a try with the patch (on LXQt but that shouldn't matter). onboard doesn't crash anymore. However, I have issues with special characters like -, /, etc. When hitting those keys, they get "stuck" and the character doesn't get printed.

For comparison I also tried with the Void Linux patch. It doesn't make a difference, the behavior is the same.

I observed that sticking keys behavior also without patch when trying the workaround changing the "Input event source" from "Xinput" to "GTK" in the onboard settings (see further above). Thus, that sticking keys issue might be a separate issue with another cause. It will need further investigation.

The MR at Alpine should be continued. It prevents onboard from crashing, which improves the situation by a lot already. This will also ease the further investigation on the sticking keys issue.

By Jakko on 2024-07-16T00:56:27

For the key sticking, I got two different error: https://pb.envs.net/?180b13591e9e0f76#41iP1g4hPgJZddxRDVGspsg4VYo4kPg8EkHa6KYjbxxe The first error always occurs only when the first time key sticking appeared, and the next when pressing more keys.

By xtex on 2024-07-16T01:26:51

I'm not familiar with that but as far as I understand in both cases it ends up at:

File "/usr/lib/python3.12/site-packages/Onboard/Keyboard.py", line 189, in press_unicode
    keysym = self._vk.keysym_from_unicode(char)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

In the source code that line is at:

https://bazaar.launchpad.net/~onboard/onboard/1.4/view/head:/Onboard/Keyboard.py#L189

As mentioned, I think this is another (additional) issue. Possibly also related to python3.12 but actually not exactly the "onboard segfaults" issue.

It would be worth to check if is there is already an existing fix for this in other distributions. E.g. patches in Debian:

https://salsa.debian.org/debian-mate-team/onboard/-/tree/master/debian/patches

The current Debian patches 1005-1008 (1009 recently got merged into 1005) have been added with comment "Mostly syntax fixes". Also the second Void Linux patch fixes syntax. I would have assumed that this is just some cosmetic fix. Although, as the word UNICODE shows up in those syntax things, they might be related to the sticking keys issue.

By Jakko on 2024-07-16T07:21:18

mentioned in issue #3014

By Jakko on 2024-07-16T07:26:32

I tried with the second Void Linux patch on top of the Alpine MR.

https://github.com/void-linux/void-packages/blob/cee826ffd36d3c9f6c1e326a3b6cfd9696953cbe/srcpkgs/onboard/patches/python-3.12.patch

Unfortunately, it didn't solve the sticking keys issue.

By Jakko on 2024-07-16T22:13:22

I have tried all of the Debian onboard patches 1.4.1-8. They didn't solve the sticking keys issue for me.

https://salsa.debian.org/debian-mate-team/onboard/-/tree/debian/1.4.1-8/debian/patches

Also I looked through onboard upstream, Ubuntu, Debian, Arch Linux, Fedora issues and packaging, it bothers me that nowhere the sticking keys issue gets mentioned. Also everyone at other distirbutions seem to be fine with the workaround changing the input event source from "Xinput" to "GTK" while this caused the sticking keys issue for me as well. I wonder whether that sticking keys issue is still covered behind the segfault issue in other distributions or if it might be a specific issue to Alpine in some way.

Still I would say the MR at Alpine should go on as it is, because it fixes the segfaulting, which would be a big step forward. The MR seems to wait for an approval by @danct12.

By Jakko on 2024-07-18T19:14:33

@xtex could you have a look at your Alpine MR? It seems blocked! Also I'm not used to pmos, how do you apply the patch directly on pmos?

By Damien Gaignon on 2024-07-20T11:08:24

You can download the prebuilt package file from Alpine's CI. For armv7: https://gitlab.alpinelinux.org/xtex/aports/-/jobs/1453943/artifacts/file/packages/community/armv7/onboard-1.4.1-r12.apk. For aarch64: https://gitlab.alpinelinux.org/xtex/aports/-/jobs/1453942/artifacts/file/packages/community/aarch64/onboard-1.4.1-r12.apk. And install them with "apk add --allow-untrusted onboardxxxxx.apk".

I am not familiar with Alpine's contributing workflow. But I think the MR needs a review, probably from @danct12. -- xtex

By xtex on 2024-07-20T21:23:19

There is a fix for the sticking keys issue by Linux Mint developper "mtwebster". (I haven't tested.)

I websearched for "onboard self._vk.keysym_from_unicode(char)" and found the following issue report for Linux Mint (based on Debian->Ubuntu). Developper "clefebvre" commented yesterday "It's to do with the removal of Py_Unicode in Python 3.12." and added today "Fixed in 1.4.1+mint4.".

https://github.com/linuxmint/mint22-beta/issues/22

For that Linux Mint onboard build "1.4.1+mint4" two patches were added:

https://gitlab.com/mint-pins/onboard/-/commits/c5b2419d457c2deaba8b756654130324aeff7394

The one "LayoutLoaderSVG.py: Fix bad exception string." is similar to the Debian patch "1006_fix-str-not-callable-in-LayoutLoaderSVG-py.patch" in Debian onboard version 1.4.1-8. I don't know what it fixes and if we neccessarily need it.

similar Debian patch: https://salsa.debian.org/debian-mate-team/onboard/-/blob/debian/1.4.1-8/debian/patches/1006_fix-str-not-callable-in-LayoutLoaderSVG-py.patch

However, the patch supposed to fix the sticking keys issue is "osk_virtkey.c: Don't use legacy Py_UNICODE for converting utf8 to keysyms.":

@xtex could you test that patch "osk_virtkey.c: Don't use legacy Py_UNICODE for converting utf8 to keysyms." and, if it indeed solves the sicking key issue, could you add it to the Alpine MR? I'm not sure if we need the other patch for LayoutLoaderSVG.py, I think I wouldn't add it if it's not directly needed for fixing the sticking keys issue (but that's just my assessment, feel free to do it your way).

By Jakko on 2024-07-20T15:27:31

@jakko I tested the Linux Mint fix on samsung-klte, and it does work! I have added the patch to the MR.

By xtex on 2024-07-23T09:49:22

Thanks to @xtex I have tested the last artifact from Alpine MR and everything is ok on an Acer A500 ^^

By Damien Gaignon on 2024-07-23T13:56:42

can also report the latest alpine MR artifacts work on edge with mate on nvidia-tegra-armv7 (tf201, grouper)

By Robert Eckelmann on 2024-07-23T14:16:35

The fix has been merged to Alpine "edge", onboard-1.4.1-r12 contains the patches. The packages for architectures armv7, x86 and riscv64 have not been built yet but that's just a matter of time.

As stable postmarketOS v24.06 is affected too, the fix needs to be backported to Alpine 3.20. @xtex: Would you be up to do this MR as well? It would be again an MR with the same two commits but this time onto the "3.20-stable" branch in Alpine. The MR title needs to be prepended by "[3.20] community/onboard: ..." and as MR decription it's sufficient to refer to the first MR.

By Jakko on 2024-07-24T19:56:08

Thank you @xtex!

https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/69702

By Jakko on 2024-07-25T20:49:59

Is segfault fixed?

By 19atlas on 2024-07-27T13:28:39

the alpine builder for armv7 and x86 have not yet packaged the new packages with the fix

But should come soon

By Robert Eckelmann on 2024-07-27T17:34:55

You can check the available package versions at https://pkgs.alpinelinux.org by looking for package version "onboard", leaving branch at "edge" and not selecting any "Arch":

https://pkgs.alpinelinux.org/packages?name=onboard&branch=edge&repo=&arch=&maintainer=

Currently on "edge" the architectures armhf, s390x, x86_64, aarch64, ppc64le are at package version 1.4.1-r12. The remaining armv7, x86, riscv64 are still at 1.4.1-r11.

The riscv64 packages are built irregular to my knowlegde because they are built on special hardware.

The packages for armv7 and x86 would supposed to be built already but there currently seems to be a problem with the builder. At https://build.alpinelinux.org those two are stuck at an error of building package "community/qt6-qtwebengine" right now.

By Jakko on 2024-07-28T10:39:56

@jakko fyi, I have opened aports#16321 for the issue with the qt6-qtwebengine build

By Robert Eckelmann on 2024-07-28T13:56:00

the build of onboard seems to be blocked by qt6-qtwebengine. therefore it is likely that all x86/armv7/rv64 packages in alpine has been blocked for ~4 days.

By xtex on 2024-07-28T13:45:19

@xtex merge request still appears didn't merged

By 19atlas on 2024-07-28T13:58:24

@19atlas we needs a review from @danct12

By xtex on 2024-07-28T14:06:01

@xtex finaly its merged 🥳🥳🥳🥳🥳🥳🥳

By 19atlas on 2024-07-29T10:16:50

https://pkgs.alpinelinux.org/packages?name=onboard&branch=edge&repo=&arch=armv7&maintainer= finalyyy

By 19atlas on 2024-08-01T15:18:05

closing this as the package has been updated

By Robert Eckelmann on 2024-08-01T21:55:32

closed

onboard segfaults

On what postmarketOS version did you encounter the issue?

On what environment did you encounter the issue?

Environments

Child items 0

Activity