Segfault with netrw on Alpine Linux after v8.2.3584

[ktprograms]

Steps to reproduce

(Steps here using docker to ensure maximum reproducibility):

1. docker run -it --rm alpine:3.15
2. apk add alpine-sdk xxd ncurses-dev lua5.3-dev python3-dev gtk+3.0-dev libx11-dev libxt-dev
3. cd ~
4. git clone --depth=1 https://github.com/vim/vim -b v8.2.3584
5. cd vim/
6. ./configure --prefix=/usr --enable-luainterp --enable-python3interp=dynamic --without-x --disable-nls --enable-multibyte --enable-gui=no --with-lua-prefix=/usr/lua5.3
7. make -j$(nproc)
8. make install
9. cd /usr/share/vim
10. curl -o vimrc https://tpaste.us/ovJr
11. vim /etc/ (Or any directory)

Expected behaviour

No segfault.

Operating system

Alpine Linux v3.15, Docker Image

Version of Vim

Started happening after v8.2.3584, still happens on master (v8.2.3770 at time of reporting). Installed package versions: https://tpaste.us/jNgm

Logs and stack traces

Notes:

• Segfault happens when there are 3 or more user defined commands anywhere in the vim initialisation procedure (even in plugins) any you open netrw. (Commenting out any one of the user defined commands in the example vimrc prevents the segfault from happening).
• Segfault doesn't happen if running vim -u /usr/share/vim/vimrc /etc/, even though that's the default vimrc.
• Segfault doesn't happen if ~/.vimrc exists (can be completely empty)
• According to GDB, the segfault happens in

  vim/src/usercmd.c

  Line 1778 in 58ef8a3

  if ((cmd->uc_argt & EX_KEEPSCRIPT) == 0)

  (line number as of v8.2.3584).
• The segfault doesn't happen on Debian (with compiling the problematic version)

[lacygoill]

What's the output of :scriptnames when you run just $ vim?
What's the output of :scriptnames when you run $ vim /etc?

[ktprograms]

@lacygoill When running vim /etc/, it segfaults so I can't get the output of :scriptnames, right? Or am I misunderstanding something?

[lacygoill]

@lacygoill When running vim /etc/, it segfaults so I can't get the output of :scriptnames, right? Or am I misunderstanding something?

Ah right. My bad. Although, maybe you could start in debug mode:

$ vim -D /etc

And ask for :scriptnames right before the crash. Not sure that's possible.

---

I'm asking these questions because I can't reproduce:

mv ~/.vim/vimrc{,moved} \
    && mkdir -p /tmp/share/vim \
    && VIM=/tmp/share/vim VIMRUNTIME=/usr/local/share/vim/vim82 vim --cmd 'set rtp-=~/.vim rtp-=~/.vim/after' /etc \
    && mv ~/.vim/vimrc{moved,}

I'm trying to find out the right invocation.

[lacygoill]

But at least the output of :scriptnames when starting $ vim alone would help me.

[ktprograms]

@lacygoill Like I noted, if there's a ~/.vimrc file, then the segfault doesn't happen, so I don't know if other things in your invocation are preventing the segfault from happening.

Also, I forgot to mention that the segfault only happens on Alpine.

[lacygoill]

@lacygoill Like I noted, if there's a ~/.vimrc file, then the segfault doesn't happen, so I don't know if other things in your invocation are preventing the segfault from happening.

That's why my command temporarily moves the vimrc so that Vim can't find it:

mv ~/.vim/vimrc{,moved} \
    ...

[lacygoill]

Also, I forgot to mention that the segfault only happens on Alpine.

OK, I'm on Ubuntu 20.04.

[ktprograms]

Output of :scriptnames: https://tpaste.us/DXRd
(When opening a dir it's the exact same list)

[ktprograms]

OK, I'm on Ubuntu 20.04.

@lacygoill can you try with the Docker steps I have in the issue?

[ktprograms]

@lacygoill With vim -D /etc, this is the line just before the segfault:

cmd: call s:VimEnter(expand("<amatch>")

Which is

vim/runtime/plugin/netrwPlugin.vim

Line 36 in dee78e1

au VimEnter * sil call s:VimEnter(expand("<amatch>"))

[lacygoill]

@lacygoill can you try with the Docker steps I have in the issue?

I'm trying but it's not easy because I don't have docker:

$ docker run -it --rm alpine:3.15

Command 'docker' not found, but can be installed with:

sudo snap install docker     # version 19.03.13, or
sudo apt  install docker.io  # version 20.10.7-0ubuntu5~20.04.2

$ sudo apt install docker

$ docker run -it --rm alpine:3.15

Command 'docker' not found, but can be installed with:

...

I had to install docker.io. But I don't understand the remaining commands, and I don't know what they install, nor where they install files, nor how to remove all of it at the end. But I'm still in the process of running all the commands. I'll need more time.

---

@lacygoill With vim -D /etc, this is the line just before the segfault:

Thank you very much. FWIW, you can step inside the s:VimEnter() function by executing the :help >step debugging command.

[ktprograms]

@lacygoill Once you install docker, all the commands after the docker run should be done inside the docker container. Once you're done testing, you can run docker system prune -a which will remove the docker stuff.

apk is the Alpine package keeper, does the same job as apt

[lacygoill]

@lacygoill Once you install docker, all the commands after the docker run should be done inside the docker container. Once you're done testing, you can run docker system prune -a which will remove the docker stuff.

Thank you, that's helpful.

---

OK, I can reproduce the crash now. I'm going to try to get some logs.

[chrisbra]

can you paste the backtrace in gdb, since you already pinpointed the problematic line in gdb?

[ktprograms]

@chrisbra I'm deep in vim debug mode in the middle of netrw right now, but as soon as I'm done I'll compile vim with debug symbols again and send the backtrace.