Long lag when completing the command argument of ssh

[szsam]

Describe the bug

There is a long lag (30 seconds) when completing the command argument of ssh.

To reproduce

Type 'ssh user@host ', then hit tab. The shell will freeze for 30 seconds.

Expected behavior

The completion should be fast. The 30-second lag is unacceptable.

Versions (please complete the following information)

• Operating system name/distribution and version: Ubuntu 20.04.5, WSL2
• bash version, echo "$BASH_VERSION": 5.0.17(1)-release
• bash-completion version, (IFS=.; echo "${BASH_COMPLETION_VERSINFO[*]}"): 2.10

Additional context

It is the following line that takes 30 seconds, where cur is "". The output of the compgen -c -- '' command contains 10k+ lines on my system.

bash-completion/completions/ssh

Line 356 in ac68847

COMPREPLY+=($(compgen -c -- "$cur"))

Debug trace

bash_completion_log.txt

[szsam]

I suggest that we do not try to complete when cur is an empty string.

[szsam]

I'm using Ubuntu on WSL (Windows Subsystem for Linux). That might be the reason why the compgen is so slow.

[akinomyoga]

Similar codes can be found in more completions within the codebase.

./completions/_svn:42:                COMPREPLY=($(compgen -c -- "$cur"))
./completions/cpio:27:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/dpkg:107:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/fio:78:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/fio:91:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/firefox:36:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/man:33:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/mr:57:                COMPREPLY=($(compgen -c -- "${cur}"))
./completions/mussh:42:            COMPREPLY+=($(compgen -c -- "$cur"))
./completions/mypy:28:            COMPREPLY=($(compgen -c -- "${cur:-py}"))
./completions/perlcritic:37:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/protoc:23:                COMPREPLY=($(compgen -c -- "$cur"))
./completions/querybts:22:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/reportbug:42:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/rpm:83:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/sqlite3:21:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/ssh:189:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/ssh:356:            COMPREPLY+=($(compgen -c -- "$cur"))
./completions/ssh:405:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/ssh:551:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/ssh-keygen:82:                        COMPREPLY=($(compgen -c -- "${cur#*=}"))
./completions/strace:93:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/tar:551:                COMPREPLY=($(compgen -c -- "$cur"))
./completions/tcpdump:31:            COMPREPLY=($(compgen -c -- "$cur"))
./completions/valgrind:86:                    COMPREPLY=($(compgen -c -- "$cur"))
./completions/vpnc:57:            COMPREPLY=($(compgen -c -- "$cur"))

I think it would be more consistent to switch the behavior referencing shopt -q no_empty_cmd_completion.

[scop]

I think it would be more consistent to switch the behavior referencing shopt -q no_empty_cmd_completion.

SGTM. We may want to wrap that in a public utility function to make it harder to forget, easier to test, and make it possible for users to override (even if that's not something we'd generally recommend).