[LU-16392] Bash completion regressions, lctl and lfs

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Affects Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Thomas Bertschinger
• Reporter: Thomas Bertschinger

Description

There are a few regressions introduced in the bash completions for lctl and lfs.

1. Attempting completion for lctl prints a message to stderr and does not generate any completions. To test this behavior interactively, type lctl <tab>. To test this behavior programmatically, run

source lustre/scripts/bash-completion/lustre; COMPLETIONS=$( _lustre_cmds lctl) && [ ! -z "$COMPLETIONS" ]

An exit status of 1 indicates the bug is present.

 

2. Completion for lfs check gives incorrect results.

 

3. Attempting completion for lfs find -component-flags erroneously prints a message to stderr.

Issue Links

• is related to: LU-9830 string truncation from snprintf in libcfs/util/parser.c::Parser_list_commands()
• is related to: LU-5170 lfs usability
• is related to: LU-8621 Parser_execarg() prints help to stderr, usage to stdout
• is related to: LU-13225 add bash completion for lfs
• is related to: LU-12734 bash completion (for now: lctl set/get_param)
• is related to: LU-930 Update or improvement to a Lustre man page, documentation, or messages

Comments

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/49484/
Subject: LU-16392 utils: use --list-commands for bash completion
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: b4cc570ad11c1c07a6e1d825787ccc62c1245ca1

[Peter Jones]

Does not affect b2_15- just master 

[Gerrit Updater]

"Thomas Bertschinger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/49484
Subject: LU-16392 utils: use --list-commands for bash completion
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 7c02a52fd57e2a0fa7e35c6cf3dad16ba12cb8d1

[Andreas Dilger]

For the sub-commands, is it possible to list only the top-level command, and then add eg. "lctl pcc --list-commands" to generate the list of sub commands for the completion (ie. it passes the sub-command as well, not just "lctl --list-commands").?

Also, in a semi-related note, I'm not sure why "--list-commands" was used here instead of using "--help" to get this list, but it would be good to allow both (though "--list-commands" should be used for the completions since it has existed much longer).

[Andreas Dilger]

I was looking at the command truncation issue as well. It looks like we are out by only a character or two in a few rare cases. It would be better to misalign the output by a few characters to fit in the longest options I think, rather than creating a new option.

[Thomas Bertschinger]

I considered resolving this by using lctl --list-commands, but there are two problems:

1. --list-commands truncates long options, thus generating incorrect completions (example: changelog_deregister)
2. lctl --list-commands lists options for sub-commands which do not apply to the top-level command, and I don't know how to work around this within the bash completion interface (example: lctl pcc add: bash would suggest both pcc and add as completions)

I have a patch in progress that adds an option --list-commands-brief which prints out subcommands without the above two problems. This has the advantage of not changing existing behavior of the CLI tools, but the disadvantage is adding a new, perhaps superfluous option. What do you think of this idea? An alternative would be modifying the format of lctl --list-commands.

Note, in this bug report I only mentioned lctl but this problem applies to lfs as well.

[Andreas Dilger]

To test this behavior programmatically, run:

source lustre/scripts/bash-completion/lustre; COMPLETIONS=$( _lustre_cmds lctl) && [ ! -z "$COMPLETIONS" ]

Thanks for this. I was wondering how we might be able to write a regression test for this without having interactive usage.

[Andreas Dilger]

Thomas, thanks for your bug report. Any chance you could investigate how to change bash completion to call "lctl --list-commands" to generate to command list instead of just a null input?

Otherwise, it should be possible to revert the above patch to restore the basic functionality, with the accompanying spew of lines for any error (which is why it was changed originally).

[Andreas Dilger]

It looks like this is a regression from patch https://review.whamcloud.com/47162 "LU-8621 utils: cmd help to stdout or short cmd error"