Re: [PATCH v3 4/6] git.c: implement --list-cmds=porcelain

["SZEDER Gábor" <szeder.dev@gmail.com>]

On Sat, Apr 21, 2018 at 6:54 PM, Nguyễn Thái Ngọc Duy <pclouds@gmail.com> wrote:
> This is useful for git-completion.bash because it needs this set of
> commands. Right now we have to maintain a separate command category in
> there.

I don't really understand this paragraph, in particular its second
sentence.  I would have described this change like this:

  To get the list of commands offered after 'git <TAB>', the
  completion script filters out plumbing commands from the list of all
  git commands.  To do so, it contains a long hard-coded list of the
  names of all known plumbing commands, which is redundant with the
  categorization in 'command-list.txt', is a maintenance burden, and
  tends to get out-of-sync when new plumbing commands are added.

  Implement 'git --list-cmds=porcelain' to list only commands
  categorized as porcelains, so we can get rid of that long hard-coded
  command list from the completion script.

But then I noticed that it's not an accurate description of the
current situation, because there is a wide grey area between
porcelains and plumbing, and the completion script doesn't "filter out
plumbing commands", but rather filters out commands that can be
considered too low-level to be useful for "general" usage.
Consequently, after 'git <TAB>' we also list:

  - some 'ancillaryinterrogators': blame, annotate, difftool, fsck,
    help
  - some 'ancillarymanipulators': config, mergetool, remote
  - some 'foreignscminterface': p4, request-pull, svn, send-email
  - even some plumbing: apply, name-rev (though 'name-rev' could be
    omitted; we have 'git describe')
  - and also all "unknown" 'git-foo' commands that can be found in
    $PATH, which can be the user's own git scripts or other
    git-related tools ('git-annex', Git LFS, etc.).

With this change we wouldn't list any of the above commands, but only
those that are explicitly categorized as 'mainporcelain'.  I'd much
prefer the current behaviour.


> Note that the current completion script incorrectly classifies
> filter-branch as porcelain and t9902 tests this behavior. We keep it
> this way in t9902 because this test does not really care which
> particular command is porcelain or plubmbing, they're just names.
>
> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
> ---
>  contrib/completion/git-completion.bash | 106 +++----------------------
>  git.c                                  |   2 +
>  help.c                                 |  12 +++
>  help.h                                 |   1 +
>  t/t9902-completion.sh                  |   6 +-
>  5 files changed, 31 insertions(+), 96 deletions(-)
>
> diff --git a/contrib/completion/git-completion.bash b/contrib/completion/git-completion.bash
> index a5f13ade20..7d17ca23f6 100644
> --- a/contrib/completion/git-completion.bash
> +++ b/contrib/completion/git-completion.bash


> @@ -859,101 +864,14 @@ __git_all_commands=
>  __git_compute_all_commands ()
>  {
>         test -n "$__git_all_commands" ||
> -       __git_all_commands=$(__git_list_all_commands)
> -}
> -
> -__git_list_porcelain_commands ()
> -{
> -       local i IFS=" "$'\n'
> -       __git_compute_all_commands
> -       for i in $__git_all_commands
> -       do
> -               case $i in
> -               *--*)             : helper pattern;;
> -               applymbox)        : ask gittus;;
<...>
> -               verify-tag)       : plumbing;;

I will be glad to see this enormous list go.

> -               *) echo $i;;
> -               esac
> -       done
> +       __git_all_commands=$(__git_list_commands all)
>  }