Posted too early by accident, the description is now complete.

I haven't put a lot of thought into this yet, but my first reaction is I am more interested in making it consistent by calling the hook in both cases. Do you want to have a go at that as well? I am imagining a few lines to add the hook before the direct call to the subcommand in _dispatchHelpCommand, not a refactor. Would be a separate competing PR.

I think some people might try using the preSubcommand hook to do some lazy setup, and that might affect the help too.

I haven't put a lot of thought into this yet, but my first reaction is I am more interested in making it consistent by calling the hook in both cases.

I don't really like this idea for two reasons:

• There is no Command instance that would be reasonable to pass to the preSubcommand hooks in the second parameter since the help command only exists virtually.

• There would be a discrepancy between this and how the preAction hooks are not called when the help option is invoked (and that is a valid comparison because the help option would have to be implemented in the action handler just like the help command would have to be implemented in a subcommand if the convenience shortcuts were not offered by the library).

  try {
    const program = new Command()
      .exitOverride()
      .hook('preAction', () => console.log('preAction'))
      .action(() => {});
    program.parse(['--help'], { from: 'user' }); // 'preAction' not logged
  } catch {}
  
  try {
    const program = new Command()
      .exitOverride()
      .addHelpCommand('help')
      .hook('preSubcommand', () => console.log('preSubcommand'));
    program.parse(['help'], { from: 'user' }); // 'preSubcommand' would be logged
  } catch {}

I was thinking about this again today, and leaning towards triggering the hook again!

#1956 (comment)

Interesting comparison with a "real" help command. I had not considered that model.

I think of the built-in prog help foo as equivalent to prog foo --help. I'm leaning towards calling preSubcommand, as it did before #1864.

(This is pretty obscure behaviour, no one may ever notice!)