[PATCH 2 of 2 V2] extdiff: add --mode option

[Yuya Nishihara]

On Sun, 13 Jan 2019 15:31:07 -0800, Ludovic Chabant wrote:
> > A separate patch seems fine. To make clear, there's no need to check if the
> > tool is console-based or not unless multiple diff processes are spawned
> > simultaneously. Currently the .gui flag is tested only by filemerge.py.
> 
> So you mean I don't need to modify the previous code path (dir-diff), nor do I need to check the gui flag if the "confirm" flag is given (i.e. one external process at a time)?
> How come? Wouldn't even one gui process be a problem if you're in a shell-only instance of hg?

I was just speaking about the current implementation (without your patch.)
Maybe I should say .gui flag isn't checked because the current extdiff never
spawns more than one processes at a time.

> > The latter looks worse for me. Instead, you can use [alias] to pass in
> > arguments to hg commands.
> 
> That's true, but then the entire extdiff configuration section would be deprecated if that was the case, no?
> Like, AFAIK there's not much difference between:
> 
> [extdiff]
> cmd.bcomp = /path/to/bcomp
> opts.bcomp = --whatever
> 
> and:
> 
> [alias]
> bcomp = extdiff -p /path/to/bcomp -o "--whatever"
> 
> I assume the point of extdiff is to be a slighly better version of an alias for the purpose of a diffing stuff, but maybe someone with a better knowledge of the history of both features can correct me.

AFAIK, it duplicates alias functionality because there wasn't no [alias]
when the extdiff extension was introduced. And I think that's the major
reason why the extdiff is still an extension. It can't be trivially ported
to a core command.

If the extdiff had the option to look for a diff tool from stock templates
(i.e. [diff-tools], [merge-tools], and maybe [extdiff]), alias could be
expressed as follows:

  [alias]
  bcomp = extdiff --tool bcomp --per-file

This should be good enough.

> More importantly, I actually just realized (maybe) why a --mode option might be better. Remember how I intended to have, say, BeyondCompare setup to do per-file-diffs by default, and I would pass a dir-diff option for the 5% cases where I want to only diff a few files in a revision?
> Well, if we have multiple flags, the extdiff/alias/whatever section of my .hgrc would specify --per-file (so I get per-file diffs by default), but if I then pass --dir (to revert back to a dir-diff), it wouldn't override the default... the extdiff command would instead get both --per-file and --dir, and would most likely throw an error because it doesn't make sense to pass those options togethers.

You can turn off the --per-file by --no-per-file. A separate --dir option
isn't needed as long as --no-per-file == --dir. See hg help flags.