Re: [PATCH v3] Enable auto-merge for meld to follow the vim-diff beharior

From: David Aguilar <davvid@gmail.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: sunlin via GitGitGadget <gitgitgadget@gmail.com>,
	git@vger.kernel.org, sunlin <sunlin7@yahoo.com>,
	"lin.sun" <lin.sun@zoom.us>
Subject: Re: [PATCH v3] Enable auto-merge for meld to follow the vim-diff beharior
Date: Tue, 30 Jun 2020 00:42:04 -0700	[thread overview]
Message-ID: <20200630074204.GA2144485@gmail.com> (raw)
In-Reply-To: <xmqqeepxfmdm.fsf@gitster.c.googlers.com>

On Mon, Jun 29, 2020 at 05:06:13PM -0700, Junio C Hamano wrote:
> "sunlin via GitGitGadget" <gitgitgadget@gmail.com> writes:
> >  mergetools/meld | 32 ++++++++++++++++++++++++++++++--
> >  1 file changed, 30 insertions(+), 2 deletions(-)
> >
> > diff --git a/mergetools/meld b/mergetools/meld
> > index 7a08470f88..91b65ff22c 100644
> > --- a/mergetools/meld
> > +++ b/mergetools/meld
> > @@ -7,13 +7,23 @@ merge_cmd () {
> >  	then
> >  		check_meld_for_output_version
> >  	fi
> > +	if test -z "${meld_has_auto_merge_option:+set}"
> > +	then
> > +		check_meld_for_auto_merge_version
> > +	fi
> 
> The detection part looks clumsy and inefficient.  More about it later.

Sorry for not noticing your reply here earlier.  I agree with everything
you wrote here, and rescind my earlier sign-off.  Combining as you
suggested below is best IMO as well.

> 
> > +	option_auto_merge=
> > +	if test "$meld_has_auto_merge_option" = true
> > +	then
> > +		option_auto_merge="--auto-merge"
> > +	fi
> >  
> >  	if test "$meld_has_output_option" = true
> >  	then
> > -		"$merge_tool_path" --output="$MERGED" \
> > +		"$merge_tool_path" $option_auto_merge --output="$MERGED" \
> >  			"$LOCAL" "$BASE" "$REMOTE"
> >  	else
> > -		"$merge_tool_path" "$LOCAL" "$MERGED" "$REMOTE"
> > +		"$merge_tool_path" $option_auto_merge "$LOCAL" "$MERGED" "$REMOTE"
> >  	fi
> >  }
> 
> The part that chooses whether to pass --auto-merge or not and
> adjusts the command line options does look sensible.
> 
> I wonder if the same "hasAutoMerge" option can be used by those who
> do *not* want the new --auto-merge behaviour to opt out of this
> change.  Is there a reason why "meld" offers the --auto-merge as an
> optional behaviour (which tells, at least to me, that the default
> behaviour is not to auto-merge and makes me assume that the default
> must be chosen by some sound reasoning, hence some users would prefer
> not to use this new behaviour with good reasons)?
> 
> I guess what I am trying to get at is, if --auto-merge is an optional
> and non-default behaviour for "meld" users, perhaps it is not a good
> idea to change the behaviour on them only because the version of meld
> they run happens to support the --auto-merge as an optional behaviour.
> 
> IOW, wouldn't it make more sense, and certainly make it safer
> without surprises to existing users, if we made the logic to
> 
>     * If mergetool.meld.useAutoMerge is not set, do not pass
>       --auto-merge whether "meld" supports the option or not.
> 
>     * If mergetool.meld.useAutoMerge is 'true', always pass it
>       without checking.
> 
>     * If mergetool.meld.useAutoMerge is 'when-able' (or come up with
>       a better name if you want, perhaps 'auto'), check if "meld"
>       accepts "--auto-merge" and decide whether to pass it or not.
> 
> perhaps?

I like the idea of having it be auto/true/false, and perhaps "auto"
would be a sensible default if more users benefit from it than not.

Sunlin, do you have an opinion on what the default should be?

> > +# Check whether we should use 'meld --auto-merge ...'
> > +check_meld_for_auto_merge_version () {
> > +	meld_path="$(git config mergetool.meld.path)"

Small sug -- this command substitution doesn't need the enclosing
quotes.

	meld_path=$(git config mergetool.meld.path || echo meld)

should be sufficient.  Are we okay with `|| echo meld`?
If so, it would let us drop this line below.

> > +	meld_path="${meld_path:-meld}"

> > +
> > +	if meld_has_auto_merge_option=$(git config --bool mergetool.meld.hasAutoMerge)
> > +	then
> > +		: use configured value
> > +	elif "$meld_path" --help 2>&1 |
> > +		grep -e '--auto-merge' -e '\[OPTION\.\.\.\]' >/dev/null
> > +	then
> > +		: old ones mention --auto-merge and new ones just say OPTION...
> > +		meld_has_auto_merge_option=true
> > +	else
> > +		meld_has_auto_merge_option=false
> > +	fi
> > +}
> 
> When not configured, we end up running "meld --help" twice for two
> options, which is not great, don't you think?  I actually think the
> part that runs "meld --help" and parses its output should be split
> out of the helper function "check_meld_for_output_version" and
> called "check_meld_supported_options" or something, so that the
> logic to see if the --output and --auto-merge options should be
> passed can be made with at most one invocation of "meld --help".
> Which may involve *NOT* having two separate helper functions
> check_meld_for_*_version for the tested features.

I'm 100% on board with this suggestion.

> 
> Oh, also, check_meld_for_*_version is nonsensical as a name for
> these helper functions (it is not the fault of this patch---it is
> mimicking the existing practice, but that does not make the function
> name not nonsensical).  The helpers do not actually want to check a
> "version", they only want to see if a feature is supported.  So
> having "feature" in the name would be good, but "version" is not.

Ditto.
-- 
David