Re: [PATCH] diff: Introduce diff.algorithm variable

From: Junio C Hamano <gitster@pobox.com>
Date: Mar 9, 2012, 18:40 UTC
Message-ID: <7vmx7p62k1.fsf@alter.siamese.dyndns.org>
In-Reply-To: <06652d553040ad6b25608dc69d632f1ee38eaeca.1331300343.git.mprivozn@redhat.com>

Michal Privoznik <mprivozn@redhat.com> writes:

> Some users have preference over various diff algorithms. However,
> now they are forced to use appropriate argument every time they
> run git-diff and tools using it. This is impractical. Therefore
> create new variable which can set preferred algorithm. Of course,

We tend to prefer avoiding these subjective and judgemental words like "forced" and "impractical" when the paragraph conveys the same reasoning even when they are toned down.

Show 7 quoted lines

> diff --git a/Documentation/diff-config.txt b/Documentation/diff-config.txt
> index 6aa1be0..1047e81 100644
> --- a/Documentation/diff-config.txt
> +++ b/Documentation/diff-config.txt
> @@ -1,3 +1,24 @@
> +diff.algorithm::
> +    Choose a diff algorithm.  The variants are as follows:

In order to avoid confusing users, we may want to mention that it is deliberate that this configuration does not apply to plumbing commands.

It may make sense to drop the enumeration of "The variants are...", and say that it can take the values valid for the "--diff-algorithm" parameter to the "diff" family of commands. We do not have to worry about keeping two enumerations in sync if we did so.

Show 11 quoted lines

> diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
> index 7d4566f..4e8bc5b 100644
> --- a/Documentation/diff-options.txt
> +++ b/Documentation/diff-options.txt
> @@ -55,6 +55,29 @@ endif::git-format-patch[]
>  --histogram::
>  	Generate a diff using the "histogram diff" algorithm.
>  
> +--diff-algorithm={histogram|myers|minimal|patience}::
> +    Choose a diff algorithm. The defaults are controlled
> +    by the `diff.algorithm` configuration variable

s/$/ for the Porcelain commands/; perhaps?

Show 10 quoted lines

> diff --git a/contrib/completion/git-completion.bash b/contrib/completion/git-completion.bash
> index fba076d..d54f3a3 100755
> --- a/contrib/completion/git-completion.bash
> +++ b/contrib/completion/git-completion.bash
> @@ -1299,6 +1299,7 @@ __git_diff_common_options="--stat --numstat --shortstat --summary
>  			--raw
>  			--dirstat --dirstat= --dirstat-by-file
>  			--dirstat-by-file= --cumulative
> +			--diff-algorithm=
>  "

So this helps me when I do "git diff --diff-al<TAB>" and expands it to "git diff --diff-algorithm=" and then...

Show 9 quoted lines

>  _git_diff ()
> @@ -1306,6 +1307,10 @@ _git_diff ()
>  	__git_has_doubledash && return
>  
>  	case "$cur" in
> +	--diff-algorithm=*)
> +		__gitcomp "myers histogram minimal patience"
> +		return
> +		;;

... when typing <TAB> again, this helps me by reminding what choices are available. Is that the idea? Nice.

But I see you repeated this four-element list three times. Can't we do it in just one place?

Show 9 quoted lines

> diff --git a/diff.c b/diff.c
> index 377ec1e..f5c965b 100644
> --- a/diff.c
> +++ b/diff.c
> @@ -34,6 +34,7 @@ static int diff_no_prefix;
>  static int diff_stat_graph_width;
>  static int diff_dirstat_permille_default = 30;
>  static struct diff_options default_diff_options;
> +static int diff_algorithm = 0;

Please let BSS take care of this initialization by omitting " = 0".

Show 7 quoted lines

> @@ -169,6 +184,12 @@ int git_diff_ui_config(const char *var, const char *value, void *cb)
>  	if (!strcmp(var, "diff.ignoresubmodules"))
>  		handle_ignore_submodules_arg(&default_diff_options, value);
>  
> +	if (!strcmp(var, "diff.algorithm")) {
> +		if ((diff_algorithm = diff_algorithm_parse(value)) < 0)
> +			die("bad diff.algorithm value: %s", value);

I am not sure if this is a good idea. What happens when Git 2.0 adds a new algorithm, the user names it in ~/.gitconfig, and an older version of Git sees this value? Do you force the user to modify ~/.gitconfig to use one of the older ones?

It might be more friendly to just warn and use the default.

Show 25 quoted lines

> @@ -3250,6 +3271,9 @@ int diff_setup_done(struct diff_options *options)
>  		DIFF_OPT_SET(options, EXIT_WITH_STATUS);
>  	}
>  
> +	/* set diff algorithm */
> +	options->xdl_opts |= diff_algorithm;
> +
>  	return 0;
>  }
>  
> @@ -3528,6 +3552,17 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
>  		DIFF_XDL_SET(options, PATIENCE_DIFF);
>  	else if (!strcmp(arg, "--histogram"))
>  		DIFF_XDL_SET(options, HISTOGRAM_DIFF);
> +	else if ((argcount = parse_long_opt("diff-algorithm", av, &optarg))) {
> +		int alg = diff_algorithm_parse(optarg);
> +		if (alg < 0)
> +			die("unknown algorithm: %s", optarg);
> +		DIFF_XDL_CLR(options, NEED_MINIMAL);
> +		DIFF_XDL_CLR(options, HISTOGRAM_DIFF);
> +		DIFF_XDL_CLR(options, PATIENCE_DIFF);
> +		options->xdl_opts |= alg;
> +		return argcount;
> +	}
> +

The last two hunks gives me a queasy feeling for some reason. I think I recently saw a patch to clean up this area to clarify that PATIENCE/HISTOGRAM are not independent option bits (but I cannot remember where) and we probably should use the approach taken by that patch.

Other than that, both the design and the implementation seems to be done reasonably competently. Don't we want to add tests for this?