Re: [PATCH v3] difftool.c: learn a new way start from specified file

"ZheNing Hu via GitGitGadget" [off-list ref] writes:

There is nothing that specifies a file in the above ;-)

	--start-from=<file>::
		Start viewing ...

This is even more important as SYNOPSIS section of this manual page
does not duplicate the list of options and their arguments.

 -t <tool>::
 --tool=<tool>::
 	Use the diff tool specified by <tool>.  Valid values include

There are many things I dislike about this patch, but do not take it
as a personal attack.  I'll try to suggest an alternative at the end,
but read along.

This may be a good UI to "difftool".

Unacceptable.  Nothing gives Git the right to squat on such a
generic name, and there is no hint that it is used to specify the
starting point of what operation.  In addition, I do not see a good
reason why we need to use an environment variable in the first
place.  We run "diff" as an external process, with GIT_EXTERNAL_DIFF
environment pointing back at us, no?  This information should be
passed from its command line.

Again, an unacceptable "hack".  "start the diff output showing from
this path" would plausibly a good feature even outside the scope of
"difftool", and the feature should not be limited to the external
diff interface.  More on this later.

Do we need these changes for the new feature to work?

I also find this "skip and discard ones earlier than the given path"
makes the utility of the feature artificially narrower than needed,
when I imagine how else this feature, or a variant of it, would be
useful in other situations.  For example, consider that there are 5
paths, and you've seen 3 of them so far before you went off, so you
are restarting from 4th file.  But wouldn't it be more useful, after
showing the 4th and 5th file, if the tool wraps around to show 1st,
2nd, and 3rd file if the user kept going?

I suspect that this feature fits the overall system much better if
it is implemented as a new step in the diffcore transformation.  The
way our diff subsystem works is roughly:

 * The front-end "diff", "diff-files", "diff-index" and "diff-trees"
   are given two "tree like things" to compare, and feeds bunch of
   <old, new> tuples to the diff internal machinery.  Each of these
   tuples are called "filepairs", and a "pair" has two "filespecs",
   one describing the contents, the mode, and the path in the "old"
   side of the "tree like things", the other describing the
   contents, the mode, and the path in the "new" side of the "tree
   like things".

 * The series of filepairs are given to the "diffcore" machinery,
   where they may be broken (i.e. a filepair that says that the old
   contents of "hello.txt" was X, and the new contents of
   "hello.txt" is Y, may become two filepairs, one that says that
   the "hello.txt" file with contents X used to be in the old tree
   but there is nothing corresponding to it in the new tree, and the
   other says that a new "hello.txt" with contents Y appeared
   without corresponding thing in the old tree), matched (i.e. there
   may originally be two filepairs, one that says path A.txt appears
   on the old side but disappeared on the new side, and the other
   that says path sub/A.txt did not exist on the old side but
   appears on the new side---these two filepairs may be merged to
   express "A.txt on the old side got renamed to sub/A.txt on the
   new side"), etc.

 * The set of filepairs processed in the "diffcore" machinery is
   given to the backend and each filepair will be shown in the
   output, as a series of patches, a diffstat, etc.

There is a step in the "diffcore" machinery called "diffcore-order".
The front-ends all feed the filepairs alphabetically to the
"diffcore" machinery, but "diffcore-order" can reorder them, so that
the original order that the "git diff HEAD --" frontend found the
changes may be to "hello.c" and then "hello.h" (because .c sorts
before .h), but the users can specify with the "-O<orderfile>"
option that they want to view the header files before the source
files.  When the set of filepairs exits the diffcore machinery, the
original "hello.c" then "hello.h" order may get modified to show
"hello.h" then "hello.c".  It probably is the simplest to model this
new feature after how "diffcore-order" does it.

So, perhaps we can introduce a new "diffcore-rotate" step, where the
filepairs before the specified location are rotated out to the end
of the filepairs?

The following is just a quick draft that is only lightly tested by
running itself with the "--rotate-to=diffcore-rotate.c" option, but
it should be sufficient to get you started.  You should add a way to
diagnose that the name given to the "--start-file" option actually
exists in the diff on the "difftool" side, because a path that does
not exist in the patch with the following code is simply ignored
(and that is very much deliberate, because we do not want it to die
while running "git log -p --rotate-to=X" where some changes may or
may not touch X).


$ ./git diff --stat -p --rotate-to=diffcore-rotate.c HEAD

 diffcore-rotate.c | 35 +++++++++++++++++++++++++++++++++++
 diffcore.h        |  1 +
 Makefile          |  1 +
 diff.c            |  4 ++++
 diff.h            |  1 +
 5 files changed, 42 insertions(+)

diff --git c/diffcore-rotate.c w/diffcore-rotate.c
new file mode 100644
index 0000000000..0d17901616
--- /dev/null
+++ w/diffcore-rotate.c

@@ -0,0 +1,35 @@
+/*
+ * Copyright (C) 2021, Google LLC.
+ * Based on diffcore-order.c, which is Copyright (C) 2005, Junio C Hamano
+ */
+#include "cache.h"
+#include "diff.h"
+#include "diffcore.h"
+
+void diffcore_rotate(const char *rotate_to_filename)
+{
+	struct diff_queue_struct *q = &diff_queued_diff;
+	struct diff_queue_struct outq;
+	int rotate_to, i;
+
+	if (!q->nr)
+		return;
+
+	for (i = 0; i < q->nr; i++)
+		if (!strcmp(rotate_to_filename, q->queue[i]->two->path))
+			break;
+	/* we did not find the specified path */
+	if (q->nr <= i)
+		return;
+
+	DIFF_QUEUE_CLEAR(&outq);
+	rotate_to = i;
+
+	for (i = rotate_to; i < q->nr; i++)
+		diff_q(&outq, q->queue[i]);
+	for (i = 0; i < rotate_to; i++)
+		diff_q(&outq, q->queue[i]);
+
+	free(q->queue);
+	*q = outq;
+}

diff --git c/diffcore.h w/diffcore.h
index d2a63c5c71..bd5959375b 100644
--- c/diffcore.h
+++ w/diffcore.h

@@ -164,6 +164,7 @@ void diffcore_rename(struct diff_options *);
 void diffcore_merge_broken(void);
 void diffcore_pickaxe(struct diff_options *);
 void diffcore_order(const char *orderfile);
+void diffcore_rotate(const char *rotate_to_filename);
 
 /* low-level interface to diffcore_order */
 struct obj_order {

diff --git c/Makefile w/Makefile
index b797033c58..031b0b88e6 100644
--- c/Makefile
+++ w/Makefile

@@ -869,6 +869,7 @@ LIB_OBJS += diffcore-delta.o
 LIB_OBJS += diffcore-order.o
 LIB_OBJS += diffcore-pickaxe.o
 LIB_OBJS += diffcore-rename.o
+LIB_OBJS += diffcore-rotate.o
 LIB_OBJS += dir-iterator.o
 LIB_OBJS += dir.o
 LIB_OBJS += editor.o

diff --git c/diff.c w/diff.c
index 69e3bc00ed..90a8d5abd0 100644
--- c/diff.c
+++ w/diff.c

@@ -5599,6 +5599,8 @@ static void prep_parse_options(struct diff_options *options)
 			  DIFF_PICKAXE_REGEX, PARSE_OPT_NONEG),
 		OPT_FILENAME('O', NULL, &options->orderfile,
 			     N_("control the order in which files appear in the output")),
+		OPT_STRING(0, "rotate-to", &options->rotate_to, N_("<path>"),
+			   N_("show the change in the specified path first")),
 		OPT_CALLBACK_F(0, "find-object", options, N_("<object-id>"),
 			       N_("look for differences that change the number of occurrences of the specified object"),
 			       PARSE_OPT_NONEG, diff_opt_find_object),

@@ -6669,6 +6671,8 @@ void diffcore_std(struct diff_options *options)
 		diffcore_pickaxe(options);
 	if (options->orderfile)
 		diffcore_order(options->orderfile);
+	if (options->rotate_to)
+		diffcore_rotate(options->rotate_to);
 	if (!options->found_follow)
 		/* See try_to_follow_renames() in tree-diff.c */
 		diff_resolve_rename_copy();

diff --git c/diff.h w/diff.h
index 2ff2b1c7f2..0801469e63 100644
--- c/diff.h
+++ w/diff.h

@@ -226,6 +226,7 @@ enum diff_submodule_format {
  */
 struct diff_options {
 	const char *orderfile;
+	const char *rotate_to;
 
 	/**
 	 * A constant string (can and typically does contain newlines to look for

Junio C Hamano [off-list ref] 于2021年2月10日周三 上午2:17写道：

"ZheNing Hu via GitGitGadget" [off-list ref] writes:

quoted

diff --git a/Documentation/git-difftool.txt b/Documentation/git-difftool.txt
index 484c485fd06c..552be097dfea 100644
--- a/Documentation/git-difftool.txt
+++ b/Documentation/git-difftool.txt

@@ -34,6 +34,9 @@ OPTIONS
      This is the default behaviour; the option is provided to
      override any configuration settings.

+--start-from::
+     Start viewing diff from the specified file.
+

There is nothing that specifies a file in the above ;-)

        --start-from=<file>::
                Start viewing ...

This is even more important as SYNOPSIS section of this manual page
does not duplicate the list of options and their arguments.

I admit my mistake,thanks for reminding.

quoted

 -t <tool>::
 --tool=<tool>::
      Use the diff tool specified by <tool>.  Valid values include

There are many things I dislike about this patch, but do not take it
as a personal attack.  I'll try to suggest an alternative at the end,
but read along.

Of course I humbly accepted your criticism,I am only a sophomore in
university after all,It is extremely important to listen to your suggestions.

quoted

diff --git a/builtin/difftool.c b/builtin/difftool.c
index 6e18e623fddf..67d2befa1210 100644
--- a/builtin/difftool.c
+++ b/builtin/difftool.c

@@ -690,7 +690,7 @@ int cmd_difftool(int argc, const char **argv, const char *prefix)
 {
      int use_gui_tool = 0, dir_diff = 0, prompt = -1, symlinks = 0,
          tool_help = 0, no_index = 0;
-     static char *difftool_cmd = NULL, *extcmd = NULL;
+     static char *difftool_cmd = NULL, *extcmd = NULL, *start_file = NULL;
      struct option builtin_difftool_options[] = {
              OPT_BOOL('g', "gui", &use_gui_tool,
                       N_("use `diff.guitool` instead of `diff.tool`")),

@@ -714,6 +714,8 @@ int cmd_difftool(int argc, const char **argv, const char *prefix)
              OPT_STRING('x', "extcmd", &extcmd, N_("command"),
                         N_("specify a custom command for viewing diffs")),
              OPT_ARGUMENT("no-index", &no_index, N_("passed to `diff`")),
+             OPT_STRING(0, "start-from", &start_file, N_("start-from"),
+                        N_("start viewing diff from the specified file")),
              OPT_END()
      };

This may be a good UI to "difftool".

quoted

@@ -724,6 +726,9 @@ int cmd_difftool(int argc, const char **argv, const char *prefix)
                           builtin_difftool_usage, PARSE_OPT_KEEP_UNKNOWN |
                           PARSE_OPT_KEEP_DASHDASH);

+     if (start_file)
+             setenv("START_FILE", start_file, 1);

Unacceptable.  Nothing gives Git the right to squat on such a
generic name, and there is no hint that it is used to specify the
starting point of what operation.  In addition, I do not see a good
reason why we need to use an environment variable in the first
place.  We run "diff" as an external process, with GIT_EXTERNAL_DIFF
environment pointing back at us, no?  This information should be
passed from its command line.

Sure,I using environment variables originally hoped can be used in
git--difftool-helper.sh, later I found it was not easy to deal with, so I am
attempt to use environment variables to transmit information in
`run_external_diff`,now it seems that this method is not very good ...

quoted

diff --git a/diff.c b/diff.c
index 69e3bc00ed8f..cdad26f99063 100644
--- a/diff.c
+++ b/diff.c

@@ -4249,6 +4249,7 @@ static void run_external_diff(const char *pgm,
                            const char *xfrm_msg,
                            struct diff_options *o)
 {
+     const char *start_file = NULL;
      struct strvec argv = STRVEC_INIT;
      struct strvec env = STRVEC_INIT;
      struct diff_queue_struct *q = &diff_queued_diff;

@@ -4272,9 +4273,17 @@ static void run_external_diff(const char *pgm,

      diff_free_filespec_data(one);
      diff_free_filespec_data(two);
+
+     start_file = xstrdup_or_null(getenv("START_FILE"));
+     if (start_file) {
+             if (strcmp(start_file, name))
+                     goto finish;
+             unsetenv("START_FILE");
+     }

Again, an unacceptable "hack".  "start the diff output showing from
this path" would plausibly a good feature even outside the scope of
"difftool", and the feature should not be limited to the external
diff interface.  More on this later.

What you said `from its command line` is using `git difftool
--rotate-to=<file>`,
Maybe I didn’t know whether to deal with the subcommands of `diff`, now I admit
that passing from parameters is better than passing environment variables.

quoted

      if (run_command_v_opt_cd_env(argv.v, RUN_USING_SHELL, NULL, env.v))
              die(_("external diff died, stopping at %s"), name);

+finish:
      remove_tempfile();
      strvec_clear(&argv);
      strvec_clear(&env);

diff --git a/diff.h b/diff.h
index 2ff2b1c7f2ca..f67c43f5af95 100644
--- a/diff.h
+++ b/diff.h

@@ -86,18 +86,18 @@ typedef void (*diff_format_fn_t)(struct diff_queue_struct *q,

 typedef struct strbuf *(*diff_prefix_fn_t)(struct diff_options *opt, void *data);

-#define DIFF_FORMAT_RAW              0x0001
-#define DIFF_FORMAT_DIFFSTAT 0x0002
-#define DIFF_FORMAT_NUMSTAT  0x0004
-#define DIFF_FORMAT_SUMMARY  0x0008
-#define DIFF_FORMAT_PATCH    0x0010
-#define DIFF_FORMAT_SHORTSTAT        0x0020
-#define DIFF_FORMAT_DIRSTAT  0x0040
+#define DIFF_FORMAT_RAW              (1U<<0)
+#define DIFF_FORMAT_DIFFSTAT (1U<<1)
+#define DIFF_FORMAT_NUMSTAT  (1U<<2)
+#define DIFF_FORMAT_SUMMARY  (1U<<3)
+#define DIFF_FORMAT_PATCH    (1U<<4)
+#define DIFF_FORMAT_SHORTSTAT        (1U<<5)
+#define DIFF_FORMAT_DIRSTAT  (1U<<6)

 /* These override all above */
-#define DIFF_FORMAT_NAME     0x0100
-#define DIFF_FORMAT_NAME_STATUS      0x0200
-#define DIFF_FORMAT_CHECKDIFF        0x0400
+#define DIFF_FORMAT_NAME     (1U<<8)
+#define DIFF_FORMAT_NAME_STATUS      (1U<<9)
+#define DIFF_FORMAT_CHECKDIFF        (1U<<10)

Do we need these changes for the new feature to work?

It has no effect on this new feature. I should put this modification
in an additional commit, right?

I also find this "skip and discard ones earlier than the given path"
makes the utility of the feature artificially narrower than needed,
when I imagine how else this feature, or a variant of it, would be
useful in other situations.  For example, consider that there are 5
paths, and you've seen 3 of them so far before you went off, so you
are restarting from 4th file.  But wouldn't it be more useful, after
showing the 4th and 5th file, if the tool wraps around to show 1st,
2nd, and 3rd file if the user kept going?

Well, it would be better if you can see 1st, 2nd, 3rd again.

I suspect that this feature fits the overall system much better if
it is implemented as a new step in the diffcore transformation.  The
way our diff subsystem works is roughly:

 * The front-end "diff", "diff-files", "diff-index" and "diff-trees"
   are given two "tree like things" to compare, and feeds bunch of
   <old, new> tuples to the diff internal machinery.  Each of these
   tuples are called "filepairs", and a "pair" has two "filespecs",
   one describing the contents, the mode, and the path in the "old"
   side of the "tree like things", the other describing the
   contents, the mode, and the path in the "new" side of the "tree
   like things".

 * The series of filepairs are given to the "diffcore" machinery,
   where they may be broken (i.e. a filepair that says that the old
   contents of "hello.txt" was X, and the new contents of
   "hello.txt" is Y, may become two filepairs, one that says that
   the "hello.txt" file with contents X used to be in the old tree
   but there is nothing corresponding to it in the new tree, and the
   other says that a new "hello.txt" with contents Y appeared
   without corresponding thing in the old tree), matched (i.e. there
   may originally be two filepairs, one that says path A.txt appears
   on the old side but disappeared on the new side, and the other
   that says path sub/A.txt did not exist on the old side but
   appears on the new side---these two filepairs may be merged to
   express "A.txt on the old side got renamed to sub/A.txt on the
   new side"), etc.

Thank you for these interpretations on "git diff" for me, it will help me
understand its principle.

 * The set of filepairs processed in the "diffcore" machinery is
   given to the backend and each filepair will be shown in the
   output, as a series of patches, a diffstat, etc.

There is a step in the "diffcore" machinery called "diffcore-order".
The front-ends all feed the filepairs alphabetically to the
"diffcore" machinery, but "diffcore-order" can reorder them, so that
the original order that the "git diff HEAD --" frontend found the
changes may be to "hello.c" and then "hello.h" (because .c sorts
before .h), but the users can specify with the "-O<orderfile>"
option that they want to view the header files before the source
files.  When the set of filepairs exits the diffcore machinery, the
original "hello.c" then "hello.h" order may get modified to show
"hello.h" then "hello.c".  It probably is the simplest to model this
new feature after how "diffcore-order" does it.

Very magical "diffcore-order", when I was looking for a solution yesterday,
I noticed that these functions similar to "diffcore_apply_filter" in
"diffcore_std",

So, perhaps we can introduce a new "diffcore-rotate" step, where the
filepairs before the specified location are rotated out to the end
of the filepairs?

Awesome idea. In this way, `difftool --rotate-to=<file>` can call
`diff --rotate-to=<file>` , user can choose the starting file, and they can
also see previous files.

The following is just a quick draft that is only lightly tested by
running itself with the "--rotate-to=diffcore-rotate.c" option, but
it should be sufficient to get you started.  You should add a way to
diagnose that the name given to the "--start-file" option actually
exists in the diff on the "difftool" side, because a path that does
not exist in the patch with the following code is simply ignored
(and that is very much deliberate, because we do not want it to die
while running "git log -p --rotate-to=X" where some changes may or
may not touch X).

Yes, I want to know why being so cautious in git log?If the file name is
wrong, why can't I make it exit? :)

Thanks for the patch!
After that, there was too little work I could do,do i just need to add
the following
 code in `diff_flush_patch_all_file_pairs`?
if (o->rotate_to && q->nr && strcmp(q->queue[0]->one->path, o->rotate_to) &&
strcmp(q->queue[0]->one->path, o->rotate_to)) {
    error(_("could not find start file name '%s'"), o->rotate_to);
        return;
}
In addition, Do I need to do the documentation and tests related to
your `diff --rotate-to`?
If so, I will finish it, otherwise, I will only do corresponding tests
for `git difftool --rotate-to`.

Thank you Junio, you are like a tireless teacher,

Happy Chinese New Year!

--
ZheNing Hu