Re: [PATCH 0/2 v4] xdiff: implement empty line chunk heuristic

Stefan Beller [off-list ref] writes:

quoted

OK, so perhaps either of you two can do a final version people can
start having fun with?

Here we go. I squashed in your patch, although with a minor change:

-               if ((flags & XDF_SHORTEST_LINE_HEURISTIC)) {
+               if ((flags & XDF_COMPACTION_HEURISTIC) && blank_lines) {

We did not need that in the "shortest line" heuristic as we know
a line with the shortest line length must exist. We do not know about
empty lines though.

Makes sense.  The last hunk of

$ git show 9614b8dcf -- update-cache.c

gives an unexpected result without "&& blank_lines" above.  Lack of
"&& blank_lines" happens to make the result slightly easier to read,
but at the cost of having an extra line in the hunk.

Thanks.

On Mon, Apr 18, 2016 at 2:22 PM, Junio C Hamano [off-list ref] wrote:

Stefan Beller [off-list ref] writes:

quoted

quoted

OK, so perhaps either of you two can do a final version people can
start having fun with?

Here we go. I squashed in your patch, although with a minor change:

-               if ((flags & XDF_SHORTEST_LINE_HEURISTIC)) {
+               if ((flags & XDF_COMPACTION_HEURISTIC) && blank_lines) {

We did not need that in the "shortest line" heuristic as we know
a line with the shortest line length must exist. We do not know about
empty lines though.

Makes sense.  The last hunk of

$ git show 9614b8dcf -- update-cache.c

gives an unexpected result without "&& blank_lines" above.  Lack of
"&& blank_lines" happens to make the result slightly easier to read,
but at the cost of having an extra line in the hunk.

So without the blank_lines check you get  (A):
    @@ -271,15 +279,14 @@ int main(int argc, char **argv)
                     if (!verify_path(path)) {
                             fprintf(stderr, "Ignoring path %s\n", argv[i]);
                             continue;
    -                }
    -                if (add_file_to_cache(path)) {
    -                        fprintf(stderr, "Unable to add %s to
database\n", path);
    -                        goto out;
                     }
    +                if (add_file_to_cache(path))
    +                        usage("Unable to add %s to database", path);
             }
    ...

and with the heuristic you get (B):

@@ -272,14 +280,13 @@ int main(int argc, char **argv)
    @@ -272,14 +280,13 @@ int main(int argc, char **argv)
                             fprintf(stderr, "Ignoring path %s\n", argv[i]);
                             continue;
                     }
    -                if (add_file_to_cache(path)) {
    -                        fprintf(stderr, "Unable to add %s to

database\n", path);
    -                        goto out;
    -                }
    +                if (add_file_to_cache(path))
    +                        usage("Unable to add %s to database", path);
             }
    ...

In case of (A) the compaction heuristic tries to shift the hunk upwards,
stopping at the first empty line or when lines miss match.
As there is no blank line, it goes until the miss match.

Personally I'd find it less readable, because the intent was not to remove

    -                }
    -                if (add_file_to_cache(path)) {
    -                        fprintf(stderr, "Unable to add %s to
database\n", path);
    -                        goto out;

but rather remove

    -                if (add_file_to_cache(path)) {
    -                        fprintf(stderr, "Unable to add %s to
database\n", path);
    -                        goto out;
    -                }

as that is the logic unit I'd think.

Although you find this instance easier to read the behavior without the
blank_lines check would result in

    Shift hunk upward as much as possible, stop at the first empty line.

For hunks without empty line this just becomes

    Shift hunk upward as much as possible.

which is 50:50 for looking good, so we kept the old behavior as
that is just as good.

Thanks,
Stefan

Thanks.