Regression introduced in 03267e8656truct object_id oid;

[This behavior was originally reported in https://stackoverflow.com/q/76022720/147356.]

Prior to commit 03267e8656, the following sequence would result in an empty commit:

    git init testrepo
    date > testrepo/testfile
    git -C testrepo add testfile
    git -C testrepo commit -m test --allow-empty --only

After commit 03267e8656, the same sequence creates a commit that contains "testfile". I believe the original behavior was correct, based on the documentation for the `--only` option:

A simple change that corrects this behavior would be to prevent git from refreshing the cache when --only is active:

    diff --git a/builtin/commit.c b/builtin/commit.c
    index 9d8e1ea91a..8c441548f8 100644
    --- a/builtin/commit.c
    +++ b/builtin/commit.c
    @@ -995,7 +995,7 @@ static int prepare_to_commit(const char *index_file, const char *prefix,
        struct object_id oid;
        const char *parent = "HEAD";
     
    -		if (!the_index.cache_nr) {
    +		if (!the_index.cache_nr && !only) {
          discard_index(&the_index);
          if (repo_read_index(the_repository) < 0)
            die(_("Cannot read index"));

And while that passes all the tests, I'm not sure if its correct; the change in 03267e8656 was correcting a memory leak and I'm not sure I completely understand the details.