Re: [PATCH 06/10] dir.c: get rid of lazy initialization

[Elijah Newren <newren@gmail.com>]

On Sun, Oct 3, 2021 at 5:46 PM Ævar Arnfjörð Bjarmason <avarab@gmail.com> wrote:
>
> Remove the "Lazy initialization" in prep_exclude() left behind by
> aceb9429b37 (prep_exclude: remove the artificial PATH_MAX limit,
> 2014-07-14).
>
> Now that every caller who sets up a "struct dir_struct" is using the
> DIR_INIT macro we can rely on it to have done the initialization. As
> noted in an analysis of the previous control flow[1] an earlier
> passing of of "dir->basebuf.buf" to strncmp() wasn't buggy, as we'd
> only reach that code on subsequent invocations of prep_exclude(),
> i.e. after this strbuf_init() had been run. But keeping track of that
> makes for hard-to-read code. Let's just rely on the initialization
> instead.

Having read through the link previously, this all makes sense to me,
but I'm not sure if this paragraph motivates the change without that
context.  Maybe another reader can comment.

> This does change the behavior of this code in that it won't be
> pre-growing the strbuf to a size of PATH_MAX. I think that's OK.
>
> That we were using PATH_MAX at all is just a relic from this being a
> fixed buffer from way back in f87f9497486 (git-ls-files: --exclude
> mechanism updates., 2005-07-24).
>
> Pre-allocating PATH_MAX was the opposite of an optimization in this
> case. I logged all "basebuf.buf" values when running the test suite,
> and by far the most common one (around 80%) is "", which we now won't
> allocate at all for, and just use the "strbuf_slopbuf".
>
> The second most common one was "a/", followed by other common cases of
> short relative paths. So using the default "struct strbuf" growth
> pattern is a much better allocation optimization in this case.
>
> 1. https://lore.kernel.org/git/87sfxhohsj.fsf@evledraar.gmail.com/
>
> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
> ---
>  dir.c | 8 --------
>  dir.h | 4 +++-
>  2 files changed, 3 insertions(+), 9 deletions(-)
>
> diff --git a/dir.c b/dir.c
> index 39fce3bcba7..efc87c2e405 100644
> --- a/dir.c
> +++ b/dir.c
> @@ -1550,14 +1550,6 @@ static void prep_exclude(struct dir_struct *dir,
>         if (dir->pattern)
>                 return;
>
> -       /*
> -        * Lazy initialization. All call sites currently just
> -        * memset(dir, 0, sizeof(*dir)) before use. Changing all of
> -        * them seems lots of work for little benefit.
> -        */
> -       if (!dir->basebuf.buf)
> -               strbuf_init(&dir->basebuf, PATH_MAX);
> -
>         /* Read from the parent directories and push them down. */
>         current = stk ? stk->baselen : -1;
>         strbuf_setlen(&dir->basebuf, current < 0 ? 0 : current);
> diff --git a/dir.h b/dir.h
> index ff3b4a7f602..e3757c6099e 100644
> --- a/dir.h
> +++ b/dir.h
> @@ -342,7 +342,9 @@ struct dir_struct {
>         unsigned visited_directories;
>  };
>
> -#define DIR_INIT { 0 }
> +#define DIR_INIT { \
> +       .basebuf = STRBUF_INIT, \
> +}
>
>  struct dirent *readdir_skip_dot_and_dotdot(DIR *dirp);
>
> --
> 2.33.0.1404.g83021034c5d

Wahoo!  Nice code cleanup.