Re: [PATCH] entry.c: use dir-iterator to avoid explicit dir traversal

[Matheus Tavares Bernardino <matheus.bernardino@usp.br>]

On Sun, Dec 8, 2019 at 3:06 PM <otalpster@gmail.com> wrote:
>
> From: Plato <otalpster@gmail.com>
>
> Replace usage of opendir/readdir/closedir API to traverse directories
> recursively, at remove_subtree() function, by the dir-iterator API. This
> simplifies the code and avoids recursive calls to remove_subtree().
>
> Signed-off-by: Plato <otalpster@gmail.com>
> ---
> Hello,
>
> This is my first patch.

Hello Plato, and welcome!
Thanks for working on this.

> I hope I cc'd the correct people and didn't mess up.
>
> The changes pass the test suite t/ and Travis CI.
> Please point out any mistakes.
>
> Thanks for your time! :)
>
>  entry.c | 32 +++++++++++++++-----------------
>  1 file changed, 15 insertions(+), 17 deletions(-)
>
> diff --git a/entry.c b/entry.c
> index 53380bb614..e7f4881d3b 100644
> --- a/entry.c
> +++ b/entry.c
> @@ -2,6 +2,8 @@
>  #include "blob.h"
>  #include "object-store.h"
>  #include "dir.h"
> +#include "iterator.h"
> +#include "dir-iterator.h"
>  #include "streaming.h"
>  #include "submodule.h"
>  #include "progress.h"
> @@ -50,29 +52,25 @@ static void create_directories(const char *path, int path_len,
>
>  static void remove_subtree(struct strbuf *path)
>  {
> -       DIR *dir = opendir(path->buf);
> -       struct dirent *de;
> -       int origlen = path->len;
> +       int ok;
> +       unsigned int flags = DIR_ITERATOR_PEDANTIC;
> +       struct dir_iterator *iter = dir_iterator_begin(path->buf, flags);
>
> -       if (!dir)
> +       if (!iter)
>                 die_errno("cannot opendir '%s'", path->buf);

Nitpick: since dir_iterator_begin() might fail for reasons other than
an opendir() error, I think the error message here could be more
generic. Maybe "failed to start iterator over %s"?

> -       while ((de = readdir(dir)) != NULL) {
> -               struct stat st;
>
> -               if (is_dot_or_dotdot(de->d_name))
> +       while ((ok = dir_iterator_advance(iter)) == ITER_OK) {
> +               if (is_dot_or_dotdot(iter->path.buf))

This check is already done by dir-iterator internally, so you may
remove it here.

>                         continue;
>
> -               strbuf_addch(path, '/');
> -               strbuf_addstr(path, de->d_name);
> -               if (lstat(path->buf, &st))
> -                       die_errno("cannot lstat '%s'", path->buf);
> -               if (S_ISDIR(st.st_mode))
> -                       remove_subtree(path);
> -               else if (unlink(path->buf))
> -                       die_errno("cannot unlink '%s'", path->buf);
> -               strbuf_setlen(path, origlen);
> +               if (unlink(iter->path.buf)) {

unlink()-ing a directory in Linux will return a EISDIR error. So I
think you still need to use S_ISDIR() to check if iter->path.buf is a
directory and call rmdir(), in this case.

However, note that the dir-iterator API gives entries in pre-order.
I.e. a directory appears before its subentries. In the use case of
remove_subtree(), though, we need to traverse in post-order, since we
have to remove the subentries before removing the directory where they
reside. My suggestion is that you add a preliminary patch,
implementing a new DIR_ITERATOR_POST_ORDER flag to dir-iterator.h, and
then use it in this patch.

You may also want to check this[1] series, which worked towards the
same goal of converting remove_subtree(). It ended up not getting
merged, back them, but some of the patches were re-used in this[2]
series which got merged. I think you could also re-use some of the
code from [1] that implements the post-order traversing and a test[3]
for remove_subtree().

Thanks,
Matheus

[1]: https://public-inbox.org/git/1493226219-33423-1-git-send-email-bnmvco@gmail.com/
[2]: https://public-inbox.org/git/cover.1562801254.git.matheus.bernardino@usp.br/
[3]: https://public-inbox.org/git/1493226219-33423-3-git-send-email-bnmvco@gmail.com/