Re: [WIP v2 2/5] mv: check if out-of-cone file exists in index with SKIP_WORKTREE bit

[Shaoxuan Yuan <shaoxuan.yuan02@gmail.com>]

On Fri, May 27, 2022 at 11:13 PM Derrick Stolee
<derrickstolee@github.com> wrote:
> > diff --git a/builtin/mv.c b/builtin/mv.c
> > index 83a465ba83..32ad4d5682 100644
> > --- a/builtin/mv.c
> > +++ b/builtin/mv.c
> > @@ -185,8 +185,32 @@ int cmd_mv(int argc, const char **argv, const char *prefix)
> >
> >               length = strlen(src);
> >               if (lstat(src, &st) < 0) {
> > +                     /*
> > +                      * TODO: for now, when you try to overwrite a <destination>
> > +                      * with your <source> as a sparse file, if you supply a "--sparse"
> > +                      * flag, then the action will be done without providing "--force"
> > +                      * and no warning.
> > +                      *
> > +                      * This is mainly because the sparse <source>
> > +                      * is not on-disk, and this if-else chain will be cut off early in
> > +                      * this check, thus the "--force" check is ignored. Need fix.
> > +                      */
>
> I wonder if this is worth the comment here, or if we'd rather see
> the mention in the commit message. You have documented tests that
> fail in this case, so we already have something that marks this
> as "TODO" in a more discoverable place.

This comment was added during my local development, it should be
removed.

> > +                     int pos = cache_name_pos(src, length);
> > +                     if (pos >= 0) {
> > +                             const struct cache_entry *ce = active_cache[pos];
> > +
> > +                             if (ce_skip_worktree(ce)) {
> > +                                     if (!ignore_sparse)
> > +                                             string_list_append(&only_match_skip_worktree, src);
> > +                                     else
> > +                                             modes[i] = SPARSE;
>
>
> > +                             }
> > +                             else
> > +                                     bad = _("bad source");
>
> style nit:
>
>         } else {
>                 bad = _("bad source");
>         }
>
> > +                     }
> >                       /* only error if existence is expected. */
> > -                     if (modes[i] != SPARSE)
> > +                     else if (modes[i] != SPARSE)
> >                               bad = _("bad source");
>
> For this one, the comment makes it difficult to connect the 'else
> if' to its corresponding 'if'. Perhaps:
>
>         } else if (modes[i] != SPARSE) {
>                 /* only error if existence is expected. */
>                 bad = _("bad source");
>         }
>
> >               } else if (!strncmp(src, dst, length) &&
> >                               (dst[length] == 0 || dst[length] == '/')) {
>
> In general, I found this if/else-if chain hard to grok, and
> a lot of it is because we have "simple" cases at the end
> and the complicated parts have ever-increasing nesting. This
> is mostly due to the existing if/else-if chain in this method.
>
> Here is a diff that replaces that if/else-if chain with a
> 'goto' trick to jump ahead, allowing some code to decrease in
> tabbing:
>
> ---- >8 ----

[cutting the proposed refactor for space]

> ---- >8 ----
>
> But mostly the reason for this refactor is that the following
> diff should be equivalent to yours:
>
> ---- >8 ----
>
> diff --git a/builtin/mv.c b/builtin/mv.c
> index d8b5c24fb5..add48e23b4 100644
> --- a/builtin/mv.c
> +++ b/builtin/mv.c
> @@ -185,11 +185,28 @@ int cmd_mv(int argc, const char **argv, const char *prefix)
>
>                 length = strlen(src);
>                 if (lstat(src, &st) < 0) {
> -                       /* only error if existence is expected. */
> -                       if (modes[i] != SPARSE) {
> +                       int pos;
> +                       const struct cache_entry *ce;
> +
> +                       pos = cache_name_pos(src, length);
> +                       if (pos < 0) {
> +                               /* only error if existence is expected. */
> +                               if (modes[i] != SPARSE)
> +                                       bad = _("bad source");
> +                               goto act_on_entry;
> +                       }
> +
> +                       ce = active_cache[pos];
> +                       if (!ce_skip_worktree(ce)) {
>                                 bad = _("bad source");
>                                 goto act_on_entry;
>                         }
> +
> +                       if (!ignore_sparse)
> +                               string_list_append(&only_match_skip_worktree, src);
> +                       else
> +                               modes[i] = SPARSE;
> +                       goto act_on_entry;
>                 }
>                 if (!strncmp(src, dst, length) &&
>                     (dst[length] == 0 || dst[length] == '/')) {
> ---- >8 ---
>
> To me, this is a bit easier to parse, since we find the error
> cases and jump to the action before continuing on the "happy
> path". It does involve that first big refactor first, so I'd
> like to hear opinions of other contributors before you jump to
> taking this suggestion.

True. I also find it easier to read. Though Victoria mentioned the
goto hazard, the gotos here decouples the huge chain and that
brings clarity and makes it easier to extend.

-- 
Thanks & Regards,
Shaoxuan