Re: [PATCH 1/2] reset: handle submodule with trailing slash

[John Keeping <john@keeping.me.uk>]

On Wed, Sep 11, 2013 at 05:54:48PM +0700, Duy Nguyen wrote:
> On Wed, Sep 11, 2013 at 3:20 PM, John Keeping <john@keeping.me.uk> wrote:
> > On Wed, Sep 11, 2013 at 08:05:44AM +0200, Johannes Sixt wrote:
> >> Am 10.09.2013 21:13, schrieb John Keeping:
> >> > When using tab-completion, a directory path will often end with a
> >> > trailing slash which currently confuses "git rm" when dealing with
> >> > submodules.  Now that we have parse_pathspec we can easily handle this
> >> > by simply adding the PATHSPEC_STRIP_SUBMODULE_SLASH_CHEAP flag.
> >> >
> >> > Signed-off-by: John Keeping <john@keeping.me.uk>
> >> > ---
> >> >  builtin/reset.c            | 5 +++++
> >> >  t/t7400-submodule-basic.sh | 6 ++++--
> >> >  2 files changed, 9 insertions(+), 2 deletions(-)
> >> >
> >> > diff --git a/builtin/reset.c b/builtin/reset.c
> >> > index 5e4c551..9efac0f 100644
> >> > --- a/builtin/reset.c
> >> > +++ b/builtin/reset.c
> >> > @@ -220,8 +220,13 @@ static void parse_args(struct pathspec *pathspec,
> >> >             }
> >> >     }
> >> >     *rev_ret = rev;
> >> > +
> >> > +   if (read_cache() < 0)
> >> > +           die(_("index file corrupt"));
> >>
> >> When the index is now read here, I would have expected hunk in this
> >> patch that removes a read_cache() invocation.
> >
> > I think that needs to look like this on top - there's also a
> > read_cache_unmerged() around line 68 but I don't think we can remove
> > that one.
> >
> > -- >8 --
> > diff --git a/builtin/reset.c b/builtin/reset.c
> > index 9efac0f..800117f 100644
> > --- a/builtin/reset.c
> > +++ b/builtin/reset.c
> > @@ -143,7 +143,6 @@ static int read_from_tree(const struct pathspec *pathspec,
> >         opt.output_format = DIFF_FORMAT_CALLBACK;
> >         opt.format_callback = update_index_from_diff;
> >
> > -       read_cache();
> >         if (do_diff_cache(tree_sha1, &opt))
> >                 return 1;
> >         diffcore_std(&opt);
> > @@ -169,7 +168,7 @@ static void set_reflog_message(struct strbuf *sb, const char *action,
> >
> >  static void die_if_unmerged_cache(int reset_type)
> >  {
> > -       if (is_merge() || read_cache() < 0 || unmerged_cache())
> > +       if (is_merge() || unmerged_cache())
> >                 die(_("Cannot do a %s reset in the middle of a merge."),
> >                     _(reset_type_names[reset_type]));
> 
> reset --soft does not go through these code paths (i.e. it does not
> need index at all). If we fail to load index index in "reset --soft" I
> think it's ok to die(). Corrupt index is fatal anyway. But "reset
> --soft" now has to pay the cost to load index, which could be slow
> when the index is big. Assuming nobody does "reset --soft" that often
> I think this is OK.
> 
> Alternatively we could load index lazily in _CHEAP code only when we
> see trailing slashes, then replace these read_cache() with
> read_cache_unless_its_already_loaded_earlier() or something.

read_cache() already has an early return if the index is already loaded
so I don't think we need to worry about a special function for that.

I'm not sure it's worth optimizing this case too heavily, but it might
be a nice change to make parse_pathspec() not rely on the index being
loaded before it is called with certain flags.