On Fri, Apr 5, 2019 at 12:04 AM SZEDER Gábor <szeder....@gmail.com> wrote:
> > struct object_id *get_commit_tree_oid(const struct commit *);
> >
> > /*
> > diff --git a/contrib/coccinelle/commit.cocci
> > b/contrib/coccinelle/commit.cocci
> > index c49aa558f0..f5bc639981 100644
> > --- a/contrib/coccinelle/commit.cocci
> > +++ b/contrib/coccinelle/commit.cocci
> > @@ -12,12 +12,12 @@ expression c;
> >
> > // These excluded functions must access c->maybe_tree direcly.
> > @@
> > -identifier f !~
> > "^(get_commit_tree|get_commit_tree_in_graph_one|load_tree_for_commit)$";
> > +identifier f !~
> > "^(repo_get_commit_tree|get_commit_tree_in_graph_one|load_tree_for_commit)$";
> > expression c;
> > @@
> > f(...) {<...
> > - c->maybe_tree
> > -+ get_commit_tree(c)
> > ++ repo_get_commit_tree(the_repository, c)
>
> So, why this change?
Because get_commit_tree() now becomes a compat wrapper (yes I'll fill
in the commit message ;) and should be avoided.
> It would also require furher changes to 'commit.cocci', in particular
> to the last semantic patch, which is supposed to ensure that
> get_commit_tree() doesn't end up on the LHS of an assignment, but with
> this change Coccinelle does suggest transfomations with
> repo_get_commit_tree() on the LHS.
Oooh.. I see now. I actually updated that then dropped, thinking that
a function call cannot be on LHS and that conversion is for
already-long-gone code anyway. But yeah when you stack that conversion
on top of this, it makes sense that we need double conversion to avoid
build error.
Since I will have to update this patch anyway, I'll update the commit
message on the first cocci patch too.
>
> > ...>}
> >
> > @@
> > --
> > 2.21.0.479.g47ac719cd3
> >
--
Duy
