git/contrib/coccinelle/commit.cocci
Nguyễn Thái Ngọc Duy a133c40b23 commit.cocci: refactor code, avoid double rewrite
"maybe" pointer in 'struct commit' is tricky because it can be lazily
initialized to take advantage of commit-graph if available. This makes
it not safe to access directly.

This leads to a rule in commit.cocci to rewrite 'x->maybe_tree' to
'get_commit_tree(x)'. But that rule alone could lead to incorrectly
rewrite assignments, e.g. from

    x->maybe_tree = yes

to

    get_commit_tree(x) = yes

Because of this we have a second rule to revert this effect. Szeder
found out that we could do better by performing the assignment rewrite
rule first, then the remaining is read-only access and handled by the
current first rule.

For this to work, we need to transform "x->maybe_tree = y" to something
that does NOT contain "x->maybe_tree" to avoid the original first
rule. This is where set_commit_tree() comes in.

Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-04-16 18:56:51 +09:00

35 lines
717 B
Plaintext

@@
expression c;
@@
- &c->maybe_tree->object.oid
+ get_commit_tree_oid(c)
@@
expression c;
@@
- c->maybe_tree->object.oid.hash
+ get_commit_tree_oid(c)->hash
@@
identifier f !~ "^set_commit_tree$";
expression c;
expression s;
@@
f(...) {<...
- c->maybe_tree = s
+ set_commit_tree(c, s)
...>}
// These excluded functions must access c->maybe_tree direcly.
// Note that if c->maybe_tree is written somewhere outside of these
// functions, then the recommended transformation will be bogus with
// get_commit_tree() on the LHS.
@@
identifier f !~ "^(get_commit_tree|get_commit_tree_in_graph_one|load_tree_for_commit|set_commit_tree)$";
expression c;
@@
f(...) {<...
- c->maybe_tree
+ get_commit_tree(c)
...>}