Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes:
> +identifier INIT_ASSIGN1 =~ "^get_worktrees$";
> +// strbuf_init(&I, ...) etc.
> +identifier INIT_CALL1 =~ "^[a-z_]*_init$";
> +// stbuf_release(), string_list_clear() etc.
strbuf?
> +identifier REL1 =~ "^[a-z_]*_(release|clear|free)$";
> +// release_patch(), clear_pathspec() etc.
> +identifier REL2 =~ "^(release|clear|free)_[a-z_]*$";
> +@@
I am hesitant to see this broad set of patterns that could match
init/release functions (and possible false positive matches).
Especially given that it ended up finding only 4 instances, all of
the same "STRBUF_INIT" followed by "strbuf_release()", which means
that all other possible matches, when they actually are found, will
be seen by developers who are not necessarily familiar with these
rules before they are inspected by those who are for correctness.
It would be nice to have a step that catch only strbuf_init(),
STRBUF_INIT, strbuf_release(), and nothing else, possibly with
another step with concrete function names, with other "presumably
functions whose name match this loose pattern are all release
functions" patterns in a separate follow-up patch so that the last
one can easily be reverted.
> +// .. A declaration like "struct strbuf buf;"...
> +(
> +- T I;
> +// ... or "struct STRBUF buf = STRBUF_INIT;" ...
> +|
> +- T I = INIT;
> +)
Presumably, if either of the above followed by foo_release(I) should
be caught, then we should catch "T I = { 0 };" followed by a release
as well. Initialization "T I = { 1, };" for a type without _INIT
macro is also the same story.
Given that, do we even need to limit the forms of declaration? The
only thing we care about is that I is new in this scope, and I is
not used otherwise, in a way other than (1) calling _init() function
on it, or (2) calling _release() function on it, before leaving the
scope, right?
Thanks.
