I (Julian Foad) wrote: > I noticed that svn_wc_copy3() and svn_wc__move2() didn't check that they > have been called with the write lock(s) required by their doc strings. Some > other WC functions do -- e.g. svn_wc_delete4() for example. > > So I added corresponding checks in svn_wc_copy3() and svn_wc__move2(), and > fixed > the ensuing test failures where we weren't acquiring a lock, in r1410044. > Something like: > > SVN_ERR(svn_wc__write_check(db, parent_path, pool)); > > Then I noticed there are still more WC APIs that say they require a lock but > don't check. For example, svn_wc__internal_remove_from_revision_control(). > And this one has a "real" caller that's broken: > svn_client_copy6().
(Oops, svn_client_copy6() isn't broken, I noticed a failure because I implemented the test wrongly. svn_wc_remove_from_revision_control2()'s doc string just said "must hold a write lock" and I checked for a lock on the parent, but in fact it should check for a lock on LOCAL_ABSPATH itself if it's a WC root, else on the parent.) > Can we do something more rigorous to ensure that we check for having a write > lock in the functions where it is logically required? Any ideas? This question still holds. - Julian
