From: Junio C Hamano Date: Fri, 03 Jun 2022 18:31:35 GMT Subject: Re: [PATCH 2/4] branch: add branch_checked_out() helper Message-ID: In-Reply-To: <5f54766e1032ebf3a331516a6dd696b997bdfdd8.1654263472.git.gitgitgadget@gmail.com> "Derrick Stolee via GitGitGadget" writes: > +int branch_checked_out(const char *refname, char **path) > +{ > + struct worktree **worktrees = get_worktrees(); > + const struct worktree *wt = find_shared_symref(worktrees, "HEAD", refname); > + int result = wt && !wt->is_bare; > + > + if (result && path) > + *path = xstrdup(wt->path); > + > + free_worktrees(worktrees); > + return result; > +} Don't you plan to call this repeatedly from the for_each_deco iteration? I am wondering if it should take the result of get_worktrees() and reuse the result of get_worktrees(). There also was another topic that was triggered by find_shared_symref() being relatively heavy-weight, which suggests a more involved refactoring. I wonder if we rather want to rewrite find_shared_symref() *not* to take the target parameter at all, and instead introduce a new function that iterates over worktrees and report the branch that is checked out (or being operated on via rebase or bisect). Then we can - create a strset out of its result, i.e. set of branches that should not be touched; - iterate over refs that point into the history being rebased (using for_each_decoration()), and consult that strset to see if any of them is being rewritten. With the API of find_shared_symref(), we'd need to iterate over all worktrees for each decoration. With such a restructuring, we can iterate over all worktrees just once, and match the result with decoration, so the problem becomes O(N)+O(M) and not O(N*M) for number of worktrees N and number of decorations M.