Junio C Hamano <gitster@xxxxxxxxx> writes: > That's quite a departure from the established practice, isn't it? > Due to recent and not so recent header shuffling (moving everything > out of cache.h, dropping "extern", etc.), "git blame" is a bit hard > to follow, but ever since 16ac8b8d (setup: introduce the > discover_git_directory() function, 2017-03-13) added the function, > we do execute the "setup" when we know we are in a repository. > > It would probably be worth mentioning that the "global state" Dscho > refers to in that commit is primarily about the current directory of > the Git process. During the discovery, we used to go up one level > at a time and tried to see if the current directory is either the > top of the working tree (i.e. has ".git/" that is a git repository) > or the top of a GIT_DIR-looking directory. That was changed in > ce9b8aab (setup_git_directory_1(): avoid changing global state, > 2017-03-13) in the same series and discusses what "global state" the > series addresses. > > If a relatively recent and oddball caller calls the function when it > does not want any of the setup donw after finding out that we could > use the directory as a repository, a new early "pure discovery" part > should be split out of the function, and both the function itself > and the oddball caller should be taught to call that pure-discovery > helper, I think. Hm, isn't discover_git_directory() that pure-discovery helper? In the ce9b8aab, Dscho created stateless machinery (setup_git_directory_1()), and in 16ac8b8d, he used that stateless machinery to create a pure-discovery helper (discover_git_directory()). It's true that the global state is primarily the cwd, but the spirit of the change is the same - setup_git_directory_1() should have no global side effects and neither should discover_git_diretory() (since it's supposed to only do discovery).
