If a locally added file in A status gets deleted from disk it will move into missing ! status. If the user then decides to delete the file we must remove the file index entry immediately rather than flagging the file for removal during the next commit operation. The file was never committed and lacks a base-blob and base-commit, so it cannot be removed during the next commit. As a result the commit operation was reporting a bogus "out of date" error. Found while diagnosing a related bug reported by gonzalo@ ok op@

If creating the /tmp display diff for a commitable item failed we would free the commitable item while it was already on the path list. Later on when the path list was freed in got_worktree_commit() a double-free would be detected and the program would be aborted. Found by gonzalo@ ok op@

This makes histedit folding work when a file is modified by commit A and then deleted in commit B, with A being folded into B. Problem reported by naddy@

This callback could return any error code. Do not mask all such errors as cancellation.

Walking the whole file index can take some time. Avoid delaying the perceived start-up time of tog by reading the file index in the background log thread. Problem pointed out by op@ with fixes from + ok jamsek

If tog is invoked in a work tree, prefix the base commit log message summary line with a '*' if the work tree is up-to-date, and with a '~' if the base commit is not up-to-date with respect to the branch tip or it contains mixed commits. While here, plug a couple worktree leaks in cmd_ref() and cmd_tree(). ok stsp@

This prevents mixing the use of got and cvg clients in the same work tree. Avoids confusion going forward while cvg differentiates itself further. tog(1) and gotadmin(1) remain compatible with both work tree formats. These tools only read work trees to locate the corresponding repository.

While checking out files, do not attempt to create every directory that exists in the repository tree without paying consideration to their existence on disk. Instead, let functions which add files worry about creating any missing parent directories. This shakes out two other bugs in broken symlink installation code which ended up passing an absolute path to add_dir_on_disk() and forgot to report progress in an unreachable code path that is now getting used. Problem reported by + ok naddy@

Nothing appears to benefit from this being non-static, and it was missing from the public API headers as well. ok stsp@

ok op@

When users run 'got add *' the shell may pick up already versioned files and trigger errors about paths being in an unexpected status. Expand the check which previously only allowed files in added status to be double-added to cover the following status codes which are all safe to ignore: A M C m This should make bulk additions of files a bit easier in most cases. Problem reported by robert@ ok jamsek

Split part of got_worktree_merge_prepare into a new function, got_worktree_merge_write_refs, since that part doesn't make sense in the fast-forward case. ok stsp@

Otherwise, 'got merge -c' can silently revert already committed changes. Also fix GOT_ERR_MERGE_COMMIT_OUT_OF_DATE by giving it a value distinct from GOT_ERR_MERGE_STAGED_PATHS. Patch by James Cook

Make 'got rm' keep empty directories which are not explicitly listed for deletion. Deleting such directories is problematic in several use cases. Avoids deleting the current working directory when the user runs "got rm *" (pointed out by Mikhail), and avoids deletion of an empty directory "foo/" after 'got rm foo/a foo/b' (pointed out by op@). ok jamsek, op

Requested by James Cook

Problem reported by James Cook with reproduction scripts Regression test case by jamsek, as a simplified version of James' scripts fix tested + ok jamsek

Otherwise we leave unversioned files behind in the work tree which may interfere with new attempts to rebase or merge the changes again. Problem found by + ok naddy@

problem found by robert@ ok jamsek, op

The commands 'got rebase -a' and 'got histedit -a' were checking out files from the wrong commit. Make them check files out from the commit we are switching the work tree to, as intended. Avoids spurious merge conflicts when the work tree is later used for another rebase operation. It also makes 'got update' right after 'rebase -a' a no-op, as it should be. Problem found by naddy@ while rebasing jca's llvm15 branch ok op, jamsek earlier version

This avoids having to run 'got update' after operations which run checkout_files() and where a file happens to have the same blob ID but its base commit differs from the commit we are updating to. Found while investigating a bug report by naddy@ ok op, jamsek (earlier version)

problem found by naddy@

problem found by naddy@ ok op@