When a parent is not present in the pack file, handle this situation gracefully rather than erroring out. Fixes a regression introduced with recent got-read-pack coloring changes where commands such as 'got send' could fail with 'got-read-pack: object not found'.

Port the parent commit coloring fix to got-read-pack, and ensure that it starts off with the same state as the main process. got-read-pack did not have access to the main process's ids array, and was thus working with a different initial state. With these changes the same commit traversal happens regardless of whether coloring is "offloaded" to got-read-pack or not (verified manually by placing debug printfs). ok jamsek

When a commit is first discovered as a commit which should be included in the pack, but is later found to be a parent of a commit which should be excluded from the pack, gotadmin pack correctly excluded the commit itself but failed to exclude this commit's parents. This bug is the reason why our test suite did not notice that gotd was not protecting references when clients did not send a pack file. In our test case, these parents are in the 'keep' set already and were never added to the 'skip' set. A useless pack was built which included those parents and nothing else. Add a test which triggers the bug via gotadmin pack. ok jamsek

Reflow lines exceeding 80 columns; replace Allman with K&R braces; and line continuation whitespace. ok stsp@

Found by martijn@ ok jamsek

Reported by naddy: if the user opens tog and quickly quits before the log thread has finished fetching work tree state for displaying dirty work tree entries, the main thread waits for the log thread to finish before acting on the user's quit input. This produces a noticeable lag between entering 'q' and tog actually terminating. Pass a pointer to the quit flag to the worktree diff cancel callback so we can act on it immediately without making the user wait. tested by naddy who confirmed it greatly reduces the lag

Make gotadmin cleanup resolve multiple levels of nested tags, in order to ensure that objects referenced via nested tags will not get deleted. Add test coverage for this edge case.

We only considered referenced commit and tag objects during cleanup. Directly referenced trees or blobs were seen as unreferenced, and deleted during cleanup. Add test coverage for this problem and fix it.

While here, fix potential NULL deref in gotadmin cleanup and pack error paths. ok stsp@

Our cleanup implementation is only safe to use after everything referenced has been packed into a single pack file. Otherwise, the algorithm we use for checking pack redundancy might remove small packs which contain objects that other objects depend on. The easy fix for this issue is to have 'gotadmin cleanup' create the required pack file before cleaning up, making cleanup safe by default. This happens to be what 'git gc' does as well. ok jamsek

provided by mark during review of previous change

We are going to run modified files through the diff engine anyway. As a bonus speed-up tweak, skip conflict marker detection in binary files found on disk.