"Phillip Wood via GitGitGadget" <gitgitgadget@xxxxxxxxx> writes: > Rebase records the oid of HEAD before rebasing and the commit created by > "--autostash" in files in the rebase state directory. This means that > the autostash commit is never reachable from any ref or reflog and when > rebasing a detached HEAD the original HEAD can become unreachable if the > user expires HEAD's the reflog while the rebase is running. Fix this by > reading the relevant files when marking reachable commits. I do not like this kind of special casing in general, but because these are our tools' droppings, I am OK to grandfather them in, as long as we promise ourselves that we will not add more of these ad-hoc "text files" that record object names, loss of which affects correctness. They should, like "git bisect", be using proper references to protect these objects instead, of course. I agree with you that we might want to add pseudorefs as a starting points of reachability traversal, but I suspect it would add unnecessary complexity we would rather not want to deal with. For example, not GC'ing what is pointed at by lines in FETCH_HEAD is OK. Excluding those objects that are only reachable from an object mentioned by a pseudoref, when a new "git fetch" is negotiating with a remote what objects need to be sent here, might be disastrous, as the pseudoref that said "this object is here and you can safely consider everything reachable from it is" will be short-lived and can go away anytime, and an auto-gc kicking in at a wrong time ... Thanks.
