priority

95

sprint

""

status

triage

id

"7ab5"

assignee

null

kind

issue

Today lock are badly managed

Today every call to repository.Open() return a repository and then caller can RLock or WLock but sometimes it’s too late.

To bypass that, I made a cacheLock which permit to freeze all cal to reposotory during an operation (creation for example). But no other code can run during a cacheLock because they will not acquire cacheLock.

Suggestion

I think wee need to decouple:

• a repository in read without FS (no worketree) => most use case + mountinf worktree is time consuming and it is done every time today
• a repository in will write with FS => second most use case should let read pass
• repository in write => no other can do anything

When a repository write, others open repository need to know that and reload (invalid cache?) of current worktree.

All of this lock state should be managed outside of a repo by the manager. Like that it not impact the way repository works.