summaryrefslogtreecommitdiff
path: root/doc/bugs/incremental_fsck_should_not_use_sticky_bit/comment_4_5ac5a10bdddf23153e8ea0a8eb60323e._comment
blob: 3b55891c4f6c5a7d08644d1f9185721063dd33a4 (plain)
1
2
3
4
5
6
7
8
9
10
[[!comment format=mdwn
 username="http://joeyh.name/"
 ip="209.250.56.64"
 subject="comment 4"
 date="2013-12-02T20:35:45Z"
 content="""
@helmut Well, true, it could overwrite the old ref with a new one at the start of a new fsck pass. If there was a separate ref for each uuid, this would work even for mixed local/remote fscking. And git should eventually drop the objects associated with the old ref, probably in `gc --auto`, which has a default --prune of 2 weeks.

The git overhead would probably be fine for fsck. I am less sure about it for the direct mode mapping files, which can be queried quite frequently. In particular inAnnex is used lots of places, often repeatedly on a lot of files, and it uses withObjectLoc, which uses associatedFiles.
"""]]