Per-file log is *very* slow with packs
Bug #172567 reported by
Lukáš Lalinský
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Bazaar |
Expired
|
Undecided
|
Unassigned |
Bug Description
Knits (best of 5 runs):
$ time bzr log bzrlib/
real 0m1.637s
user 0m1.520s
sys 0m0.068s
Packs (best of 5 runs):
$ time bzr log bzrlib/
real 0m13.334s
user 0m10.869s
sys 0m0.292s
IMO, this is quite important regression.
To post a comment you must log in.
Any chance you could attach a callgrind/lsprof run with the difference?
My guess is that reading the per-file graph is having to read the complete graph of all files. So it is just time spent loading the .tix indexes.
bzr --lsprof-file ../packs.callgrind log bzrlib/ win32utils. py > /dev/null win32utils. py > /dev/null
bzr --lsprof-file ../knits.callgrind log bzrlib/
And then attach the files here.