| Submitter | Tony Tung |
|---|---|
| Date | July 21, 2016, 5:57 a.m. |
| Message ID | <a216bfe47256fa157f7b.1469080661@andromeda.local> |
| Download | mbox | patch |
| Permalink | /patch/15965/ |
| State | Accepted |
| Headers | show |
Comments
I love this 100ms speed-up. If we want to support the old format, I think it is doable by making "branches" lazy and testing "if rname in bookmarks" first. Alternatively, we may want to find a way to disable "branches" more completely. I prefer dropping support. Will let smf decide. Excerpts from ttung's message of 2016-07-20 22:57:41 -0700: > # HG changeset patch > # User Tony Tung <tonytung@merly.org> > # Date 1469080576 25200 > # Wed Jul 20 22:56:16 2016 -0700 > # Node ID a216bfe47256fa157f7bc1fb80e10d91918f95b6 > # Parent 183d5b47f7472b839a96e7be735bb7fbf8068969 > remotenames: remove support for old remotenames file format > > This can cost up to 100ms when we don't need it on startup time. > > Things that I've based this number on: 5 runs each of hg status, hg bookmark. > > diff --git a/remotenames.py b/remotenames.py > --- a/remotenames.py > +++ b/remotenames.py > @@ -1229,10 +1229,6 @@ > if not vfs.exists('remotenames'): > return > > - # needed to heuristically determine if a file is in the old format > - branches = repo.names['branches'].listnames(repo) > - bookmarks = repo.names['bookmarks'].listnames(repo) > - > f = vfs('remotenames') > for line in f: > nametype = None > @@ -1254,14 +1250,6 @@ > if not rname: > continue > > - # old format didn't save the nametype, so check for the name in > - # branches and bookmarks > - if nametype is None: > - if rname in branches: > - nametype = 'branches' > - elif rname in bookmarks: > - nametype = 'bookmarks' > - > yield node, nametype, remote, rname > > f.close()
Jun Wu <quark@fb.com> writes: > I love this 100ms speed-up. If we want to support the old format, I think it > is doable by making "branches" lazy and testing "if rname in bookmarks" > first. Alternatively, we may want to find a way to disable "branches" > more completely. > > I prefer dropping support. Will let smf decide. I'm fine with dropping support (this is a very experimental extension). I'm still processing the journal patch so I'm behind on reviewing this one.
Sean Farley <sean@farley.io> writes: > Jun Wu <quark@fb.com> writes: > >> I love this 100ms speed-up. If we want to support the old format, I think it >> is doable by making "branches" lazy and testing "if rname in bookmarks" >> first. Alternatively, we may want to find a way to disable "branches" >> more completely. >> >> I prefer dropping support. Will let smf decide. > > I'm fine with dropping support (this is a very experimental extension). > I'm still processing the journal patch so I'm behind on reviewing this > one. This is now pushed. Thanks!
Patch
diff --git a/remotenames.py b/remotenames.py --- a/remotenames.py +++ b/remotenames.py @@ -1229,10 +1229,6 @@ if not vfs.exists('remotenames'): return - # needed to heuristically determine if a file is in the old format - branches = repo.names['branches'].listnames(repo) - bookmarks = repo.names['bookmarks'].listnames(repo) - f = vfs('remotenames') for line in f: nametype = None @@ -1254,14 +1250,6 @@ if not rname: continue - # old format didn't save the nametype, so check for the name in - # branches and bookmarks - if nametype is None: - if rname in branches: - nametype = 'branches' - elif rname in bookmarks: - nametype = 'bookmarks' - yield node, nametype, remote, rname f.close()