[PATCH 6 of 9 V4] manifest: change manifestlog mancache to be directory based
Durham Goode
durham at fb.com
Tue Oct 25 20:15:37 UTC 2016
On 9/21/16 1:31 PM, Martin von Zweigbergk wrote:
> On Tue, Sep 20, 2016 at 4:47 PM, Durham Goode <durham at fb.com> wrote:
>> # HG changeset patch
>> # User Durham Goode <durham at fb.com>
>> # Date 1474415164 25200
>> # Tue Sep 20 16:46:04 2016 -0700
>> # Node ID 69b91c12d904f329eff6618ac43f5cfef631e8dc
>> # Parent 52a206d772021194ab4356aeba2c73650851a624
>> manifest: change manifestlog mancache to be directory based
>>
>> In the last patch we added a get() function that allows fetching directory level
>> treemanifestctxs. It didn't handle caching at directory level though, so we need to
>> change our mancache to support multiple directories.
>>
>> diff --git a/mercurial/manifest.py b/mercurial/manifest.py
>> --- a/mercurial/manifest.py
>> +++ b/mercurial/manifest.py
>> @@ -1021,9 +1021,16 @@ class manifestlog(object):
>> usetreemanifest = opts.get('treemanifest', usetreemanifest)
>> self._treeinmem = usetreemanifest
>>
>> + # A dictionary containing the cache for each directory
> nit: "dictionary" doesn't add much. I'd rather say what is cached
> ([tree]manifestctx instances, right?).
Will fix
>
>> + self._dirmancache = {}
>> +
>> # We'll separate this into it's own cache once oldmanifest is no longer
>> # used
>> - self._mancache = repo.manifest._mancache
>> + self._dirmancache[''] = repo.manifest._mancache
> IIUC, the root manifest's cache will be shared, but subdirectory logs
> will not be? I think the cache is not used very frequently, so maybe
> it's not a big deal? Do you know a good test case for it? Rebase?
> We're looking for a case where both manifest._mancache and
> manifestlog._dirmancache will be used, right? Perhaps there are no
> such cases so it doesn't matter? So it's probably fine, just wanted to
> point it out in case you can think of problem that I did not see.
It's a bit trickier to share the directory level mancache since the
directory level revlog may not have even been constructed yet when we
construct the ctx (since it's not constructed until we call
ctx.read()). I intend to get the entire manifest refactor in asap once
the we unfreeze, and by the end the manifest class and it's cache will
all be gone, so this is a very temporary measure.
More information about the Mercurial-devel
mailing list