This is an archive of the discontinued Mercurial Phabricator instance.

Differential D9778

reverse-branch-cache: switch to doubling allocating scheme
ClosedPublic

Authored by joerg.sonnenberger on Jan 14 2021, 8:19 PM.

Details

Reviewers
marmoute
pulkit
Group Reviewers
hg-reviewers
Commits
rHG1726a53a8494: reverse-branch-cache: switch to doubling allocating scheme
Summary

In preperation for updating the reverse-branch-cache incrementally
whenever a new changeset comes in, avoid bad performance on resize with
Python 3.7 (and likely other 3.x versions).

Diff Detail

Repository
rHG Mercurial
Branch
default
Lint
No Linters Available
Unit
No Unit Test Coverage

Event Timeline

marmoute accepted this revision.Jan 18 2021, 11:08 AM
marmoute added a subscriber: marmoute.

This is clearer. For extra clarity I would probably use something like:

extendsize = (requiredsize * 2) - rbccur

This give about the same result but is simpler to read.

pulkit accepted this revision.Jan 21 2021, 6:22 AM
This revision is now accepted and ready to land.Jan 21 2021, 6:22 AM

Revision Contents
Changeset List

PathPackages
Mmercurial/branchmap.py (15 lines)

Diff 24864

mercurial/branchmap.py

# Revision branch info cache # Revision branch info cache
_rbcversion = b'-v1' _rbcversion = b'-v1'
_rbcnames = b'rbc-names' + _rbcversion _rbcnames = b'rbc-names' + _rbcversion
_rbcrevs = b'rbc-revs' + _rbcversion _rbcrevs = b'rbc-revs' + _rbcversion
# [4 byte hash prefix][4 byte branch name number with sign bit indicating open] # [4 byte hash prefix][4 byte branch name number with sign bit indicating open]
_rbcrecfmt = b'>4sI' _rbcrecfmt = b'>4sI'
_rbcrecsize = calcsize(_rbcrecfmt) _rbcrecsize = calcsize(_rbcrecfmt)
_rbcmininc = 64 * _rbcrecsize
_rbcnodelen = 4 _rbcnodelen = 4
_rbcbranchidxmask = 0x7FFFFFFF _rbcbranchidxmask = 0x7FFFFFFF
_rbccloseflag = 0x80000000 _rbccloseflag = 0x80000000
class revbranchcache(object): class revbranchcache(object):
"""Persistent cache, mapping from revision number to branch name and close. """Persistent cache, mapping from revision number to branch name and close.
This is a low level cache, independent of filtering. This is a low level cache, independent of filtering.
if 'branchinfo' in vars(self): if 'branchinfo' in vars(self):
del self.branchinfo del self.branchinfo
def _setcachedata(self, rev, node, branchidx): def _setcachedata(self, rev, node, branchidx):
"""Writes the node's branch data to the in-memory cache data.""" """Writes the node's branch data to the in-memory cache data."""
if rev == nullrev: if rev == nullrev:
return return
rbcrevidx = rev * _rbcrecsize rbcrevidx = rev * _rbcrecsize
if len(self._rbcrevs) < rbcrevidx + _rbcrecsize: requiredsize = rbcrevidx + _rbcrecsize
self._rbcrevs.extend( rbccur = len(self._rbcrevs)
b'\0' if rbccur < requiredsize:
* (len(self._repo.changelog) * _rbcrecsize - len(self._rbcrevs)) # bytearray doesn't allocate extra space at least in Python 3.7.
) # When multiple changesets are added in a row, precise resize would
# result in quadratic complexity. Overallocate to compensate by
# use the classic doubling technique for dynamic arrays instead.
# If there was a gap in the map before, less space will be reserved.
self._rbcrevs.extend(b'\0' * max(_rbcmininc, requiredsize))
pack_into(_rbcrecfmt, self._rbcrevs, rbcrevidx, node, branchidx) pack_into(_rbcrecfmt, self._rbcrevs, rbcrevidx, node, branchidx)
self._rbcrevslen = min(self._rbcrevslen, rev) self._rbcrevslen = min(self._rbcrevslen, rev)
tr = self._repo.currenttransaction() tr = self._repo.currenttransaction()
if tr: if tr:
tr.addfinalize(b'write-revbranchcache', self.write) tr.addfinalize(b'write-revbranchcache', self.write)
def write(self, tr=None): def write(self, tr=None):